Page 1

LEXSEE 50 STANFORD L REV. 1055
Copyright (c) 1998 The Board of Trustees of Leland Stanford Junior University
Stanford Law Review
April, 1998
50 Stan. L. Rev. 1055
LENGTH: 37292 words
ARTICLE: Handcuffing the Cops? A Thirty-Year Perspective on Miranda's Harmful Effects on Law Enforcement
NAME: Paul G. Cassell * and Richard Fowles **
BIO:
* Professor of Law, University of Utah College of Law. cassellp@law.utah.edu.
** Associate Professor of Economics, University of Utah. Helpful suggestions were provided by Ron Allen,
Akhil Amar, Ian Ayres, Ron Boyce, Craig Bradley, George Brower, Lionel Frankel, William Gangi, Joe Grano,
Sam Gross, Yale Kamisar, Richard Leo, Eric Rasmusen, Morgan Reynolds, Stephen Schulhofer, David
Sklansky, Bill Stuntz, Lee Teitelbaum, George Thomas III, Richard Uviller, Lloyd Weinrieb, participants in
presentations at the American Bar Association ("ABA") 1996 National Convention, Northwestern University,
Stanford University, UCLA, and the University of Utah, and discussants on
<crimprof@chicagokent.kentlaw.edu>, an electronic mail distribution list for criminal law professors. David
Huth provided valuable assistance in running preliminary regression equations, and Katarzyna Celinska helped
collect several data series and provided much other valuable statistical analysis. Research assistants Karen
Korevaar and Bret Hayman were also vital to the success of this project. The article was supported by the
University of Utah College of Law Research Fund and the University of Utah Research Committee. Although
this article was a collaborative effort, Professor Cassell had initial responsibility for data collection and legal
analysis, whereas Professor Fowles had initial responsibility for statistical analysis.
SUMMARY:
... In this article, Professors Cassell and Fowles find this claim to be supported by FBI data on crime clearance rates. ...
Part II then considers the only available long-term measure of Miranda's effects: crime clearance rate data. ... The most
recent empirical study of confession rates in this country is the Cassell-Hayman 1994 study in Salt Lake County, Utah,
reporting an overall confession rate of only 33%. ... The clearance rate appears to be a reasonable if understated
surrogate measure for the confession rate. ... This variable, converted to a crime rate by dividing the number of index
crimes by the resident population for the country, is identified as CRIME RATE. ... In 1995, for example, the violent
crime clearance rate was 45.4%. ... There is no reason to suspect, for example, that a heroin-induced robbery would be
harder to solve than other robberies. ... The FBI has reported clearance rate data for "violent" crime (i.e., nonnegligent
homicide, rape, robbery, and assault) and "property" crime (i.e., burglary, larceny, and auto theft) from 1969 to date,
although from 1969-1972 the property crime category appears to include only clearance of larcenies $ 50 and over. ...
As a result, we used revised crime rate data to explain unrevised clearance rate data. ...

Page 2
50 Stan. L. Rev. 1055, *1055

TEXT:
[*1055]
After the Supreme Court's 1966 decision in Miranda v. Arizona, critics charged that it would "handcuff the cops."
In this article, Professors Cassell and Fowles find this claim to be supported by FBI data on crime clearance rates.
National crime clearance rates fell precipitously in the two years immediately after Miranda and have remained at lower
levels in the decades since. Multiple regression analysis reveals that other possibly confounding factors - such as the
rising crime rate and baby boom children reaching crime prone-years in the 1960s - do not account for much of the
post-Miranda decline in clearance rates. Rather, the cause of the decline was most likely the Supreme Court's broad new
restrictions on police questioning. The authors conclude that Miranda has in fact "handcuffed" the police and that
society should begin to explore ways of loosening these shackles.

I believe the decision of the Court ... entails harmful consequences for the country at large. How serious these
consequences may prove to be only time can tell... The social costs of crime are too great to call the new rules anything
but a hazardous experimentation.
- Justice John Harlan, dissenting in Miranda v. Arizona n1
[*1056] [*1057]
Introduction
On June 13, 1966, the United States Supreme Court ignited a firestorm of controversy with Miranda v. Arizona, which
became its most famous criminal law decision. n2 Miranda established new rules for interrogation of criminal suspects,
requiring among other things that officers deliver the now-famous warnings to suspects and obtain a waiver of their
constitutional rights to remain silent and speak to an attorney before beginning any custodial interrogation. Critics
immediately predicted that the requirements would put "handcuffs on the police" n3 and prevent the prosecution of
countless dangerous criminals. n4 Justice Harlan warned in his dissenting opinion that the decision would "entail[]
harmful consequences for the country at large. How serious these consequences may prove to be only time can tell." n5
Justice White likewise predicted that "in some unknown number of cases the Court's rule will return a killer, a rapist or
other criminal to the streets and to the environment which produced him, to repeat his crime whenever it pleases him."
n6 After extensive hearings, the Senate Judiciary Committee agreed with the dissenting justices, stating that "crime will
not be effectively abated so long as criminals who have voluntarily confessed their crimes are released on mere
technicalities." n7 And, during the 1968 presidential campaign, then-candidate Richard Nixon charged that Miranda
"had the effect of seriously [*1058] ham stringing [sic] the peace forces in our society and strengthening the criminal
forces." n8
In spite of these predictions of substantial harm to law enforcement, few researchers ever attempted to confirm or
refute the claims by gathering hard statistics on Miranda's effects. n9 In what has been aptly described as an "empirical
desert," n10 we have little knowledge about what police interrogation looked like shortly after Miranda, much less what
it looks like today. The lack of data, however, has not deterred Miranda's academic supporters from asserting that the
decision has had only a "negligible" effect on law enforcement n11 and that "the Miranda dissenters' fears [of harm to
law enforcement] did not prove justified." n12 Typical of the view from academe is the claim in one of the nation's
leading criminal procedure hornbooks that "little has changed since Miranda was decided." n13 But careful
examination of these claims reveals that they typically rest on selective citation of a few atypical studies or carefully
culled anecdotal information. Has Miranda handcuffed the cops over the last thirty years? In truth, even the most
knowledgeable defenders of Miranda have offered little beyond informed guesses about this fundamental issue. Yet
they have been quite successful in spinning the straw of speculation into the gold of a generally accepted, empirically
confirmed statement about Miranda's effects. n14

Page 3
50 Stan. L. Rev. 1055, *1058

The question of Miranda's practical effect has far more than academic significance. The Supreme Court views
Miranda as a pragmatically grounded, prophylactic rule - "a carefully crafted balance designed to fully [*1059] protect
both the defendant's and society's interests." n15 If the costs of the decision are greater than the Court has recognized,
then the balance is upset and the Court would presumably need to rethink the doctrine's current contours. Even the
strongest defenders of Miranda seem to agree that the number of criminals who go free "does matter" in assessing the
ruling. n16
Justice Harlan's dissent was surely right on one point: that "only time [could] tell" what the real effects of Miranda
would be. n17 The aim of this article is to draw on the more than thirty years of experience with Miranda to
quantitatively assess its effect on law enforcement's ability to solve crimes. n18 Although some have suggested that
questions about Miranda's effects "probably lie beyond the capacity of social science methodology to answer," n19 we
are not so skeptical. Social science has powerful analytic tools that can and should be brought to bear on this question.
A brief road map of our article may be in order. Part I reviews the limited evidence about Miranda's effects. Part II
then considers the only available long-term measure of Miranda's effects: crime clearance rate data. The clearance rate
is the rate at which police declare crimes solved. It is properly and generally regarded as a conservative measure of
Miranda's harmful effects. Yet contrary to the prevailing conventional wisdom, crime clearance rates fell precipitously
immediately after Miranda and have remained at lower levels ever since. This suggests that the decision prevented
police from solving a substantial number of crimes.
Part III presents a multiple regression time series analysis of crime clearance rates. After discussing the
appropriateness of regression analysis for determining Miranda's effects on clearance rates, we propose a model based
on the crime rate, expenditures on law enforcement, the percentage of the population in the crime-prone younger years,
and other relevant variables. Even controlling for these factors, police ability to solve crimes declined sharply after
Miranda. The reductions in clearance rates are observed for the [*1060] composite groupings of "violent" and
"property" crimes, as well as for the individual crime categories of robbery, vehicle theft, larceny, and (in most models)
burglary. Only the categories of murder, rape, and assault did not show significant clearance rate reductions due to
Miranda. The findings recur across a variety of specifications of the regression equations with respect to such issues as
the timing of the Miranda effect and the length of the time series.
Part IV defends the position that Miranda was an important cause of the 1966-1968 drop in clearance rates. This
was the contemporaneous view of the Federal Bureau of Investigation ("FBI") and law enforcement officers on the
streets. Moreover, Miranda was the Warren Court's most substantial restriction on police investigative techniques. Apart
from Miranda, other possible causes are poor candidates for explaining the sharp decline in clearance rates over that
three-year period, a conclusion that is consistent with other available data.
Part V then explains why a Miranda-induced drop in clearance rates should be regarded as a social cost. It first
refutes the argument that clearance rates might have fallen because police officers were generally using
unconstitutionally coercive questioning techniques before Miranda that disappeared after the decision. It then disproves
Professor Kamisar's hypothesis that the post-Miranda clearance rate drop was merely a harmless reduction in purely
"paper" clearances.
Finally, Part VI places our findings in the context of the ongoing debate about the wisdom of Miranda. It concludes
that Miranda has in fact handcuffed the cops and that society should begin to explore other, less costly ways of
regulating police interrogation.
I. Our Limited Knowledge About Miranda
Immediately after Miranda, a handful of researchers attempted to measure the decision's effects. The studies generally
suggested that fewer suspects gave confessions after the decision. In another publication, one of us collected the limited
empirical evidence on Miranda's social costs in terms of lost criminal cases. n20 Examining direct information on

Page 4
50 Stan. L. Rev. 1055, *1060

Miranda's effects - the studies of confession rates "before-and-after" the decision - the earlier article concludes that
Miranda significantly depressed the confession rate. n21 For example, a 1967 research effort in Pittsburgh revealed that
con- [*1061] fession rates there fell from 49% before the decision to 30% after. n22 Averaging this and other reliable
before-and-after studies leads to the conclusion that confession rates fell by about 16% in the year or two after Miranda,
n23 meaning that the decision results in a lost confession in about one out of every six criminal cases. Because
confessions are needed to convict in about one out of every four criminal cases, Miranda results in the loss of criminal
convictions in about 3.8% of all criminal cases (16% confession rate drop multiplied by the 24% confession necessity
rate). n24
This tentative estimate of Miranda's harmful effects came solely from the before-and-after studies conducted
directly after Miranda, which are not universally accepted as accurate measures. At least one commentator believes the
studies to be "so crudely designed, so ineptly executed, and so thoroughly riddled with the most elementary
methodological defects" as to be inadequate for careful statistical analysis. n25 Moreover, because the studies rely on
data from the months immediately surrounding Miranda, they fail to capture Miranda's long-term effects. n26 As a
result, defenders of Miranda argue that the studies overstate the harm to law enforcement, because they do not capture
later adjustments by police organizations to minimize the harms of Miranda. n27 On the other hand, critics of Miranda
contend that the studies understate the harmful effects, because full implementation of and compliance with the decision
was not immediate. n28
To gain a better, long-term perspective on Miranda's effects, we need some solid statistical indicator that extends
beyond 1967 and, indeed, into the 1990s. In theory, the ideal study would be to review confession rates since 1967 to
see whether, despite initial declines after the decision, they have since "rebounded" - in other words, a before-and-after
study of confession rates over several decades rather than over several months. Confession rates are the preferred
measure of Miranda's effects because Miranda directly affects the ability of police to obtain confessions. Unfortunately,
statistics on confession rates have not been regularly collected on a nationwide basis ei- [*1062] ther before or after
Miranda. Instead, the only figures that exist were gathered by individual researchers for particular cities on a one-time
basis. Although broad generalizations are hazardous, before-Miranda confession rates in this country were probably
somewhere around 55-60%. n29 After Miranda, the few studies extant reveal lower confession rates. The most recent
empirical study of confession rates in this country is the Cassell-Hayman 1994 study in Salt Lake County, Utah,
reporting an overall confession rate of only 33%. n30 Richard Leo's 1993 study from Berkeley, California found an
in-custody questioning success rate by detectives of 64%. n31 This percentage, when adjusted to account for suspects
not questioned and the greater efficacy of questioning by detectives of suspects in-custody, so as to be comparable to
earlier studies, translates into an overall confession rate of about 39%. n32 A 1977 study of six cities reported a
"confession" rate of 40.3%. n33
Taken together, the limited data suggest that confession rates in the years after Miranda are lower than confession
rates in the years before, implying that Miranda has in fact impeded law enforcement. n34 But this conclusion, too, has
been criticized by defenders of Miranda because it rests on studies from individual cities that may not be generalizable
across the country. n35 Responding to such criticisms is difficult, because it is quite true that existing data on
confession rates are limited. In short, the existing empirical research has arguably not resolved, at least to the
satisfaction of Miranda's defenders, the question of whether Miranda has hindered law enforcement over the long haul.
Only a national, long-term assessment of Miranda's effects will re- [*1063] spond to concerns and shed light on the
impact of Miranda on police effectiveness.
II. Declining Clearance Rates After Miranda
A. Clearance Rates As a Measure of Miranda's Effects
Since regularly collected, long-term data on confession rates are unavailable, we must search for a second-best
alternative. The strongest candidate for such a statistic is the crime "clearance" rate, the rate at which police "clear" or
solve crimes. Since at least 1950, the FBI has collected clearance rate figures from around the country and reported

Page 5
50 Stan. L. Rev. 1055, *1063

them annually in the Uniform Crime Reports ("UCR"). n36 Because of this extended range of data, clearance rates
might permit a broad perspective on Miranda's effects.
The clearance rate appears to be a reasonable if understated surrogate measure for the confession rate. n37
Sometimes police officers, lacking evidence to clear a case, will bring a suspect in, deliver Miranda warnings,
interrogate, and - if no confession results - release him, leaving them with insufficient evidence to clear the case. n38 If
Miranda prevented the confession, by discouraging the suspect from talking or otherwise, the crime may never be
cleared. As the leading police interrogation manual explains, "Many criminal cases, even when investigated by the best
qualified police departments, are capable of solution only by means of an admission or confession from the guilty
individual or upon the basis of information obtained from the [*1064] questioning of other criminal suspects." n39
The most recent field research on police interrogations found that "virtually every detective ... insisted that more crimes
are solved by police interviews and interrogations than by any other investigative method." n40 Confessions are also
sometimes necessary to solve multiple crimes committed by the same perpetrator. For example, even if police can arrest
and convict a robber for one robbery, without a confession they may not be able to clear four other robberies he also
committed. n41
Clearance rates have been widely viewed as statistics that would reveal Miranda's effects - especially by defenders
of Miranda. n42 For example, a widely cited passage in Professor Stephen Schulhofer's influential 1987 article praising
Miranda claimed that, although some of the before-and-after studies suggested declining confession rates after Miranda,
"within a year or two [clearance] rates were thought to be returning to pre-Miranda levels." n43 Many other supporters
of Miranda have likewise viewed clearance rates as an appropriate indicator of Miranda's benign effects. n44 Similarly,
law enforcement officers at the time of Miranda believed that the decision would lower their clearance rates. n45 We
have been unable to locate any article arguing that clearance rates would not capture at least some of the decision's
effects. n46
[*1065] While an apparent consensus exists that clearance rates at least partially gauge Miranda's impact, one note
of caution should be sounded. Police can record a crime as "cleared" when they have identified the perpetrator and
placed him under arrest, even where the evidence is insufficient to convict, or even to indict. As a result, clearance rates
fail to measure any of Miranda's harmful effects that manifest themselves at stages of the criminal justice process after a
crime has been cleared. It seems quite likely that, for some lost confessions, police may still be able to "clear" the crime
but lack sufficient evidence to obtain a conviction. n47 Therefore, clearance rates are a quite conservative measure of
Miranda's harmful effects on the conviction of criminals.
In theory, one could begin to measure the understatement of Miranda's harms by measuring the rate at which
cleared cases are prosecuted. If confession rates fell after Miranda, prosecutors might charge fewer suspects because the
lack of a confession made the prosecution more difficult. n48 Thus, a reduction in confessions could lower the charging
rate in cleared cases. This possibility is, in practice, a moot point because of the practical problem of collecting adequate
data. The FBI's data on charging decisions are woefully inadequate for statistical analysis, swinging wildly from year to
year during the 1960s. n49 The FBI stopped reporting charging figures in the 1970s.
One other theoretical possibility for measuring Miranda's impact would be to investigate conviction rates rather
than clearance rates. Convictions have the advantage of resting on court adjudications of guilt or innocence. n50
[*1066] However, conviction rates would probably miss many of Miranda's effects. n51 More significantly, conviction
rate data in this country are notoriously bad. n52 The basic problem is that police agencies, the source for FBI data, are
poorly situated to report on ultimate court outcomes. Perhaps for this reason, the FBI stopped reporting conviction
figures in 1978. n53
Through a process of elimination, then, the choice for a long-term evaluation of Miranda boils down to the
understated measure of clearance rates.
B. The National Clearance Rate Trend

Page 6
50 Stan. L. Rev. 1055, *1066

Did clearance rates go up or down after Miranda? The conventional academic wisdom is that Miranda had no
noticeable effect on crime clearance rates. n54 That wisdom is perhaps most prominently embodied in Professor
Schulhofer's 1987 article, Reconsidering Miranda, which said that clearance rates were "thought to be returning to
pre-Miranda levels" shortly after the decision and that "study after study confirmed this trend." n55 Although
Schulhofer has since repudiated this position, n56 his 1987 article has been cited repeatedly as proof that Miranda has
not hampered law enforcement. For example, Professor Yale Kamisar, perhaps Miranda's leading academic supporter,
has concluded that Schulhofer's article "effectively refutes [the] contention" that Miranda has harmed law enforcement.
n57
[*1067] The support for these claims about lack of impact on clearance rates is vanishingly thin. Boiling down
Schulhofer's secondary sources to their primary authorities, one discovers that he has referenced only two studies with
clearance rate data. n58 Neither of these studies provide support for the thesis that clearance rates have returned to
pre-Miranda levels. n59 Indeed, the few other statistical analyses of post-Miranda clearance rates that we have found
suggest that clearance rates fell. In New York City, in February 1967, the Deputy Commissioner of the New York
Police Department reported that clearance rates dropped about 10% in 1966. n60 He attributed the drop in part to
"recent Supreme Court decisions that had limited the admissibility of confessions in court." n61 In "Seaside City,"
California, crime clearance rates dropped about 3% after Miranda. n62
The studies finding post-Miranda declines in clearance rates in various cities suggest that something is amiss with
the conventional wisdom. But results from any particular city might be idiosyncratic. To gain a broader view of
Miranda's effects, we should examine evidence from across the country. Surprisingly, even thirty years after the
decision, the national data from the FBI's UCR remain to be systematically examined. The data show that, in fact, crime
clearances rates fell sharply all over the country immedi- [*1068] ately after Miranda and have remained at these lower
levels ever since. In 1965, the year preceding Miranda, the UCR noted that the national clearance rate for the "grand
total" crimes n63 was "virtually unchanged from 1964." n64 The next year in June, the Court handed down Miranda.
At the end of 1966, the UCR (which usually describes police performance in decidedly upbeat terms) acknowledged a
substantial drop in clearances. n65 Indeed, the UCR observed that the drop in clearance rates from 1965-1966 was
equal to the entire drop in clearance rates from 1961-1965. n66 The 1966 drop in clearances was "universally reported
by all population groups and by all geographic divisions." n67 The following year, 1967, the UCR continued to report
widespread bad news. n68 Again a clearance rate drop was "universally reported by all population groups and by all
geographic divisions." n69 In 1968, the UCR acknowledged another fall in clearance rates. n70 In 1969, the UCR
reported that most clearance rates declined slightly, n71 and in 1970 the UCR reported that clearance rates were
unchanged. n72 Clearance rates have remained roughly stable since 1970. n73
A long-term perspective on clearance rates comes from plotting the FBI's annual figures. Figure 1 depicts the
national crime clearance rate from 1950-1995 for violent crimes (i.e., nonnegligent homicide, forcible rape, aggravated
assault, and robbery). As can be seen, violent crime clearance rates were fairly stable from 1950-1965, generally
hovering above 60%. They even increased slightly from 1962-1965. Then, in the three years following Miranda, the
rates fell dramatically - to 55% in 1966, to 51% in 1967, to [*1069] 47% in 1968. n74 Violent crime clearance rates
have hovered around 45% ever since. Because Miranda probably took effect over several years - while both police
practices and suspect volubility adjusted to the new rules n75 - simple visual observation of the long-term trends
suggests that Miranda substantially harmed police efforts to solve violent crimes. Moreover, contrary to the notion that
clearance rates returned to pre-Miranda levels, n76 violent crime clearance rates in fact have been permanently
depressed since the decision.
[SEE FIGURE IN ORIGINAL] [*1070] A similar conclusion is suggested in Figure 2, which shows the annual
crime clearance rate for the same period for the property crimes of burglary, vehicle theft, and larceny. The rate at
which police cleared property crimes fluctuated somewhat from 1950-1960, declined from 1961-1965, fell at an
accelerating rate from 1966-1968, and then generally stabilized. Here again, clearance rates dropped during the critical
post-Miranda period, although somewhat less dramatically than violent crime clearance rates.

Page 7
50 Stan. L. Rev. 1055, *1070

[SEE FIGURE IN ORIGINAL] [*1071]
III. Regression Analysis of the Clearance Rate Decline

A. Using Regression to Sort Through Competing Causes
A more thorough analysis of the hypothesis that Miranda caused the declines in crime clearance rates must contend
with other competing causes. If another factor - call it the "X factor" - was responsible for the fall in clearance rates,
then Miranda would be absolved of responsibility. For example, in two new brief essays Professor Schulhofer now
concedes that, contrary to his earlier suggestions, clearance rates did indeed fall after Miranda. n77 He maintains,
however, that there is no basis for blaming the decline on Miranda; instead he blames other factors such as rising crime
rates during the 1960s for the decline. n78
The standard technique for sorting through such competing possibilities is multiple regression analysis. In this
section, we report our multiple regression equations for crime clearance rates, using an interrupted time series design.
n79 Specifically, we analyze whether, controlling for other relevant factors, there was a detectable change in clearance
rates at the time of Miranda. Before diving into the complexities of multiple regression equations, however, one
important point must be emphasized. Although sophisticated econometric techniques are available for analyzing the
data, simple visual observation has its place as well. n80 The graphs in the previous section demonstrate that there was
a sharp, post-Miranda drop in clearance rates; that overall picture nicely fits the handcuffing-the-cops theory advanced
by Miranda's critics. Even standing alone, the trend lines are important evidence suggesting Miranda had an adverse
effect. We wonder whether the many defenders of Mirandawho concluded that the decision had no adverse effects on
the basis of a posited stable post-Miranda clearance rate n81 will [*1072] now, consistent with their methodological
approach, rethink their position and agree that Miranda was indeed harmful to police efforts.
We turn to time series analysis out of necessity, as the Miranda decision precludes the use of other common
statistical techniques. The preferred methodology for assessing a social policy is a true experiment in which one
jurisdiction at random is subjected to the new policy, while another "control" jurisdiction is not. n82 Unfortunately,
such research is not possible with Miranda. On June 13, 1966, the Supreme Court required all jurisdictions across the
country to follow the prescribed interrogation procedures. Since then, police agencies have generally followed the
Miranda requirements with little innovation. n83 Comparison of a control group with a subject group is thus not
possible. n84
We also used time series analysis because another similar and commonly used statistical technique, cross-sectional
analysis, is unavailable to us. In this technique, the impact of legal rules is analyzed by state-by-state "cross-sectional"
analysis, where data from states following one legal regime are compared with data from states that do not. Professor
Schulhofer suggests using this methodology to assess Miranda's impact, claiming that "professional econometric studies
of the impact of legal rules almost invariably rely on analysis of cross-sectional, not aggregate, data." n85 One of the
studies he cites as proof of an "almost invariable" practice of cross-sectional analysis, however, is in fact a national time
series analysis similar to ours. n86 The other studies he cites involve not a legal norm applied across the country, but
rather legal rules that vary from jurisdiction to jurisdiction, permitting a true cross-sectional comparison. n87 It would
make no sense to [*1073] apply that methodology here. There simply are no "control" jurisdictions unaffected by
Miranda to observe, n88 at least in this country. n89 Moreover, as a practical matter, obtaining appropriate
cross-sectional data for the variables in our equations appears to be impossible, n90 and such data as exist might be
contaminated by other problems. n91
In any event, interrupted time series analysis is quite appropriate for assessing the effect of a legal reform. n92 This
statistical technique is commonly [*1074] used for assessing the effects of legal changes. n93 Standard statistical texts
suggest that this technique is well-suited for such issues, provided that care is used in analyzing the data and other
factors not covered in the regressions are considered. n94

Page 8
50 Stan. L. Rev. 1055, *1074

B. A Model of Crime Clearance Rates
The first step in developing any regression model is to identify relevant variables for the equations. For our dependent
variable, we decided to use national clearance rates using country-wide FBI data. Although FBI data have been
criticized on various grounds, they remain the only data available to resolve the question of Miranda's effects. n95 In
any event, the literature generally suggests that FBI data, at least for the last few decades, are of sufficient quality for
time series analysis. n96
[*1075] FBI clearance rate data in particular have been criticized. To clear a crime, police need not actually obtain
a conviction, but only determine to their own satisfaction that the crime has been solved. n97 Unsurprisingly, clearance
rate statistics have been attacked on the grounds that police can manipulate them in an effort to look good, n98 or that
police may simply disagree as to what crimes should be regarded as "cleared." n99 A study in New York City, for
example, found that the number of clearances per burglary varied widely from precinct to precinct, probably because
individual police commanders defined clearances differently. n100 Another study found large variances in the way that
clearances were recorded. n101
Although such deviations might present a serious problem for analyses of police effectiveness in particular
jurisdictions, our focus here is on a Supreme Court decision binding on police forces around the country. Accordingly,
we use the aggregate national clearance rate, comprised of clearance reports from thousands of agencies. n102 Even if a
particular city reported rates in a questionable fashion, our results would be unaffected if any such manipulations did
not change significantly in the several years surrounding Miranda or if any changes in the manipulations were relatively
small in comparison to the total number of reports nationally - reasonable assumptions both. n103 We also report
results for the aggregated categories of clear- [*1076] ance rates for "violent crimes" and "property crimes," for which
inter-jurisdictional variations in categorizing crimes should be minimized. n104
Another potential problem is that the FBI database varies slightly from year to year, depending on which cities
report their figures to the FBI. But, again, this problem is relatively small for national data. n105 We defer discussion
of other issues surrounding our data to the Appendix. n106
What factors influence clearance rates? The literature on clearance rates is surprisingly thin, particularly in
assessing clearance rates over time. n107 A few studies analyze clearance rates in particular jurisdictions. These studies
suggest that existing models of clearance rates are underdeveloped, as they rarely meet with much success in explaining
clearance rates. n108 The available literature does, however, offer a few possible control variables.
The factor most commonly cited as affecting the clearance rate is the crime rate. The standard argument is that as
police officers have more crimes to solve, they will be able to solve a smaller percentage of them n109 - variously
called the "overload" theory, n110 the "overtaxing" theory, n111 or the "system strain" the- [*1077] ory. n112
Although the theory has intuitive appeal, the empirical support is mixed. n113 Moreover, crime rates rose throughout
the 1960s and later, a pattern that does not correspond to the sharp 1966-1968 decline in clearance rates. n114 To
control for the number of crimes, we collected FBI data on the estimated number of FBI "index" crimes committed
across the country each year from 1960-1994. Because consistent FBI data are unavailable before 1960, we estimated an
extension of the FBI data back to 1950, as explained at greater length in the Appendix. n115 This variable, converted to
a crime rate by dividing the number of index crimes by the resident population for the country, is identified as CRIME
RATE.
Apart from the crime rate, the factors most often cited as influencing clearance rates are law enforcement personnel
and expenditures on law enforcement. n116 With more personnel and resources available, the argument goes, more
crimes should be cleared. Here again, despite the intuitive appeal of the hypothesis, the studies are mixed. n117 To
control for any influences [*1078] these factors may have, we collected data on the number of law enforcement
employees per capita ("POLICE PERSONNEL") n118 and the dollars spent on police protection per capita by state and
local governments, adjusted for inflation by the consumer price index ("POLICE DOLLARS (REAL)").

Page 9
50 Stan. L. Rev. 1055, *1078

Professor Schulhofer has suggested that clearance rates would respond not simply to changes in law enforcement
manpower and expenditures, but also to interactions between these variables and the overall number of crimes - what he
calls the "capacity" of the system. n119 To test this theory, we added to the equations PERSONNEL CAPACITY,
which was defined as the rate of police employees per capita divided by the crime rate for index crimes, and DOLLAR
CAPACITY (REAL), which was defined as the number of inflation-adjusted dollars spent on police protection per
capita divided by the crime rate for index crimes. n120
Criminal justice variables are not the only ones that might affect clearance rates. Other variables have been
identified in the criminal justice literature as having some bearing on clearance rates or, more generally, on crime rates.
Perhaps the most salient of these factors is the number of persons in the crime-prone younger years. n121 Most crimes
will be committed by [*1079] persons who are in adolescence or early adulthood. n122 Increases in the number of
young persons, particularly in connection with the post-World War II "baby boom," have been linked with changes in
crime rates. n123 The age band commonly identified with this effect is fifteen to twenty-four, n124 which also
corresponds to readily available census data. Since it is conceivable that more juveniles on the streets might similarly
influence clearance rates, we included a variable to take into account this age band ("JUVENILES"). n125 A control
variable of this type has been used in other criminal justice research. n126
Changes in various socioeconomic variables are also plausible candidates for affecting clearance rates. Criminal
justice literature identifies variances in the unemployment rate as a possible explanation for crime rate fluctuations.
n127 Similarly, changes in income levels and labor force partici- [*1080] pation might be associated with crime rates
n128 and clearance rates, particularly since such factors might be viewed as a measure of the opportunity cost of
committing a crime. Accordingly, we included variables for the labor force participation ("LABOR FORCE
PARTICIPATION"), unemployment rate ("UNEMPLOYMENT"), disposable per capita real income ("PER CAPITA
INCOME (REAL)").
As a measure of changing social circumstances that might be related to crime, we also added a variable that
measured live births to unmarried mothers, n129 converted to a per capita rate by dividing by resident population
("BIRTHS TO UNMARRIED WOMEN").
It has also been suggested that increasing urbanization during the 1960s was an explanation for rising crime rates.
n130 Conceivably, urbanization could have some bearing on clearance rates as well. n131 For instance, clearance rates
for index crimes are generally higher in smaller cities. n132 A few studies of clearance rates, however, have found slim
predictive power in urbanization. n133 To control for the possible effects of urbanization, we added a variable for the
percent of the resident population residing in urban areas ("URBANIZATION").
In response to a preliminary version of our equations, Professor Schulhofer suggested that we should have
controlled for the distribution of crimes committed in large and small cities. n134 Because smaller cities have higher
clearance rates, Schulhofer reasons that a shift in the distribution of crimes [*1081] could bias our results. n135 To
test this hypothesis, we controlled for the percentage of violent crimes committed in small cities, as reported in the UCR
("CRIME IN SMALL CITIES"). n136
As a final control, we added a standard time trend variable, identified as "TREND OVER TIME." Although some
cautions have been raised about such a variable, n137 we thought it might be useful to control for long-term,
time-related trends apart from Miranda.
Finally, to capture the effects of the Miranda decision, we included a "dummy" variable in the equations
("MIRANDA"), which was assigned the value of 0 before Miranda and 1 after. Because we were working with yearly
data and because Miranda was handed down on June 13, 1966 (roughly halfway through 1966), deciding what to do
with the 1966 value of the MIRANDA variable was an issue. As a first approximation of Miranda's effects, we assigned
MIRANDA the value of 0 for years before 1965, 0.5 for 1966, and 1 for 1967 and following years. This is the simplest
model, which ordinarily serves as a starting point for time series research. n138 We discuss alternative assumptions

Page 10
50 Stan. L. Rev. 1055, *1081

about when Miranda began to affect clearance rates at greater length below. n139
We thought that these variables would contain the most important influences on crime clearance rates over the last
several decades. Although other variables could be included in the equations, there are certain statistical advantages to a
parsimonious construction. n140 We discuss below whether "omitted" variables could have influenced our conclusions.
n141 Wherever reasonably possible, we have also chosen the least complicated statistical technique for analysis. Our
intent in resorting to econometrics is to clarify the factors that affect clearance rates, not to obfuscate them.
[*1082] Using standard ordinary least squares ("OLS") regression techniques, it is possible to develop an equation
to explain national crime clearance rates as follows:
[SEE TABLE IN ORIGINAL]
where i runs from 1950-1995 and the independent variables are as described above. This is a "reduced form"
equation, which assumes that there are no "simultaneity" problems (that is, that the dependent clearance rate variables
do not affect any of the independent variables), an assumption we review below. n142
C. Crime Clearance Rates by Aggregate and Individual Crime Categories
Table I reports our findings for the aggregate categories of violent and property crimes and shows a statistically
significant coefficient n143 associated with MIRANDA for both equations. For those not familiar with interpreting
regression equations, the table shows a statistically significant effect on crime clearance rates from MIRANDA. The
coefficient associated with the MIRANDA variable in the violent crime equation is -6.731, which suggests that crime
clearance rates would be 6.7% higher without the shift captured in the MIRANDA variable. In 1995, for example, the
violent crime clearance rate was 45.4%. n144 The regression equations suggest that the rate would have been 52.1%
(45.4% + 6.7%) in the absence of Miranda. Similarly, for prop- [*1083] erty crimes, the coefficient is -2.272, which
indicates that the clearance rate for property crimes in 1995 would have been 19.9% (17.6% + 2.3%). n145
The numbers in Table I rest on the aggregate categories of "violent" and "property" crime. There is a danger, of
course, that such aggregations may produce misleading results. n146 In particular, it is possible that one component
part of the aggregate may have changed in response to Miranda, creating the spurious impression that the entire
category had changed.
[SEE TABLE IN ORIGINAL] [*1084] This problem can be eliminated by simply disaggregating violent and
property crimes and running separate regressions on the component crimes. Figure 3 depicts clearance rates for the
violent crimes of nonnegligent homicide, rape, and aggravated assault.
[SEE FIGURE IN ORIGINAL]
The three crimes exhibit a long-term downward trend, but, unlike the other violent crimes, robbery has a sharp
downward break in the 1966-1968 period. Robbery is depicted separately, because its clearance rate is so much lower,
in Figure 4.
[*1085]
[SEE FIGURE IN ORIGINAL]
As can be seen, Figure 4 shows the sharp reduction in 1966-1968. This suggests that robbery clearances in
particular may have been depressed by Miranda. The regression results reported in Table II confirm this conclusion.
n147
[*1086]

Page 11
50 Stan. L. Rev. 1055, *1086

[SEE TABLE IN ORIGINAL]
The only individual violent crime for which the MIRANDA variable has a statistically significant effect is the
crime of robbery.
Turning to property crimes (i.e., burglary, larceny, and vehicle theft), Figure 5 shows that all exhibit a downward
trend following Miranda. Vehicle theft clearance shows a particularly sharp drop in the 1966-1968 period, while the
drop in burglary and larceny clearances extends from 1961-1968.
[*1087]
[SEE FIGURE IN ORIGINAL]
The regression results reported below reflect these visual observations. As can be seen in Table III, the MIRANDA
variable has a statistically significant downward effect on clearance rates for all of the property.
[SEE TABLE IN ORIGINAL] [*1088] [*1089]
D. Explaining the Pattern
Our equations suggest a "Miranda effect" n148 on clearance rates for robbery, larceny, vehicle theft, and burglary, but
not homicide, rape, or assault. What could explain this pattern? n149 No doubt, the reasons are complex, but we
venture some tentative interpretations.
At first glance, notice that what might in part be loosely denominated as crimes of passion or aggression (i.e.,
murder, rape, and assault) were apparently unaffected by Miranda, while what are more often crimes of deliberation
(i.e., robbery, larceny, vehicle theft, and possibly burglary) were affected. These categories, of course, are gross
oversimplifications, as there are obviously coolly calculated murders and impulsive car thefts. But if the generalizations
are more often than not correct, they might correspond with the empirical evidence suggesting that Miranda more
substantially affects police success in dealing with repeat offenders and professional criminals. n150 It seems probable
that professional criminals would most often commit crimes such as robbery, burglary, and vehicle theft.
Still another explanation rests on the point that police may more often clear certain kinds of crimes through
confessions. A study of the New York City Police Department around the time of Miranda compiled ratios of clearances
to arrests across crime categories. n151 The ratio of clearances to arrests is well in excess of one for some crimes specifically burglary, grand larceny, grand larceny vehicle, and robbery. n152 For burglary, for example, the ratio was
3.778, meaning that, for each burglary arrest, police cleared more than three burglaries. No doubt a substantial number
of these additional clearances came from confessions. n153 On the other hand, for other crimes - specifically homicide,
rape, and assault - the ratio was quite close to one. Presumably murderers and rapists rarely confess to more than one
crime. This suggests that confessions may play a more important role for crimes such as burglary, vehicle theft, larceny,
and robbery, and that clearance for [*1090] these crimes are, therefore, more susceptible to changes in confession
procedures.
Another possible partial explanation is that police may be able to shift resources to maintain high clearance rates for
the most serious and less numerous crimes (e.g., murder and rape) at the expense of lower clearance rates for less
serious and more numerous crimes (e.g. larceny and vehicle theft). Police agencies are frequently judged by their
effectiveness in solving the most notorious crimes, especially murders. n154 As a result, maximum "detective-power"
is allocated to solve homicides. n155 If Miranda affected clearances generally, one would expect police to respond. To
the extent resources affect clearance rates, n156 police should be able to maintain high clearance rates for the most
serious crimes by allocating more resources to solve them, but at the cost of lower clearance rates in crime categories
less visible to the public. n157

Page 12
50 Stan. L. Rev. 1055, *1090

If the shifting resources theory is correct, it places an interesting gloss on the only available before-and-after
Miranda study that specifically analyzes individual crime categories. Researchers in Pittsburgh found that, after
Miranda, the confession rate in homicide cases fell 27.3 percentage points, robbery cases 25.7 points, auto larceny 21.2
points, burglary and receiving goods 13.7 points, and forcible sex offenses 0.5 points. n158 Interestingly, the Pittsburgh
study found no change in homicide clearances after Miranda and a huge drop in robbery clearance rates. n159 In the
eighteen months before Miranda, Pittsburgh police cleared 94.3% (50 of 53) of all homicides; in the thirteen months
after, they also cleared 94.4% (34 of 36) of all homicides - even though confessions for homicide fell sharply. n160
While it would have been feasible for the Pittsburgh police to devote the necessary resources to solve thirty-four
homicides with fewer confessions, it would have been virtually impossible to deal with the vast decline in the number of
robbery confessions. Before Miranda, the Pittsburgh police cleared 45.1% (970 of 2152) of all robberies; after Miranda,
they cleared only 30.8% (556 of 1805). n161 If the [*1091] Pittsburgh numbers are indicative of what happened
around the country, then there may have been a shift in police resources on a national basis toward homicide offenses at
the expense of other less serious offenses (e.g., robbery, burglary, larceny, and vehicle theft). This could explain why
our analysis did not find a Miranda effect for homicide.
Another reason why no Miranda effect for homicide was found may be that we did not account for changing
homicide patterns. Homicide, more than any other violent crime, exhibits a long-term decline in clearance rates. It is
quite likely that at least some portion of this drop is attributable to the increase in the proportion of felony-type murders
and the corresponding decline in murders within the family or as a result of what the FBI describes as "romantic
triangles and lovers' quarrels." n162 Presumably the family and romantic homicides are easier to solve, whereas the
felony-type murders, often committed by strangers, are less so. We have not accounted for these changes because of the
difficulty in obtaining data for the relevant time period. n163 It is possible that these changes have obscured any
Miranda effect in the homicide regressions.
If changes in the patterns of homicides could have obscured a Miranda-induced drop in clearance rates, could
changes in the patterns of other crimes have caused a drop in clearance rates that coincided with Miranda? It seems
improbable that crime patterns would have changed suddenly enough to explain the kind of sharp 1966-1968 drop that
we observed, for example, for robbery. In any event, there were two crimes for which patterns were relatively stable
during the late 1960s: robbery and larceny. n164 Since we found a Miranda effect on both of these crimes, our Miranda
effect does not appear to be an artifact of changing crime patterns.
[*1092]
E. Specification of the Equations
In this section we consider various questions that might be asked about the specification of our regression equations.
1. The timing of the Miranda effect.
The basic equations just reported assume that Miranda's impact was fully and immediately felt halfway through 1966.
Yet not all of Miranda's effects would have been felt immediately on and after June 13, 1966, the date the Court
announced its decision. Instead, it is more likely that the decision took hold over several years. n165 Police compliance
with Miranda was far from immediate and uniform throughout the country, as the requirements of the decision were
"implemented slowly and imperfectly in many jurisdictions." n166
The four available empirical studies immediately following Miranda all found substantial deviations from the
prescribed procedures. A study by student observers in New Haven in the summer of 1966 found that police failed to
follow important parts of the Miranda regime, such as the requirements of obtaining an affirmative waiver of rights
from a suspect or stopping questioning upon a suspect's request to see counsel. n167 While the observers reported
improved compliance with Miranda over the summer, they observed full compliance in only a fraction of the
interrogations. n168 Similarly, a 1966 study based on interviews of suspects interrogated in the District of Columbia

Page 13
50 Stan. L. Rev. 1055, *1092

concluded that police had "failed to observe the spirit and often the letter of Miranda." n169 Noncompliance with
Miranda was also suggested by a field observation of D.C. police immediately after the decision. n170 A 1967 study by
the Vera Institute in New York City, based in part on audiotapes of police interrogations, concluded that whether police
complied with Miranda was "open to serious question." n171 Finally, a 1969 study of interrogations in Denver reported
general police compliance with Miranda, with some excep- [*1093] tions. n172 Interviews with suspects suggested
some possible police noncompliance, however. n173 These four studies are the only assessments of Miranda
compliance shortly after the decision in which researchers either observed police interrogations or talked to persons who
had been interrogated. n174 Interestingly, all report incomplete implementation.
Of course, proving substantial noncompliance with Miranda in and around 1966 does not fully answer our question
about the timing of Miranda's effect. The remaining question is when, if at all, did police begin to substantially comply?
Because of the dearth of police interrogation studies since 1970, n175 iron-clad evidence on police compliance with
Miranda is unavailable. The prevailing view, however, is that police now follow Miranda's requirements n176 and that
they began doing so by around 1969. As one scholar who reviewed the available information concluded,
implementation problems were largely a "transitional" feature, and "by the end of the 1960s, Miranda advice had
become an accepted part of police work." n177 This conclusion suggests it is reasonable to assume that Miranda effects
manifested themselves over several years.
Focusing on suspects rather than police also suggests that Miranda's effects unfolded over several years.
Presumably over time suspects gained [*1094] experience, either directly or through the "grapevine," in how to take
advantage of the new interrogation rules. Precise empirical evidence on this point is hard to come by, but there are some
suggestions that suspects would adjust to Miranda-style requirements over several years. n178
One last issue is whether confessions might have been declining before June 1966 because of other Warren Court
decisions. In particular, in June 1964, the Supreme Court decided Escobedo v. Illinois, n179 suppressing a confession
because the police had denied Escobedo's request to see a lawyer after the investigation had begun to focus on him. The
ruling has been aptly described as having an "accordion-like quality," both "broad" and "confining," n180 and it is clear
that "the scope and meaning of the decision was a matter of strong and widespread disagreement." n181 The limited
available evidence suggests that Escobedo had little, if any, effect on confession rates. n182 Nonetheless, it remains
possible that Escobedo or its subsequent lower court interpretations n183 had some limited effect on confession rates
before Miranda, an issue that can be investigated with various equations.
Although it seems likely that Miranda's effects were not instantaneous, the available information does not establish
precisely what impact structure is appropriate. Some care is required on this point, because in theory the selection of
impact structure could make a difference to regression conclusions. n184
[*1095] We believe the best view is that Miranda took hold over the 1966-1968 period. However, to assess the
competing possibilities, we tested a variety of different impact structures for the MIRANDA variable. We hasten to
emphasize that we tested these alternative models only to rebut potential criticism of our model. There is considerable
wisdom in those who caution against efforts to "fish" for an impact structure that will produce the desired outcome.
n185 The attempt here, however, was not to obtain a result, but to test the strength of our conclusions by examining
what difference alternative assumptions would have made.
We assumed that Miranda's impact would be faithfully modeled in one of the following:
July 1966. As a baseline basic model, we used the Miranda impact structure discussed previously, which assumed
that Miranda took effect almost immediately on July 1, 1966; that is, MIRANDA takes the value 0 before 1966, 0.5 in
1966, and 1 in 1967 and subsequent years. This is known as an abrupt permanent model. n186
1966-1968. MIRANDA takes the value of 0 before 1966, 0.333 in 1966, 0.666 in 1967, and 1 in 1968 and
following years. Based on the available evidence of a short-term lag in police compliance with Miranda, this is probably

Page 14
50 Stan. L. Rev. 1055, *1095

the best model of Miranda's effects. This is known as a gradual permanent model. n187
1966+. MIRANDA takes the value of 0 before 1966 and the value of 1 in 1966 and subsequent years.
1967+. MIRANDA takes the value of 0 before 1967 and the value of 1 in 1967 and subsequent years.
1964-1968. MIRANDA takes effect from 1964-1968, assuming the value of 0 before 1964, 0.100 in 1964, 0.333 in
1965, 0.666 in 1966, 0.900 in 1967, 1 in 1968 and subsequent years. This model captures any pre-Miranda effect from
Escobedo or police "anticipation" of Miranda. n188
1964-1969. MIRANDA takes effect from 1964-1969, assuming the value of 0 before 1964, 0.166 in 1964, 0.333 in
1965, 0.500 in 1966, 0.666 in 1967, [*1096] 0.832 in 1968, and 1 in 1969 and subsequent years. This model spreads
out the Miranda effect over a five-year period.
As can be seen in Table IV, the sign of the MIRANDA variable is generally not sensitive to these alternative
specifications of Miranda's impact. n189
[SEE TABLE IN ORIGINAL]
Violent crime, property crime, larceny, and vehicle theft remain statistically significant across all assumptions, and
robbery and burglary are significant across virtually all assumptions. All of these categories are statistically significant
in the expected best (1966-1968) model. The only thing that fluctuates across assumptions is the value of the
MIRANDA coefficient. Generally the coefficients are larger as the impact is extended over a longer period of time,
which is consistent with the theory that Miranda's harmful effects took several years to fully develop. n190 In short, our
findings are not [*1097] dependent on assumptions concerning the timing and phase-in of the Miranda effect.
To test the strength of the competing hypothesis of Miranda's defenders, that "police officers adjusted to Miranda
over time and that any negative impacts quickly dissipated," n191 we also decided to model several alternative impact
structures. If Miranda did indeed affect clearance rates, even if only briefly, and its effects truly disappeared quickly,
then that impact would be faithfully modeled in one of the following:
Gradual temporary. As a baseline basic model of the theory offered by Miranda's defenders, we tested the
assumption that Miranda's impact would have been felt partly in 1966, completely in 1967, dissipated halfway in 1968,
and disappeared in 1969; that is, MIRANDA takes the value 0 before 1966, 0.5 in 1966, 1 in 1967, 0.5 in 1968, and 0 in
1969 and subsequent years. This is a gradual temporary model. n192
Extended gradual temporary. This model also assumed a gradual temporary effect, but further assumed that
Miranda's (and Escobedo's) effects were felt over a more extended period of time. Here MIRANDA assumed the value
of 0 before 1964, 0.25 in 1964, 0.5 in 1965, 0.75 in 1966, 1 in 1967, 0.75 in 1968, 0.5 in 1969, 0.25 in 1970, and 0 in
subsequent years. This is an extended gradual temporary model.
Abrupt temporary. This model assumed that Miranda's effects were immediately felt, but then faded away. Here
MIRANDA assumes the value of 0 before 1966, 1 in 1966, 0.75 in 1967, 0.5 in 1968, 0.25 in 1969, and 0 in 1970 and
following years. This is an abrupt temporary model. n193
As can be seen in Table V, these models do not fit the data well.
None of the coefficients are statistically significant at the conventional 95% confidence level, suggesting that the
theory that negative impacts were only temporary and then quickly dissipated does not fit the facts. n194
[*1098]
[SEE TABLE IN ORIGINAL]

Page 15
50 Stan. L. Rev. 1055, *1098

2. The length of the time series.
One caution that has been raised about the use of interrupted time series is the possibility that specification of the
beginning point or ending point of the data could affect the results. n195 Ordinarily, one should use all available data to
analyze an issue. n196 It can be instructive, however, to examine whether conclusions change over different time
periods because, it has been argued, "if the results of a time series regression analysis are a faithful representation of
underlying causal processes, the values of the estimated coef- [*1099] ficients will be independent of the specific time
period chosen for the analysis." n197
In our equations, two time periods are plausible candidates for exclusion from the analysis. First, we could run our
equations from 1960 forward, dropping the 1950s. The quality of our data improves significantly in 1960, as the FBI
has reported a consistent crime rate series ever since. The series has to be estimated if we want to include pre-1960 data.
n198 The FBI also revised some of its crime definitions in 1958, meaning that earlier data might not be strictly
comparable. n199 Running the equations from 1960 forward also eliminates potential problems in our "national" data
resulting from the admission of Hawaii and Alaska to the Union in 1959. n200 To answer these concerns, we reran the
regressions from 1960-1995. Second, Professor Schulhofer has suggested that extending the analysis into the 1980s and
1990s introduces "post-1980 data that could suppress the significance of variables that were causally important during
the critical 1960-1970 time period." n201 Why this might be so is unexplained. Nonetheless, we have considered this
possibility as well, by also running the regressions from 1950-1980.
Table VI reports the results of these respecifications, using the basic impact structure, over alternative lengths of
data. As can be seen, our results are insensitive to both respecifications. The MIRANDA variable remains statistically
significant across both data sets for violent crimes, robbery, property crimes, larceny, and vehicle theft. The only
difference is that the MIRANDA variable is not statistically significant at the standard 95% confidence level for
burglary using only 1950-1980 data.
[*1100]
[SEE TABLE IN ORIGINAL]
3. The appropriateness of a linear model.
In a brief essay anticipating the findings of our study, Professor Schulhofer has suggested that our linear regression
equations might not accurately measure the impact of an increasing crime rate on the criminal justice system. He argues
that the tide of rising crime during the 1960s might be analogized to "mounting pressure [that] finally took its toll - the
dam finally broke." n202 To account for such possibilities, Schulhofer argues that the regression equations must
therefore take into account "reserve capacities, lagged responses and the like." n203
We are skeptical about such an explanation for falling clearance rates. There is nothing in our equations that would
suggest that a linear model is inappropriate, as all of the summary statistics suggest that linear modeling provides a good
fit. n204 In contrast, a nonlinear, quantum change model is [*1101] unlikely to be a good predictor of aggregate
national crime statistics, which rest on the activities of thousands of criminal justice agencies across the country. Even if
one such agency should reach some kind of "breaking point," it is not clear why all other agencies would reach exactly
the same point at the same time. Criminal justice research has often used linear models to explain relationships between
such variables as crime and clearance rates, especially when modeling national data. n205 Finally, and perhaps most
important, the positing of an unspecified quantum change model of non-Miranda-induced changes in clearance rates is
almost equivalent to adopting, as an article of faith, that Miranda had no effect. Testing such a claim is virtually
impossible, as it requires a set of assumptions about the nature of the quantum shift that are unlikely to be susceptible to
observation and empirical review. For example, Schulhofer suggests that we model a "reserve" capacity of police
forces. n206 But, based on the objectively available data, there was no such capacity. In 1965, for example, roughly
75% of all index crimes were not cleared n207 and were presumably keeping law enforcement officers busy. What sort

Page 16
50 Stan. L. Rev. 1055, *1101

of "reserve" capacity should we add for 1965? Why should we assume that it would disappear in 1966, rather than 1967
or 1968 or 1969, etc.? Why don't the various impact structures we modeled account for any such "reserves"? Moving
from observable phenomena (such as the number of crimes and number of officers) to unobservable phenomena (such
as "reserve capacities") promises to make the Miranda debate turn on one's initial predispositions rather than on actual
empirical findings. One could simply characterize our Miranda effect as a "quantum shift in reserve capacity effect," but
we doubt that any but the most faithful adherents to the conventional wisdom about Miranda will be persuaded.
4. The issue of simultaneity.
All of our equations have been of the reduced form model, that is, we have specified only a single causal equation. This
requires the assumption that crime clearance rates were explained by various variables, but not vice [*1102] versa. It is
possible, of course, that crime clearance rates might influence some of the explanatory variables. In particular, a drop in
the clearance rate might cause crime rates to rise, because the lower the chance of apprehension, the greater the
expected rewards to crime. n208 Such a possibility is known as "simultaneity." Our model of crime clearance rates,
resting as it does on the reduced form model, does not capture these possible interactions. There is a considerable
literature suggesting that failure to consider simultaneity can lead to problems in predicting crime rates and the deterrent
effects of police. n209 However, the issue posed here is a slightly different one: whether failure to control for possible
simultaneity affects our conclusions about the MIRANDA variable.
Our conclusions are unlikely to have been substantially affected for several reasons. First, it is possible that there is
no simultaneity - that is, clearance rates might not affect crime rates. While we do not necessarily subscribe to this
counterintuitive theory, it is interesting that the literature on police effects on crime is conflicting. n210 Second, even if
simultaneity is an [*1103] issue, results consistent with those reported here might still be found. Simultaneity can bias
results not only upward, but also downward, or leave them unchanged. We do not think that simultaneity would explain
away the kinds of dramatic shifts in crime clearance rates that we are studying here. The only sophisticated
simultaneous equations study of the effect of the Supreme Court on crime rates reaches results similar to ours. n211 In
addition, our results are insensitive to the variables that we include or exclude in the equations. n212 Third, we do not
believe that simultaneity threatens our general findings. Reduced form equations can still generate policy-relevant
conclusions even if unable to resolve specific relationships among variables. n213
We acknowledge that, in theory, simultaneous equations for our crime clearance rate model would be the most
desirable way to proceed. In practice, however, the specification of such models is often highly dependent on
underlying assumptions. Indeed, some commentators have suggested that it is simply impossible to effectively model
simultaneous effects. n214 This is a quagmire that we hope, at least for present purposes, to avoid. n215
5. Extreme bounds analysis.
While OLS regression is a widely used statistical technique to quantify the relationship between sets of variables,
regression estimates can hinge on model specification (that is, which variables are included or excluded in the model).
Model specification can create the serious practical problem of "fragility" if results depend on the inclusion or exclusion
of certain variables [*1104] in an equation. n216 For example, our Miranda result would be fragile if it disappeared
when certain other explanatory variables were dropped from our equation.
Standard statistical significance tests cannot detect fragility. Yet conventional reporting of regression results often
presents only the "best-fitting" results or those results that conform to a researcher's prior beliefs. Since the
methodology for searching for the model is often not published along with the final model, the reader may think the
reported statistics represent a straightforward estimation process. Selective reporting of conventional inferential
statistics based on these types of searches may be misleading, since the sampling properties of the resultant estimators and thus those of the reported statistics - are not well-known.
A theoretically straightforward solution to these problems is to examine multiple combinations of the variables and

Page 17
50 Stan. L. Rev. 1055, *1104

see if the results are affected. The practical econometric technique that allows for the consideration of an infinite array
of models is known as extreme bounds analysis ("EBA"). n217 EBA computes the maximum and minimum
coefficients that could be obtained using maximum likelihood estimation over all possible linear combinations of
explanatory variables. The highest and lowest estimates are called the upper and lower bounds. If a negative or positive
coefficient could be generated from a regression depending on the variables included, the bounds straddle zero; these
bounds are known as "fragile" or "loose" bounds. If all possible combinations of the variables yield an unambiguous
sign on the coefficient of interest, the upper and lower bounds do not straddle zero; these bounds are known as
"nonfragile" or "tight" bounds. An illustration may clarify this. We have reported various equations showing that the
MIRANDA variable has a negative sign, suggesting that Miranda harmed clearance rates. If the inclusion of certain
variables is necessary to generate that negative sign, then the effect would be fragile. On the other hand, if all models
generated a consistent negative effect, then the result would be nonfragile.
To test the extreme bounds of the Miranda result here, we divided the explanatory variables into two subsets: focus
and doubtful. n218 The focus set was comprised of criminologically important variables that are always considered.
These variables include MIRANDA, CRIME RATE, POLICE PERSONNEL, POLICE DOLLARS (REAL),
PERSONNEL CAPACITY, [*1105] DOLLAR CAPACITY (REAL), and a constant term. The doubtful set was
comprised of variables that may or may not be included in every regression: labor force participation, unemployment,
per capital income (real), births to unmarried mothers, urbanization, juvenile population, crime in small cities, and trend
over time. n219
Table VII presents EBA results for the focus coefficients for each of the clearance rate dependent variables, using
the standard July 1966 start for the MIRANDA variable.
[SEE TABLE IN ORIGINAL]
The results demonstrate that the Miranda effect is robust with respect to model specification. As shown in the dark
shaded boxes, the bounds do not straddle zero for violent crimes, robbery, property crimes, larceny, and vehicle theft. In
other words, regardless of the model chosen, that is, the variables included in these equations, MIRANDA always had a
negative effect on clearance rates. For burglary, too, the EBA analysis suggested that a negative effect was more likely.
n220 For all of the other criminal justice system [*1106] variables, the results were fragile for every crime category that is, one could generate either a positive or negative effect by manipulating the variables included in the equation.
That the MIRANDA variable alone was able to produce nonfragile results suggests that our finding is not a mere
artifact of our models. Moreover, the nonfragile results in are surprising, since EBA is designed to expose the
fundamental fragility of regression modeling in the presence of collinear data. n221 Evidence of a Miranda effect is
accordingly not only statistically significant but also robust with respect to model specification.
F. Summary of the Range of the Miranda Effect
Having considered various models for the Miranda effect, it might be useful to set out a short summary of our findings
and the range of the possible effect. Table VIII displays the pertinent information.
[SEE TABLE IN ORIGINAL] [*1107] In the first column, we set out the 1995 clearance rate for the various
crime categories - for example, a 24.2% clearance rate for robbery. n222 Then we set out the range of the Miranda
effect found in the extreme bounds analysis just discussed. For example, depending on the model specification, Miranda
reduced robbery clearances somewhere from 1.6-7.2%. n223 To provide some context to these figures, we next set out
the percentage increase in clearances without the Miranda effect. For example, given that only 24.2% of robberies were
cleared in 1995, increasing the clearance rate by 1.6-7.2% would mean that 10.1-22.2% more robberies would be
cleared. n224 Because there may be some interest in the absolute number of crimes affected, n225 in the last column
we estimate how many more crimes would have been cleared in 1995 in the absence of the Miranda effect. n226 Our
equations suggest, for instance, that between 8000 and 36,000 more robberies would have been solved in 1995 in the
absence of the Miranda effect. It is worth emphasizing again that these estimates are quite conservative because they

Page 18
50 Stan. L. Rev. 1055, *1107

capture only Miranda's impact on crime clearances, missing some of the effects on prosecutions and convictions at later
points in the criminal justice system. n227 They are also conservative because they rely on regression analysis of
national crime data, which can only detect significant and simultaneous changes in aggregate police performance across
the country.
IV. Miranda As the Cause of the Clearance Rate Decline
So far we have presented evidence that, controlling for several major criminal justice and socioeconomic variables,
crime clearance rates for robbery, larceny, vehicle theft, and burglary fell significantly after Miranda. The question
immediately arises whether Miranda caused those drops. Regression analysis can never establish causality. Instead, the
causal conclusion can only come from combining the information provided by the regression equations with theory and
other information to determine whether a causal interpretation is a reasonable one. n228 The potential obstacle to con[*1108] cluding that Miranda was the cause of the 1966-1968 drop in clearance rates is the problem of alternate
causality - that is, perhaps some other change in society, unaccounted for in our regression equations, was responsible
for the reduction in clearance rates.
In interrupted time series analysis, especially without control groups, it is important to consider such alternative
causes. n229 We have made an aggressive effort to identify possible "omitted" variables that might influence our
conclusions n230 and discuss the various possibilities of such an omission in this section. n231 Before turning to these
other, thus-far-unconsidered candidates for the "X factor," we should first discuss briefly what we are looking for. To be
a strong candidate for the X factor, the variable must change sharply in the critical 1966-1968 period when clearance
rates fell - but not elsewhere. For example, Figure 4 shows that robbery clearances plunged from 37.6% in 1965 to
26.9% in 1968 and have remained relatively stable since. Our regression equations suggest that about half of this drop is
explained by a structural shift at the time of Miranda, controlling for such criminal justice variables as crime rates and
police personnel and such socioeconomic variables as unemployment and income. Now that we have used regression
analysis to rule out the influences of those factors, we next look to the relevant theoretical, anecdotal, and logical
explanations that could reveal what was the cause of the shift in the clearance rate: Miranda or something else.
A. Law Enforcement Assessment of the Drop
In assessing plausible causal factors for the decline in clearance rates, it is useful to first examine the FBI's
contemporaneous explanations. The FBI would have ample incentive to provide explanations for apparently declining
police performance. During the critical 1966-1968 period, the UCR authors listed the following explanatory causes of
falling clearance rates: "court decisions which have resulted in restrictions on police investigative and en- [*1109]
forcement practices, sharply increasing police workloads not limited to crime increases, an almost static ratio of police
to population not commensurate with the sharp rise in crime, and constant increasing criminal mobility." n232 The
regression equation controls for two of the factors cited: the increase in police workloads and the static ratio of police
strength. The last factor cited - increased mobility of those committing crimes - may have some long-term explanatory
power, n233 but seems an unlikely explanation for relatively sudden shifts in crime clearance rates. n234 Increasing
mobility could only affect clearances over the long haul. That leaves the first factor - "court decisions which have
resulted in restrictions on police investigative and enforcement practices" - as the logical candidate for explaining the
post-Miranda drop in clearance rates.
Confirming the FBI's assessment were the assessments from those who would know firsthand: law enforcement
officers who questioned suspects both before and after Miranda's constraints were imposed. Perhaps the best interviews
of officers on the streets were done by Otis Stephens and his colleagues in Knoxville, Tennessee, and Macon, Georgia
in 1969 and 1970. n235 Virtually all of the officers surveyed believed that Supreme Court decisions had adversely
affected their work, and most attributed this negative influence first and foremost to Miranda. n236 Similarly, in New
Haven, Connecticut, Yale students interviewed most of the detectives involved in the interrogations they observed
during the summer of 1966 and an additional twenty-five detectives. n237 They reported that "the detectives
unanimously believe [Miranda] will unjustifiably [help the suspect]." n238 They also reported that "the detectives

Page 19
50 Stan. L. Rev. 1055, *1109

continually told us that the decision would hurt their clear- [*1110] ance rate and they would therefore look
inefficient." n239 Also, law student Gary L. Wolfstone sent letters in 1970 to police chiefs and prosecutors in each state
and the District of Columbia. Most agreed that Miranda at least raised obstacles to law enforcement. n240 And, in
Seaside City, James Witt interviewed forty-three police detectives before 1973. n241 Witt reported that the detectives
"were in almost complete agreement over the effect that the Miranda warnings were having on the outputs of formal
interrogation. Most believed that they were getting many fewer confessions, admissions and statements." n242 Witt
also found the detectives to be "quick to refer to a decline in their clearance rate when discussing problems emanating
from the Miranda decision." n243
Concerns about Miranda continue to be expressed by law enforcement. n244 As one experienced detective recently
explained, "Most police are not really crazy about Miranda. It always gives the criminal that extra incentive not to say
anything." n245 But the more telling evidence is what the police thought at the time Miranda was handed down, when
they had experience both with and without the new rules. Those firsthand, contemporaneous reports are strong evidence
that Miranda was the cause of the clearance rate drop.
B. Other Supreme Court Decisions
Although police contemporaneously identified Miranda as a major obstacle, defenders of Miranda might argue that it is
impossible to single out Miranda as the most harmful decision among the Warren Court's various rulings. n246 This
argument suggests that the MIRANDA variable is more properly denominated a "Warren Court" variable. Even if this
assertion were true, our findings would still be of some importance, as legal academics have generally denied that the
Warren Court decisions impeded law enforce- [*1111] ment. n247 Nonetheless, there are strong reasons for believing
that Miranda was the Warren Court decision primarily - although perhaps not exclusively - responsible for declining
clearance rates.
Other Court decisions from the same time seem unlikely to have had as much effect on clearance rates. Ehrlich and
Brower cataloged the "significant" Warren Court decisions that possibly affected law enforcement. n248 In the critical
1966-1968 period, they identify, in addition to Miranda, two other Court decisions. The first, United States v. Wade,
n249 is one of several "lineup" cases decided by the Court in 1967 that suggested that the right to counsel extended to
such proceedings. n250 But lineups are probably implicated in a small percentage of cases overall, certainly a much
smaller percentage than interrogation. Moreover, the Court cut back on the applicability of the doctrine in 1972, n251
with the result that it "is now largely ineffectual." n252 Therefore, if the lineup decisions were responsible for the drop
in clearance rates, those rates should have later rebounded. Such a rebound does not exist. n253 The other significant
case identified by Ehrlich and Brower is United States v. Katz, n254 which involved the unusual investigative
technique of placing an electronic listening device on a telephone booth used frequently by a suspect. Although the
decision is doctrinally important because of its famous suggestion that the Fourth Amendment "protects people, not
places," n255 its effect on day-to-day police operations was probably relatively small; the Court specifically concluded
that the law enforcement agents could have done exactly the same thing, provided they obtained a warrant first. n256
Other potentially important cases during the 1966-1968 time period did not restrict police investigations, but rather
court adjudicatory processes - which begin only after a crime has been cleared. For example, [*1112] Hoffa v. United
States n257 dealt with the right to counsel before an indictment. n258 Similarly, In re Gault n259 extended various
due process procedures to juvenile court trials. n260 Chapman v. California n261 set new rules for appellate courts
reviewing whether the denial of certain constitutional rights during trial is "harmless error." n262 None of these
otherwise important decisions could have had much effect on police clearance rates.
It is also conceivable that earlier decisions affected clearance rates during the 1966-1968 period. n263 Professor
Schulhofer has cited the Court's decisions in Mapp v. Ohio n264 and Gideon v. Wainwright n265 as possible
competing impediments to law enforcement. n266 But the timing of these rulings makes them implausible candidates
to explain away the post-Miranda clearance rate drop. Mapp was decided in 1961 and Gideon in 1963. It seems unlikely
that they could explain, for example, the sudden changes in robbery clearance rates that appeared in 1966-1968, but not

Page 20
50 Stan. L. Rev. 1055, *1112

earlier. n267 Mapp's main effects, if any, n268 were probably felt in the early 1960s. n269 It is hard to under[*1113] stand why Mapp would cause clearance rates to begin falling at an accelerated pace some five to seven years
after the decision. Also, Gideon is a particularly poor candidate to influence clearance rates because it dealt not with
police investigations but with later court adjudications.
It is also possible that the Court's decisions on vagrancy and related issues might have influenced clearance rates.
Professor William Stuntz has explained how police once had rather substantial authority to stop and arrest persons on
loitering, vagrancy, and related grounds, n270 authority that is now restricted. n271 Although this declining authority
to arrest might result in declining opportunities for interrogation and hence confessions, the sharp 1966-1968 clearance
rate drop probably did not stem from these restrictions - the timing is off. The leading case striking down vagrancy laws
is Papachristou v. City of Jacksonville, n272 decided in 1972. Also, arrest rates for vagrancy, "suspicion," disorderly
conduct, and drunkenness declined over a much longer time period, from well before Miranda to well after, without
sharp changes in the late 1960s. n273 Finally, restrictions on police arrests for vagrancy would only indirectly affect
clearance for index crimes such as robbery.
[*1114]
C. Sundry Other Possible Causes
Apart from judicial decisions, there remain several other conceivable causes for the 1966-1968 clearance rate declines.
But none of them appears to be nearly so strong a candidate as Miranda.
One possible factor is changing patterns in illegal drug use. If drug use increased dramatically around the time of
Miranda, and if this led to a significant increase in crimes, and if drug-related crimes are harder to clear, then perhaps
this phenomenon could have caused the clearance rate drops. All of these links are questionable. Here again, the timing
does not quite fit. Use of illegal drugs appears to have increased from the early 1960s through the 1970s and 1980s.
n274 While data on illegal drug usage are notoriously unreliable, one possible measure is provided by the arrest rates
for narcotics violations. The data show sharply increasing numbers of arrests from 1965 (the first year for which
comparable data are available) to 1974, then a leveling off, then another sharp rise from 1980-1989. n275 The available
data thus do not suggest a unique, sharp 1966-1968 change in drug usage. Even if there had been a sharp surge in drug
usage during this limited period, the connection between drug use and crime is unclear. n276 Finally, it must be
remembered that the regression equations already control for the number of crimes committed. Thus, for drug usage to
be the X factor, drug-related crimes would need not only to have increased around the time of Miranda, but also to have
been uniquely hard to clear. There is no reason to suspect, for example, that a heroin-induced robbery would be harder
to solve than other robberies.
Another possibility is that police-citizen relations deteriorated during the 1960s, affecting crime clearance rates
around the time of Miranda. Police clashes with Vietnam War protesters and demonstrators for racial justice may have
reduced citizen cooperation, thereby drying up witnesses and leads needed to clear crimes. n277
[*1115] Although worsening relations could conceivably have played a role in declining clearance rates during the
1960s, we think it an unlikely candidate for triggering much of the 1966-1968 decline. Although quantification of
public attitudes is difficult, it is improbable that police-citizen relations would have deteriorated substantially across the
country over such a short period of time. Indeed, the Gallup Poll suggests increased respect for the police around the
time of Miranda. In April 1965, 70% of respondents across the country had a great deal of respect for the police, a
percentage that rose to 77% in August 1967. n278 Although we have been unable to locate consistent polling data on
public confidence in the police throughout the decade, the polls we have found reported high public confidence in the
police. n279 Any decline in trust for the police would have been a more long-run phenomenon, extending from the late
1950s through the mid-1970s, from the earliest days of the civil rights protests to the waning days of Vietnam War
protests and Watergate disillusionment. Such a decline would therefore not explain the sharp decline in clearance rates
of the late 1960s. Also, the late 1960s, like the years before and after, were a time of rising public concern about crime,

Page 21
50 Stan. L. Rev. 1055, *1115

n280 which might have produced greater, rather than lesser, willingness to help the police. Finally, it is quite possible
that declining confidence in the police was more concentrated in some parts of the country than others and in some
kinds of geographical areas than in others (e.g., large cities). Yet the sudden drops in clearance rate in 1966 and 1967
were reported by all population groups and all geographic divisions. n281
Even if public willingness to assist the police deteriorated in the 1960s, it seems likely to have improved at some
point since then. Indirect measures of public support for law enforcement - such as public support for the death penalty
or public concern about crime - have risen since the 1960s, n282 while [*1116] crime clearance rates have remained
relatively stable. Similarly, the Gallup Poll also found that the public who rated the honesty and ethics of police as high
or very high rose from 37% in 1977 to 49% in 1996. n283
A final problem for the public attitude explanation is that it seems unlikely to work for some crimes. For example,
motor vehicle theft is a crime that is generally and consistently reported to the police because of insurance requirements.
n284 Our Miranda effect on motor vehicle clearance seems accordingly unlikely to have been caused by declining
citizen cooperation.
Professor Schulhofer has advanced a hypothesis that appears to be the opposite of the worsening police relations
explanation for declining clearance rates. He suggests that reporting of crimes to the police might have actually
increased after Miranda, thereby bringing weaker, harder to solve cases into the system and, in turn, lowering clearance
rates. n285 In effect, this claim implies that relations improved rather than deteriorated, because crime reporting is
generally linked to greater confidence in the police. n286 However, Schulhofer's claim of increased reporting is made
without evidentiary support, and there is empirical evidence suggesting reporting has, in fact, remained constant, at least
since 1973. n287 Moreover, the claim suffers from many of the same flaws as the worsening police relations
hypothesis, in that it uses what would seem to be a long-term trend to explain a short-term de- [*1117] cline in
clearances. Also, the increased reporting theory fails to explain why clearances of motor vehicle theft declined after
Miranda, when that particular crime is usually consistently reported. n288
Another possible explanation for the drop in clearance rates would be a change in the definition of crimes
"cleared." n289 But FBI data collection criteria appear to have remained the same throughout the relevant period. n290
A related hypothesis is that improved police professionalism might have caused clearance rates to decline by reducing
artificial inflation of clearance rates. For example, Professor Schulhofer has claimed that "the overall movement for
greater professionalism ... occurred, coincidentally or not, almost exactly at the time of Miranda." n291 It would,
indeed, be a "coincidence" to find dramatic and apparently heretofore unrecognized changes in record-keeping
procedures that caused significant reductions in clearance rates. Professionalism is a quintessential long-term trend,
unlikely to have changed sharply during the relevant period. Bolstering this view is a recurrent theme in the police
literature - that police bureaucracies resist change imposed from the outside. n292 While it is possible to cite reports
recommending reforms at about the time of Miranda, n293 reform efforts - on record-keeping and other matters - were
not narrowly confined to the late 1960s. n294
[*1118] None of these explanations, either collectively or individually, begins to compete with Miranda as the
cause of the clearance rate declines. But, as a final note, it is worth observing that Miranda's defenders have articulated
a welter of conflicting theories on clearance rates. Professor Schulhofer's shifting claims are illustrative. His initial
position, widely adopted by Miranda's defenders, was that clearance rates did not decline permanently after Miranda,
n295 a position that collapsed in the face of FBI data to the contrary. n296 Schulhofer next maintained that "we need
only turn to trends in levels of crime and police resources during the [post-1965] period" to understand the clearance
rate decline. n297 However, empirical evidence demonstrated that the 1966-1968 drop in clearance rates did not follow
the pattern of rising crime rates throughout the 1960s and early 1970s. n298 Subsequently, Schulhofer adopted his
current view that, while "many forces contributed to clearance rate trends ... there is no reason to think that one
particular factor - Miranda - was among the factors playing a causal role." n299 In addition to rising crime rates and
reduced law enforcement resources, he has pointed to the limited "capacity" of the criminal justice system, urbanization,
the proportion of crimes in small cities, and other less easily quantifiable factors. n300 To the extent that his claims are

Page 22
50 Stan. L. Rev. 1055, *1118

empirically testable, they have been disproven: Our regression analysis suggests that, even controlling for all the
quantifiable factors, Miranda had a significant depressing effect on clearance rates, an effect found more consistently
than that for Schulhofer's competing causes. n301 The robustness of the Miranda explanation compared to the fragility
of the testable, competing explanations is itself strong evidence in favor of the handcuffing-the-cops position.
[*1119]
D. The Logic of Miranda As a Cause
Apart from Miranda, then, we have been unable to identify any strong candidate as the cause for the post-Miranda drop
in clearance rates. Because we have been focusing on a host of specific and narrow issues, it might be useful to step
back for a moment to look at the big picture: Is it logical to view Miranda as a causal factor?
The conclusion that Miranda caused a significant part of the 1966-1968 decline in clearance rates is supported by a
wide range of information. To begin with, there is an obvious, sharp drop in clearance rates between 1966 and 1968, at
exactly the time when the drop should have occurred if Miranda harmed law enforcement. n302 Next, the regression
equations indicate that the drop for robbery, vehicle theft, larceny, and burglary cannot be explained by major criminal
justice or socioeconomic variables. n303 Indeed, this finding remains statistically significant over a variety of
alternative formulations. n304 In addition, both the FBI and the police on the street contemporaneously identified
Miranda as a hindrance to clearing crimes. n305 These law enforcement reports are corroborated by declining
confession rates reported in the before-and-after studies of Miranda's impact, n306 lower confession rates reported in
this country after Miranda, n307 and higher confession rates reported in countries not following the Miranda rules.
n308
Finally, plain old common sense points to precisely the same result. Our conclusion is simply that when the
Supreme Court imposed unprecedented restrictions on an important police investigative technique, the police became
less effective. This was exactly what the dissenters predicted in Miranda, n309 a claim that the majority did not refute.
n310 In short, we are not asserting anything unusual; instead, we are merely suggesting the obvious. The proposition
that Miranda harmed law enforcement should be uncontroversial. Instead, it is the contrary academic wisdom that
receive our skepticism.
[*1120]
V. The Clearance Rate Decline As a Social Cost
Part IV suggested that Miranda caused a significant drop in crime clearance rates. This part contends that the drop in
clearance rates should be regarded as a social cost.
A. The Implausibility of Declining Coercion
Sometimes it is argued that clearance rates declined after Miranda, because the police were forced to abandon
unconstitutionally coercive questioning techniques. n311 On this view, declining clearance rates measure not the social
cost of criminals unfairly escaping, but rather the price of the social benefit of police abandoning impermissible
questioning techniques. Note that this argument implicitly concedes that clearance rates did fall because of Miranda.
However, this explanation of why Miranda produced the drop is far-fetched for several reasons.
First, genuinely coerced confessions were statistically rare at the time of Miranda. n312 Of course, one cannot
consult an FBI tally of the number of coerced confessions each year. Yet it appears to be common ground that, as the
result of increasing judicial oversight and police professionalization, coercive questioning methods began to decline in
the 1930s and 1940s. n313 By the 1950s, coercive questioning had, according to a leading scholar in the area,
"diminished considerably." n314 Thus, the Supreme Court's decision curtailing police interrogation techniques

Page 23
50 Stan. L. Rev. 1055, *1120

addressed a problem "that was already fading into the past." n315 Chief Justice Warren's majority opinion in Miranda,
though noting historical police abuses, acknowledged that such abuses are "undoubtedly the exception now" and that
"the modern practice of in-custody interrogation is psychologically rather than physically oriented." n316 Furthermore,
empirical surveys n317 provide good support for Professor Gerald Rosenberg's assessment: "Evidence is hard to come
by but what evidence there is suggests that any reductions that have been achieved [*1121] in police brutality are
independent of the Court and started before Miranda." n318
Coercive police questioning is not only limited to physical brutality, but also includes other techniques. It seems
unlikely, however, that such forms of coercion were so widespread that their elimination would have greatly changed
clearance rates. n319 In addition, statistics on motions to suppress confessions confirm that coercive techniques were
infrequently used by the time of Miranda. We should find frequent challenges to the voluntariness of confessions even
before Miranda if coerced confessions were in fact frequent. n320 Such motions, however, appear to have been rare
around the time of Miranda. n321
Besides the relative scarcity of coercion, there is another reason to believe that clearance rate reductions were not
caused by fewer coerced confessions: the nature of the Miranda rules themselves. Miranda was not particularly
well-designed as a shield against coercion. As Justice Harlan pointed out in his Miranda dissent, "Those who use
third-degree tactics and deny them in court are equally able and destined to lie as skillfully about warnings and
waivers." n322 In other words, police who used coercive tactics to obtain involuntary confessions would not
necessarily have felt compelled after Miranda to change their already-improper methods. n323 And even if they did so,
it was unlikely to have been so rapid a change as to produce a quick decrease in confession rates. Considering the low
coerciveness of question- [*1122] ing even before Miranda and the ineffectiveness of the Miranda requirements in
preventing coercion, it is unlikely that the 1966-1968 clearance rate drop is a manifestation of a change from a coercive
to a noncoercive system. n324
B. Components of the Clearance Rate Change
Even if crime clearance rates fell after Miranda, defenders of Miranda might argue that law enforcement was not really
harmed. Instead, they might claim, all that changed was the ability of law enforcement officers to get suspects to
confess to, and thus "clear," old unsolved crimes. Professor Yale Kamisar has taken this view. Responding to a
preliminary presentation of our thesis, he was quoted as saying that our interpretation of the clearance rate drop is too
simple in that "clearance rates can be very misleading because before Miranda, there is evidence police were using
so-called confessions to "clear' crimes that were never prosecuted." n325 In other words, clearance rates appeared high
before Miranda because police could get a suspect to confess many crimes, even though there was never sufficient
evidence to be sure that the suspect committed all of those crimes. n326
It is important to recognize that the minor premise of Professor Kamisar's argument is that the confession rate did in
fact decline after Miranda. Clearance rates for burglary, for example, would fall only if police were less successful in
interrogating burglars. Kamisar's argument, therefore, appears to concede implicitly that Miranda is an impediment to
productive police investigation.
[*1123] Professor Kamisar is analytically correct in suggesting that clearance rates could have fallen because law
enforcement was hampered in its ability get confessions to previously unsolved crimes. To examine this possibility
more precisely, it is important to understand that the clearance rate consists of two components: (1) the number of
arrests per crime and (2) the number of clearances per arrest. n327 Kamisar's ultimate claim is that a change in the
second component - clearances per arrest - is primarily responsible for the observed clearance rate drop n328 and,
apparently, that such a change is not a major law enforcement concern. Both of these positions are doubtful.
The theory that only the rate of clearances per arrest fell after Miranda is itself too simple. At bottom, this theory is
simply speculation about one possible explanation for the clearance rate drop after Miranda. No hard evidence suggests
that there were, at the time of Miranda, sufficient numbers of confessions to multiple crimes to account for the clearance

Page 24
50 Stan. L. Rev. 1055, *1123

rate changes we have reported. n329 More specifically, the available data suggest that clearance rates fell immediately
after Miranda not only because police were obtaining fewer statements from suspects but also because they were
making fewer arrests per crime. In other words, both clearances per arrest and arrests per crime fell after Miranda. FBI
data for violent crimes from 1965 n330 to 1991 on this issue are plotted in Figure 6.
[*1124]
[SEE FIGURE IN ORIGINAL]
As can be seen, there was a substantial decline in the number of arrests per violent crime at the time of Miranda. In
fact, as far as we can tell from the available data, more than half of the 1966-1968 drop in violent crime clearance rates
stemmed from reductions in arrests per crime. The number of arrests per crime fell sharply from 1965-1972 and has
continued to do so modestly since then. Clearances per arrest, on the other hand, have bounced around over the last
several decades. Although more research into this area is plainly warranted, n331 it is hard to read the data as showing
a permanent, [*1125] post-Miranda drop in clearance rates stemming only, or even primarily, from changes in
clearances per arrest.
Even were Professor Kamisar correct in pinpointing the cause of the clearance rate drop, his other claim - that
society is unharmed when police officers solve fewer crimes from each arrest - is wrong. Fewer cleared crimes leave
police investigators less able to focus effectively on crimes that require their attention. They are forced to spend hours
or days attempting to solve crimes that would have been solved by talking with a suspect for a few minutes. n332 The
paucity of police resources makes this unnecessary diversion a possible concern. n333 Multiple confessions are also
useful to help police officers discern the modus operandi (or MO) of professional criminals. n334
Uncleared crimes also harm crime victims, who never receive the reassurance that their attackers have been
apprehended. Evidence suggests that the principle harms suffered by victims of crime include psychological trauma.
n335 Without a clearance, they will likely continue to fear that they will be victimized again. n336 Victims also have a
hard time putting the crime behind them. Victims understandably want to see justice done, n337 and that is often not
possible until the crime is cleared. Victims of property crimes also benefit from each clearance, because the confession
creates the possibility of the return of stolen property. n338 For all these reasons, the harm to victims from declining
clearances is indeed a legitimate social concern. n339
Finally, a drop in clearances per arrest will inevitably be followed by a drop in convictions. If police clear fewer
crimes because they obtain fewer [*1126] statements from suspects, subsequent prosecutions will be impaired.
Kamisar conceivably could be correct that it makes little difference whether an arrested burglar confesses to four crimes
or to one. But it makes a considerable difference whether he confesses to one crime or none, as a confession is quite
important to prosecutors. n340 Miranda does not generally reduce the number of confessions from four to one; it
reduces them to zero by occasionally blocking police interrogation entirely. n341 It seems hard to view such declining
clearances as anything other than a serious social cost.
VI. Moving Beyond Miranda
Having analyzed the data on crime clearance rates, we are now in a position to return to the question we posed at the
outset: Has Miranda, in fact, handcuffed the cops? The answer to the question turns on what "handcuffed" means. It is
clear that our analysis provides no support for the overheated rhetoric of some of Miranda's detractors, who predicted
immediately after the decision that law enforcement would grind to a halt. n342 But we understand Miranda's
pragmatic critics today to make a more modulated claim: that Miranda has seriously impeded police effectiveness in
ways that could be avoided through reasonable changes in the Miranda rules.
Our findings support this more tempered attack. Our regression equations and accompanying causal analysis
suggest that, without Miranda, the number of crimes cleared would be substantially higher - by as much as 6.6-29.7%

Page 25
50 Stan. L. Rev. 1055, *1126

for robbery, 6.2-28.9% for burglary, 0.4-11.9% for larceny, and 12.8-45.4% for vehicle theft. n343 Moreover, applied
to the vast numbers of cases passing through the criminal justice system, these percentages would produce large
numbers of cleared crimes. As many as 36,000 robberies, 82,000 burglaries, 163,000 larcenies, and 78,000 vehicle
thefts remain uncleared each year as a result of Miranda. n344
To put Miranda's costs into some perspective, it is useful to compare them to the costs of the Fourth Amendment
exclusionary rule, long considered a major - if not the major - judicial impediment to effective law enforcement. n345
At the time the Supreme Court created the "good faith" exception to the Fourth Amendment exclusionary rule, it cited
statistics indicating that the rule resulted in the failure to convict from 0.6-2.35% of individuals [*1127] arrested for
felonies. n346 The Court pointed out that these "small percent- ages ... mask a large absolute number of felons who are
released because the cases against them were based in part on illegal searches or seizures." n347 Measured through
clearances, a quite conservative measure of Miranda's effects, n348 there appear to be more uncleared crimes from
Miranda than lost convictions from the search and seizure exclusionary rule. n349 Exact comparisons of the
percentages are problematic because of differing variables under study - persons arrested in the exclusionary rule
studies and crimes cleared in our Miranda study. However, this asymmetry likely understates the relative size of the
Miranda effect, because there are more crimes cleared than there are arrests made. n350 While other technical
adjustments would also have to be made to strictly compare the two figures, n351 it seems quite likely that Miranda's
costs exceed those of the exclusionary rule, especially outside of contraband offenses. n352
Taking the perspective of victims of crime underscores the severity of Miranda's costs. n353 A society concerned
for victims is obligated to do its best to avoid such miscarriages of justice as when a child abuser is set free because of a
Miranda technicality. n354 Hearing about cases in which confessions are obtained and then suppressed puts a human
face on Miranda's costs. But the far more frequent form of Miranda suppression occurs even before that, when a suspect
fails to even make the statement that could prevent the crime from going unsolved or unpunished. In concentrating on
statistical quantification of Miranda's costs, we hope we have not underplayed the human tragedies involved. As
Professor Gerald Caplan has concluded, the statistical studies "reduce crime to something remote and abstract, a string
of numbers, an event that one reads about in the newspapers, something that happens in another part of town. There is
no hint of rape as a [*1128] nightmare come alive, or robbery as a ruinous matter." n355 Moreover, we have not
discussed the inequitable distribution of the social consequences of these unsolved crimes. It is likely that these costs
fell most heavily on those in the worst position to bear them, including racial minorities and the poor. n356
Finally, it is a truism that an unnecessary cost is a cost that is too high. n357 If we can protect other values in our
criminal justice system while eliminating the costs of Miranda, regardless of whatever size they might be, then there is
no reason to cling to its constraints. What converts Miranda's harm into an undeniable tragedy is that these uncleared
crimes are largely unnecessary.
Today, with the benefit of thirty years of subsequent interpretations, we know the Miranda mandate is not a
constitutional requirement. n358 Rather, the Court has held specifically that Miranda rules are only "safeguards,"
whose purpose is to reduce the risk that police will violate the Constitution during custodial questioning. As explained
in Michigan v. Tucker, n359 Miranda established a "series of recommended "procedural safeguards' ... The [Miranda]
Court recognized that these procedural safeguards were not themselves rights protected by the Constitution but were
instead measures to insure that the right against compulsory self-incrimination was protected." n360 Based on this
rationale, the Tucker Court found that failure to give warnings that complied with Miranda "did not abridge
respondent's constitutional privilege ... but departed only from the prophylactic standards later laid down by this Court
in Miranda to safeguard that privilege." n361 Thus, the Miranda rules can be changed without impinging on the Fifth
Amendment.
The Miranda rules have no firm roots in the constitutional history of this country. n362 For many years, the
Constitution was generally understood as prohibiting the states from introducing coerced or involuntary statements
[*1129] against a defendant. Courts assessed "voluntariness" by looking at such factors as police pressure, the length of
questioning, and the suspect's mental and physical condition. Police compliance with an elaborate set of

Page 26
50 Stan. L. Rev. 1055, *1129

warning-and-waiver requirements was not thought relevant to that determination. As one ardent defender of Miranda
has conceded, the decision was a "radical departure" from the assumptions of the times. n363
The Court has taken Miranda beyond the bounds of the Constitution based on a purely pragmatic, cost-benefit
assessment. The Miranda decision, the Court has claimed, "embodies a carefully crafted balance designed to fully
protect both the defendant's and society's interests." n364 Thus, the Court weighs the costs and benefits, as it sees them,
of extending the "safeguards" of Miranda. Although the Court has never articulated with any precision what "costs" it is
willing to tolerate in this cost-benefit calculation, n365 it has in all likelihood significantly underestimated their
magnitude. Presumably the Court has simply plugged into its equation the supposed "negligible" costs of the decision,
as recounted in the conventional academic wisdom, n366 and determined that the equally unspecified "benefits"
outweigh them. Even under such charitable assumptions, the Court's conclusion is quite debatable. n367 But, if
substantial numbers of uncleared crimes are counted among the costs of Miranda, it is far more doubtful that the
calculus would produce the same conclusion.
The Court's calibration of Miranda's costs and benefits becomes even more problematic when the possibility of
reasonable, less harmful approaches to regulating police questioning is factored in. When the Court announced Miranda
in 1966, significant efforts to reform the rules regarding interrogations were already underway. n368 The decision itself
seemed to invite the continued exploration of such alternatives, promising that "our decision in no way creates a
constitutional straitjacket which will handicap sound efforts at reform." n369 This has proven to be an empty promise.
In the quarter of a century since Miranda, reform efforts have been virtually non- [*1130] existent. The reasons are not
hard to imagine. No state is willing to risk possible invalidation of criminal convictions by departing from Miranda. As
a result, the Office of Legal Policy concludes:

The Miranda decision has petrified the law of pre-trial interrogation for the past twenty years, foreclosing the possibility
of developing and implementing alternatives that would be of greater effectiveness both in protecting the public from
crime and in ensuring fair treatment of persons suspected of crime... Nothing is likely to change in the future as long as
Miranda remains in effect and perpetuates a perceived risk of invalidation for any alternative system that departs from
it. n370
The failure to explore other approaches cannot be attributed to lack of viable options. The states might, for example, be
allowed to videotape interrogations as a substitute for the Miranda procedures. Videotaping would better protect against
police brutality, end the "swearing contest" about what happened in secret custodial interrogation, and allow suspects
who are manipulated into falsely confessing to prove their innocence. n371 At the same time, even when coupled with
limited warnings of rights, videotaping does not appear to significantly depress confession rates. n372 Videotaping thus
might be the "win-win" solution to the problem of regulating police interrogations envisioned in Miranda's
encouragement to "Congress and the States to continue their laudable search for increasingly effective ways of
protecting the rights of the individual while promoting efficient enforcement of our criminal laws." n373
As another example, the states might be allowed to bring an arrested suspect before a magistrate for questioning.
Such a proposal was made more than sixty years ago by Paul Kauper; n374 similar proposals have since been made by
Judge Walter Schaefer, Judge Henry Friendly, and Akhil Amar and Renee Lettow. n375 Questioning under the
supervision of a magistrate would [*1131] offer more judicial oversight than Miranda but might also be structured in
such a way as to result in more evidence useful for society. However, just as with videotaping, the threat of lurking
constitutional issues n376 and the Court's failure to indicate whether this might be a permissible alternative to Miranda,
has relegated this proposal to nothing more than a favorite "hypothetical" for criminal procedure professors. n377
Finally, the Court might simply abandon the grand social experiment of Miranda and return to the long-standing
"voluntariness" test for the admissibility of confessions. n378 After all, this was the generally prevailing approach to
assessing confessions in this country for almost two centuries and is supported by the conventional notion that the

Page 27
50 Stan. L. Rev. 1055, *1131

Constitution ought to be interpreted consistently with the Framers' intent. n379 The voluntariness standard is also
adopted by an explicit (although as of yet largely untested) congressional directive, making it the touchstone for
admitting confessions in federal cases. n380
The evidence presented in this article argues in favor of earnestly considering these alternatives to Miranda. Justice
Harlan's dissenting opinion warned that the decision was a hazardous experiment with the country's safety, with
"harmful consequences" that "only time can tell." n381 The ex- [*1132] periment's results are now in. The clearance
rate data collected in this study - coupled with the other evidence concerning Miranda's effect n382 - strongly suggest
that Miranda has seriously harmed society by hampering the ability of the police to solve crimes. Indeed, based on
national crime clearance rates, a rough and conservative measure of its impact, Miranda may be the single most
damaging blow inflicted on the nation's ability to fight crime in the last half century. In short, it appears that Miranda
has, as its critics charge, handcuffed the cops. It is time to consider removing those shackles and regulating police
interrogation in less costly ways.
[*1133]
Appendix. Data Sources for the Regression Variables
It has been suggested that "it is of the utmost importance when reporting statistical results that authors report in the
fullest possible manner all relevant information that would enable others to judge the quality of the results and, if
necessary, to replicate them." n383 To that end, we explain in some detail the methodology for compiling our data.
A. The Dependent Variables - Clearance Rates
The dependent variables for the regressions were national crime clearance rates, published in the FBI's annual UCR.
1. National totals.
From 1958-1972, the FBI published a national clearance rate for total crimes. n384 This "grand total" included the
seven index crimes (i.e., murder, rape, robbery, assault, burglary, larceny $ 50 and over, and auto theft), but also
included manslaughter by negligence and larceny under $ 50. The FBI stopped reporting the clearance rate for total
crimes in 1973.
From 1965 to date, the FBI has published a national clearance rate for the seven index crimes. It is possible to
extend the crime index clearance rate backward to 1962 by taking the percentages reported in the crime clearance table
and "backing out" manslaughter by negligence and larcenies of less than $ 50. However, the reporting format before
1962 does not easily permit this adjustment, because only clearance rates for total larcenies were reported.
From 1965-1972, the index crime of larceny was defined as larceny of $ 50 and over. In 1973, the FBI changed the
crime index definition to include all larcenies, not just larcenies of $ 50 and over. n385 This means that there is a
discontinuity between the 1972 and 1973 index crime clearance rate series. n386 As a result, we analyzed clearance
rates for individual crime categories and the aggregate categories of violent and property crimes rather than for the
crime index.
2. Violent and property crimes.
The FBI has reported clearance rate data for "violent" crime (i.e., nonnegligent homicide, rape, robbery, and assault)
and "property" crime (i.e., burglary, larceny, and auto theft) from 1969 to date, although from 1969-1972 the property
crime category appears to include only clearance of larcenies $ 50 and over. For pre-1969 data, we used Dean James
Alan Fox's pub- [*1134] lished data for 1950-1974. n387 Dean Fox's data are based on a weighted average of reported
material from the FBI. We then switched to the FBI data from 1975 to date. Despite this "splicing," the Fox compilation

Page 28
50 Stan. L. Rev. 1055, *1134

corresponded quite well with the reported FBI violent crime data, and corresponded exactly with the property data. For
example, in 1974, the last year of Fox's tabulation, Fox's violent crime clearance rate was 44.9%; whereas the FBI's was
45.2%. The 1974 property crime clearance rate as tabulated in both sources was 18.5%. n388 The difference in the
violent crime category may stem from Fox's elimination 1957 of statutory rape from the rape category, and hence the
violent crime category in 1957. n389 In view of the small difference that this appeared to make in the overall violent
crime data, we made no adjustment to the post-1975 FBI data.
3. Individual crimes.
The FBI has reported clearance data for individual crime categories continuous from 1950 to date. A few caveats
should be noted. For the murder clearance rate, we used only "murder and nonnegligent manslaughter," excluding
"manslaughter by negligence," apparently comprised primarily of traffic fatalities. In addition, while the FBI has
continuous reported data on rape, in 1958 it eliminated statutory rape from the total. n390 We made no adjustment for
this change in the individual rape data, because we ran our data from 1960 forward, thus eliminating this issue. n391
For larceny, while the FBI has sometimes reported clearance rate data for larcenies $ 50 and over, it has consistently
reported data for total larcenies. Accordingly, total larceny data, and not the inconsistently defined crime index larceny
data, were used for the larceny clearance rate in this study.
4. City versus agency data.
From 1950-1979, the FBI only reported clearance rates for cities, not for the entire country. As a result, we used city
clearance rate data for the dependent variables in our equations. However, the independent variables generally rest on
data not just from cities but from the entire country.
While this slight mismatch might be a problem in theory, in practice we think it is not. The clearance rates for the
entire country probably track clearance rates for cities rather closely for three reasons. First, UCR figures are crime
figures, and cities have a large percentage of the crimes in the [*1135] country. n392 Thus, data from cities
substantially affect the UCR trends. n393 Second, clearance rates do not vary as greatly from jurisdiction to jurisdiction
as crime rates. For example, in 1995 the clearance rate for violent crimes for all agencies was within 5% of the
clearance rate for cities, and the clearance rates for property crimes were virtually indistinguishable. n394 However,
crime rates showed a wider variation. n395 Finally, it does not seem unreasonable to assume that trends affecting city
clearance rates would generally affect other police agencies as well.
To test the convergence of city clearance data with national clearance data, we compared the two sets of data from
1980, when the FBI began reporting these numbers, to the present. The correlation coefficient for the two series is
0.985. It therefore seems unlikely that our use of city data significantly affects our conclusions. n396
B. Independent Variables
1. Miranda variables.
Various dummy variables were used to measure Miranda's effects starting around July 1, 1966, as explained in more
detail in the text. n397
2. Crime rates.
Since 1950, the FBI has published an annual total number of crimes for the United States in the UCR. Because not all
law enforcement agencies reported figures to the FBI during the 1950s, the totals during those years are estimates. In
1975, the FBI undertook a revision of its earlier published figures in order to produce a consistent time series. As a
result, a continuous series of revised data from 1960 to date is now available for the FBI's seven index crimes (the
"official revision"). n398 In order to take advantage of the [*1136] FBI's latest calculation of total crime numbers, we

Page 29
50 Stan. L. Rev. 1055, *1136

have used the revised series here. It should be noted that even the unrevised data have been used in other criminal
justice research. n399
Unfortunately, the FBI does not appear to have published a revision of the clearance rates. As a result, we used
revised crime rate data to explain unrevised clearance rate data. This does not affect our conclusions, however, because
our other option using unrevised crime rate data would lead to the same results; the unrevised crime rate data do not
explain the changes in clearance rates. n400
To extend the revised series back from 1960-1950, we used an FBI revision that has not been formally published
(the "unpublished revision"). The revision is reprinted in Brower's dissertation and extends from 1933-1972. n401 For
all of the seven index crimes except larceny, the unpublished revision figures appeared to correspond closely to the
FBI's official revised figures for the overlap period of 1960-1972. n402 In order to splice the unpublished 1950-1959
data with the official revision, we first determined the ratio of the unpublished revision to the official revision figures
for each year for each index crime during the overlap period. n403 The mean ratio during the overlap period was then
applied to the 1950-1959 unpublished revision data for each index crime.
The unpublished revision for larceny did not appear as suitable for time series analysis for three reasons. First,
during the overlap period, the unpublished crime rate was approximately one-third of that in the official revision. n404
It is unclear whether the changed definition of larceny (e.g., in excess of $ 50 only) or some other factor produced this
disparity. Second, in contrast to the other crime categories, the ratio of the unpublished revision to the official revision
shifted significantly during the overlap period, from 27.3% in 1960 to 44.3% in 1972. n405
[*1137] In view of these difficulties with the unpublished revision of the larceny data, we used an alternative
methodology to construct a data series of estimated total larcenies for 1950-1959. The FBI had earlier published
estimates of total larcenies in the United States between 1950 and 1957, but this series is also not directly comparable to
the official revision. n406 To splice these two data sets together, we started from the 1960 total larcenies reported in the
official revision n407 and then used the reported change in the rate of total larcenies for U.S. cities to recalculate the
totals back to 1956. n408 We then compared the 1956 and 1957 numbers in this series to the FBI's earlier estimated
larceny data for 1956 and 1957. A ratio for each of the two years was determined and then averaged. We then applied
the average ratio the 1950-1955 estimated data to complete the series.
With the number of crimes determined, we then calculated crime rates by dividing by resident population for each
year. The official FBI revision does this for 1960 to date. We used Census data for the years 1950-1959 to calculate
crime rates during that period.
3. Number of police personnel.
Basic data for the number of police personnel come from FBI figures, which report the number of full-time law
enforcement employees, including [*1138] civilians, per one thousand inhabitants for all cities. n409 This series has
been used in other criminal justice research. n410
Because of variation in the dates at which the number of employees was determined (e.g., July 1, October 31, or
December 31), Dean Fox standardized the figures to July 1 from 1950-1974. n411 This article uses the Fox figures for
these years to ensure consistency during the critical period surrounding Miranda. For the years from 1975-1993, the FBI
figures were used, although they were determined as of October 31. We did not attempt to adjust these figures to July 1,
because we found that the Fox standardized figures mirrored the FBI figures for the overlap years of 1964-1974. In
addition, this article uses the rate of police personnel rather than the total number of personnel to eliminate spikes in the
data that would otherwise occur from differences in the number of agencies reporting. n412
In 1980, the FBI began reporting data not only for city police but also for police throughout the country
(denominated "total agencies"). n413 While the total agency percentage covers a substantially larger portion of the

Page 30
50 Stan. L. Rev. 1055, *1138

population, n414 this article uses only the police personnel figures for cities to maintain consistency in the time trend
and also to maintain consistency with the clearance rate numbers, which also come from city data. This approach does
not appreciably influence our results because the reported police rates for total agencies roughly track the city data.
n415
4. Dollars spent on police protection.
A data series on annual government expenditures for police protection year is available from the U.S. Bureau of the
Census compilation of govern- [*1139] ment finances. n416 This series has been used in other criminal justice
research. n417 The particular series used here was developed by adding local expenditures to state expenditures.
Federal expenditures were not included because they are not likely to be closely associated with crime clearance rates
for index crimes in major cities. A missing observation for 1951 was derived by averaging the immediately preceding
and following observations. Data for 1994 and 1995 were derived by assuming a continuation of the rate of expenditure
growth from 1990-1993. The series includes Alaska beginning in 1959 and Hawaii beginning in 1960. n418 However,
because these states' overall expenditures were such a small fraction of the total, no adjustment appeared necessary.
The Consumer Price Index ("CPI") was used to adjust expenditures for inflation. While the CPI's accuracy is
debated, n419 it appears to be a logical adjusting measure. n420
5. Socioeconomic variables.
Several socioeconomic variables were included in the equations. The unemployment rate is the standard annual
measurement collected by the U.S. Bureau of Labor Statistics. n421 It is expressed as the percentage of the civilian
labor force, aged sixteen or older, that is unemployed.
The civilian labor force participation rate comes from the U.S. Bureau of Labor Statistics and reflects the
employment status of the civilian population. n422 Income numbers come from the U.S. Bureau of Economic Analy[*1140] sis. n423 They reflect disposable personal income per capita, in constant 1987 dollars. This measure has been
used in other criminal justice research. n424 Data on live births to unmarried women come from the U.S. National
Center for Health Statistics, which measures births to residents based on marital status. n425 The percentage of the
resident population residing in urban areas was derived by fitting a third-degree polynomial function to the decadal data
from the U.S. Bureau of the Census. n426
6. Crime in small cities.
The percentage of violent crime in small cities was added to respond to a criticism of our earlier model raised by
Professor Schulhofer. n427 Accordingly, we tracked his specification of the variable, which was "the portion of the
total [of violent crimes] that was reported in cities with less than 250,000 residents." n428 Professor Schulhofer kindly
provided us with his data on this series, which he extracted from the FBI's UCR. n429
7. Population variables.
Population data comes from the U.S. Bureau of the Census. n430 The series for resident population was used,
consistent with the UCR. n431 The population data was used to generate a crime rate, officer rate,
dollars-on-law-enforcement rate, juvenile rate, and illegitimate birth rate.
Data on the population aged fifteen to twenty-four comes from the U.S. Bureau of the Census. n432 In 1959,
Alaskan and Hawaiian figures were included. This made a small difference in size of the percentage, increasing it by
roughly 0.4%. To eliminate, albeit imperfectly, this small incompatibility, the ratio of the population with fifty states to
the population with forty-eight states was determined for 1959. The 1950-1958 forty-eight-state [*1141] population
percentage was then multiplied by that ratio. During the 1970s, the Census Bureau appears to have categorized their age

Page 31
50 Stan. L. Rev. 1055, *1141

group differently, reporting data on the age groups 14-24, rather than for 15-24. For these years, we calculated the ratio
of persons aged 14-24 to persons aged 15-24 for 1970 and 1980. We then estimated population data for the 1970s by
assuming that the average moved constantly between the two ratios during the decade. n433
8. Time trend variable.
A time trend variable was constructed using standard statistical techniques.
C. Miranda and the Relative Importance of the Variables
As shown previously, the MIRANDA variable is robust with respect to model specification. n434 Additionally, we
tested the statistical importance of the Miranda effect from a predictive perspective using a forward stepwise regression
procedure. Stepwise regression is an automatic search mechanism for determining an optimal specification based on a
sequence of regression models. n435 The forward search method begins by finding the most important explanatory
variable in terms of reducing the total variation of the dependent variable. At each stage, an explanatory variable is
considered for inclusion and the program examines whether or not any of the preceding variables should be dropped in
light of the new performance of the most recent variable. n436
As shown in Table IX (violent crimes) and Table X (property crimes), the MIRANDA variable ranked high in
order of explanatory power (in the top three) in the regression models for which a statistically significant Miranda effect
was found. This supports the hypothesis that Miranda is a relatively important factor in explaining these clearance rates.
[SEE TABLES IN ORIGINAL] [*1142] [*1143] [*1144]
D. Components of the Clearance Rate
As discussed previously, n437 a clearance rate consists of two components: (1) arrests per crime and (2) clearances per
arrest.
[SEE EQUATION IN ORIGINAL]
While analytically this formula is straightforward, n438 finding a consistent data series for arrests is problematic.
For instance, Professor Fox's tabulation of FBI data reveals some significant gyrations from 1959-1963. n439 Fox's
arrest data also appear to be flawed during the critical 1965-1966 juncture. He reports that arrest rates for violent crimes
increased from 110.6 per 100,000 in 1965 to 139.9 per 100,000 in 1966, n440 a striking 26% increase for a year in
which crime rates themselves went up only 11%. n441 Fox probably never ironed out fluctuations in the arrest data
because they were ultimately excluded from his model. n442
Fortunately, the FBI has published a revised series of arrest rates for 1965-1992 that appears to be internally
consistent. n443 To derive (1), arrests per crime, we converted the arrest rates for violent crimes to actual arrests for
violent crimes by multiplying by Census Bureau population data. The [*1145] number of arrests per violent crime was
then derived by dividing by the FBI's reported number of violent crimes. n444
To derive (2), clearances per arrest, we multiplied the clearance rate by the FBI's reported number of violent
crimes. We then divided by the number of actual arrests for violent crimes as determined above.
These conversions required certain assumptions, particularly with respect to the number of clearances per arrest. As
with other aspects of our study, we assumed that the clearance rate from cities is comparable to that from other parts of
the country. n445 We also assumed that the cities used for the clearance rate are comparable to the cities used for the
arrest rate. The revised arrest rate series relies on a larger population base than does the clearance rate series. n446
There appears to be no way to determine the extent to which these issues might confound our analysis of clearances per
arrest. However, the percentage for arrests per crime is not affected by either of these problems.

Page 32
50 Stan. L. Rev. 1055, *1145

Legal Topics:
For related research and practice materials, see the following legal topics:
Criminal Law & ProcedureInterrogationMiranda RightsCustodial InterrogationCriminal Law &
ProcedureInterrogationMiranda RightsSelf-Incrimination PrivilegeCriminal Law &
ProcedureInterrogationVoluntariness
FOOTNOTES:

n1. 384 U.S. 436, 504, 517 (1966) (Harlan, J., dissenting).
n2. A 1974 ABA survey of lawyers, judges, and law professors found that Miranda was the third most
notable decision of all time, trailing only Marbury v. Madison and United States v. Nixon and leading Brown v.
Board of Education. See Jethro K. Lieberman, Milestones! 200 Years of American Law: Milestones in Our
Legal History at vii (1976).
n3. More Criminals to Go Free? Effect of High Court's Ruling, U.S. News & World Rep., June 27, 1966, at
32, 33 (quoting Los Angeles Mayor Samuel W. Yorty).
n4. .See id. (including a statement by Fred E. Inbau, Professor of Criminal Law at Northwestern University,
that law enforcement officials would choose not to prosecute a number of cases because of Miranda).
n5. Miranda, 384 U.S. at 504 (Harlan, J., dissenting).
n6. Id. at 542 (White, J., dissenting).
n7. S. Rep. No. 90-1097, at 25 (1968), reprinted in 1968 U.S.C.C.A.N. 2112, 2123.
n8. 114 Cong. Rec. 12,936, 12,937 (1968) (Mr. Mundt reading into the record Richard M. Nixon, Toward
Freedom from Fear (1968)); see also Liva Baker, Miranda: Crime, Law and Politics 248 (1983) (citing Nixon
campaign speeches attacking Miranda).
n9. See Gerald Caplan, Book Review, 40 Wayne L. Rev. 279, 281 (1993) (reviewing Joseph D. Grano,
Confessions, Truth, and the Law (1993)) ("We still lack sufficient data about Miranda's impact on the
administration of justice."); George C. Thomas III, Is Miranda a Real-World Failure? A Plea for More (and
Better) Empirical Evidence, 43 UCLA L. Rev. 821, 837 (1996) ("We need more empirical evidence [about
Miranda]. What we have so far raises more questions than it answers.").
n10. H. Richard Uviller, Tempered Zeal: A Columbia Law Professor's Year on the Streets with the New
York City Police 198 (1988).

Page 33
50 Stan. L. Rev. 1055, *1145

n11. See, e.g., Welsh S. White, Defending Miranda: A Reply to Professor Caplan, 39 Vand. L. Rev. 1, 20
(1986) (claiming a "widely shared perception that Miranda's effect on law enforcement has been negligible");
Yale Kamisar, 30 Years Later: Miranda Does Not Look So Awesome Now, Legal Times, June 10, 1996, at 22,
50 (reporting that post-Miranda studies indicated "that the impact of the Miranda rules on conviction rates had
been quite negligible").
n12. Yale Kamisar, Remembering the "Old World" of Criminal Procedure: A Reply to Professor Grano, 23
U. Mich. J.L. Reform 537, 585 (1990).
n13. 1 Wayne R. LaFave & Jerold H. Israel, Criminal Procedure 484 (1984 & Supp. 1991) (citation
omitted); see also Paul G. Cassell, Miranda's Social Costs: An Empirical Reassessment, 90 Nw. U. L. Rev. 387,
389 & nn.2-5 (1996) (collecting similar statements).
n14. See Richard A. Leo, The Impact of Miranda Revisited, 86 J. Crim. L. & Criminology 621, 645-46
(1996) (noting the "conventional wisdom" on Miranda's lack of harm but cautioning that the limited studies will
not support it).
n15. Moran v. Burbine, 475 U.S. 412, 433 n.4 (1986).
n16. Stephen J. Schulhofer, Miranda's Practical Effect: Substantial Benefits and Vanishingly Small Social
Costs, 90 Nw. U. L. Rev. 500, 505 (1996); see also Yale Kamisar, The "Police Practice" Phases of the Criminal
Process and the Three Phases of the Burger Court, in The Burger Years: Rights and Wrongs in the Supreme
Court 1969-86, at 143, 150 (Herman Schwartz ed., 1987) (noting that striking a balance between the defendant's
and society's interests "is the way Miranda's defenders - not its critics - have talked about the case").
n17. Miranda v. Arizona, 384 U.S. 436, 504 (Harlan, J., dissenting).
n18. We will not discuss other, perhaps less quantifiable, harms of Miranda. See, e.g., Grano, supra note 9,
at 199-222 (1993) (making a strong case for overruling Miranda as an "illegitimate" decision without reference
to its costs in terms of damage to law enforcement); Gordon Van Kessel, The Suspect As a Source of
Testimonial Evidence: A Comparison of the English and American Approaches, 38 Hastings L.J. 1, 129 (1986)
(discussing other costs of Miranda, such as consumption of police and judicial resources and the undermining of
public confidence in the criminal justice system).
n19. Caplan, supra note 9, at 281.
n20. See Cassell, supra note 13. For discussion of this assessment, compare the criticism of this analysis in
Schulhofer, supra note 16, with its defense in Paul G. Cassell, All Benefits, No Costs: The Grand Illusion of
Miranda's Defenders, 90 Nw. U. L. Rev. 1084 (1996).
n21. The term "confession rate" is used generally to embrace not only confessions but also incriminating
statements useful to the prosecution. See, e.g., Thomas, supra note 9, at 825.

Page 34
50 Stan. L. Rev. 1055, *1145

n22. See Cassell, supra note 13, at 396 (reporting results from and discussing Richard H. Seeburger & R.
Stanton Wettick, Jr., Miranda in Pittsburgh - A Statistical Study, 29 U. Pitt. L. Rev. 1, 12 tbl.1 & 13 tbl.3
(1967)).
n23. See Cassell, supra note 13, at 417, 418 tbl.1. In other words, if the confession rate was 60% before
Miranda, it was 44% (60% - 16%) after. But cf. Schulhofer, supra note 16, at 538, 539 tbl.1 (analyzing a
somewhat different set of studies and concluding that the before-and-after data show confession rates fell at
most 9.7% after Miranda).
n24. See Cassell, supra note 13, at 437-38. But cf. Schulhofer, supra note 16, at 545 & tbl.2 (concluding that
convictions are lost in 1.1% of cases if the post-Miranda regime is compared to one in which no warnings are
given (9.7% confession rate drop multiplied by 19% confession necessity rate)).
n25. Leo, supra note 14, at 676 n.243.
n26. See Cassell, supra note 13, at 450-51.
N27. See, e.g., Schulhofer, supra note 16, at 507-10.
n28. See Cassell, supra note 20, at 1088-89 (raising this possibility).
n29. See Paul G. Cassell & Bret S. Hayman, Police Interrogation in the 1990's: An Empirical Study of the
Effects of Miranda, 43 UCLA L. Rev. 839, 871 (1996); see also Christopher Slobogin, Criminal Procedure:
Regulation of Police Investigation: Legal, Historical, Empirical and Comparative Materials 6 (Supp. 1995)
(concluding that a 64% confession rate is "comparable to pre-Miranda confession rates"). But cf. George C.
Thomas III, Plain Talk About the Miranda Empirical Debate: A "Steady-State" Theory of Confessions, 43
UCLA L. Rev. 933, 935-36 (1996) (deriving a lower estimate with which to compare studies).
n30. See Cassell & Hayman, supra note 29, at 869 tbl.4. For an interesting although ultimately unpersuasive
argument that the Salt Lake County confession rate is actually higher, see Thomas, supra note 29, at 944-53.
n31. See Richard A. Leo, Police Interrogation in America: A Study of Violence, Civility and Social Change
255-68 (1994) [hereinafter Leo, Police Interrogation in America] (unpublished Ph.D. dissertation, University of
California (Berkeley)) (on file with the Stanford Law Review). Leo's informative dissertation has been published
in part as Richard A. Leo, Inside the Interrogation Room, 86 J. Crim. L. & Criminology 266 (1996) [hereinafter
Leo, Inside the Interrogation Room], and Leo, supra note 14.
n32. See Cassell & Hayman, supra note 29, at 926-30 (discussing Leo, Police Interrogation in America,
supra note 31, at 255-68). For criticism of the downward adjustment of Leo's figures, see Thomas, supra note 29,
at 953-54.
n33. See Gary D. LaFree, Adversarial and Nonadversarial Justice: A Comparison of Guilty Pleas and Trials,
23 Criminology 289 tbl.2 (1985).

Page 35
50 Stan. L. Rev. 1055, *1145

n34. See Cassell & Hayman, supra note 29, at 871-76.
n35. See Thomas, supra note 29, at 954-56 (raising this possibility).
n36. See, e.g., Federal Bureau of Investigation, U.S. Dep't of Justice, Uniform Crime Reports, Crime in the
United States 1995 (1996) [hereinafter UCR-[year]. Before 1958, the UCR was published twice a year. The first
edition of each year was entitled "Semiannual Bulletin" and included data for the first six months of the year.
The second edition, the "Annual Bulletin," contained data for the whole year. Citations herein are to the Annual
Bulletin for the given year, unless "Semiannual Bulletin" is specified. Crime clearance rates for 1950-1974 are
helpfully collected in James Alan Fox, Forecasting Crime Data: An Econometric Analysis 83-86 tbl.A-1 (1978).
n37. See Gerald N. Rosenberg, The Hollow Hope: Can Courts Bring About Social Change? 328 (1991)
("One of the ways in which police are often evaluated is through the "clearance' rate ... One of the major ways
this occurs is through confessions."). For an explanation of how clearance rates understate the effect on
confessions, see notes 47-49 infra and accompanying text.
n38. Obtaining statistics on the frequency of such interrogations is difficult. A 1967 study in Pittsburgh
found that detectives held 73 of 74 suspects after interrogation. See Seeburger & Wettick, supra note 22, at 24.
But the study appears to have examined only files of cases that had been cleared, see id. at 6, and thus likely
missed all of the cases in which a suspect refused to confess and the confession was necessary for clearing the
case. In New York City in 1967, only 4% of suspects were released after interrogation, and most of these had
not talked. See Vera Inst. of Justice, Taping Police Interrogations in the 20th Precinct, N.Y.P.D. 68 & n.27
(1967) [hereinafter Taping Police Interrogations]. This study appears primarily to have involved cases in which
suspects had been arrested and thus sheds little light on suspects not arrested.
n39. Fred E. Inbau, John E. Reid & Joseph P. Buckley, Criminal Interrogation and Confessions at xiv (3d
ed. 1986) (some letters capitalized in original).
n40. Leo, Police Interrogation in America, supra note 31, at 373.
n41. See Michael Wald, Richard Ayres, David W. Hess, Mark Schantz, Charles H. White- head II,
Interrogations in New Haven: The Impact of Miranda, 76 Yale L.J. 1519, 1595 (1967) (giving examples of such
clearances). The issue of multiple clearances from a single arrest is discussed at greater length below. See notes
325-331 infra and accompanying text.
n42. See, e.g., Seeburger & Wettick, supra note 22, at 20 (discussing clearance rate as a measure of
Miranda's effects); see also James W. Witt, Non-coercive Interrogation and the Administration of Criminal
Justice: The Impact of Miranda on Police Effectuality, 64 J. Crim. L. & Criminology 320, 330-31 (1973) (same).
n43. Stephen J. Schulhofer, Reconsidering Miranda, 54 U. Chi. L. Rev. 435, 456 (1987) [hereinafter
Schulhofer, Reconsidering Miranda] (linking the supposedly rebounding clearance rates to the notion that
Miranda "posed no barrier to effective law enforcement"); see also Stephen J. Schulhofer, The Fifth Amendment
at Justice: A Reply, 54 U. Chi. L. Rev. 950, 954 n.17 (1987) [hereinafter Schulhofer, Fifth Amendment]
(arguing that apparently steady clearance rates coupled with other evidence refute the notion that Miranda has

Page 36
50 Stan. L. Rev. 1055, *1145

harmed law enforcement).
n44. See, e.g., Special Comm. on Criminal Justice in a Free Soc'y, American Bar Ass'n, Criminal Justice in
Crisis 63 n.53 (1988) (collecting evidence that Miranda has not harmed law enforcement, including clearance
rates); White, supra note 11, at 18 n.93 & 19 n.99 (citing clearance and confession rates to argue Miranda did
not harm law enforcement); cf. Leo, supra note 14, at 645 (suggesting that Miranda has not significantly affected
crime clearance rates, although "in some instances" they may have dropped).
n45. See notes 232-245 infra and accompanying text (collecting law enforcement views on Miranda and
clearance rates).
n46. Otis Stephens apparently suggests that there is "doubtful utility" to measuring Miranda's effects
through "clearance rates, conviction rates, and the incidence of confession." Otis H. Stephens, Jr., The Supreme
Court and Confessions of Guilt 181 n.55 (1973). However, he appears primarily concerned with "differences in
the record-keeping procedures" of individual departments, see id., which is a concern that has little application
to a review of aggregate, national data for clearance rates, see note 103 infra and accompanying text. Professor
Schulhofer has also recently suggested that, to measure Miranda's effects, one should look to clearance totals
(that is, the total number of cleared crimes), in addition to clearance rates (that is, cleared crimes divided by
crimes reported to the police). See Stephen J. Schulhofer, Bashing Miranda Is Unjustified - And Harmful, 20
Harv. J.L. & Pub. Pol'y 347, 361-64 (1997) [hereinafter Schulhofer, Bashing Miranda]; Stephen J. Schulhofer,
Miranda and Clearance Rates, 91 Nw. U. L. Rev. 278, 285 (1996) [hereinafter Schulhofer, Clearance Rates].
However, this suggestion is misguided. See Paul G. Cassell, Miranda's "Negligible" Effect on Law Enforcement:
Some Skeptical Observations, 20 Harv. J.L. & Pub. Pol'y 327, 336-38 (1997) (explaining why clearance rates
rather than clearance totals must be used to assess law enforcement effectiveness).
n47. See Cassell, supra note 13, at 398-99 (arguing that clearance rates understate Miranda's adverse effect
on law enforcement).
n48. See Cassell & Hayman, supra note 29, at 908 tbl.15 (finding a statistically significant difference in
prosecutorial charging decisions between suspects who were successfully interrogated and those who were not).
n49. Compare, e.g., UCR-1965, supra note 36, at 103 tbl.12 (50,980 persons charged with violent crimes,
derived by summing four violent crime categories), and UCR-1966, supra note 36, at 104 tbl.16 (44,641 persons
charged with violent crimes, derived by summing four violent crime categories), with UCR-1967, supra note 36,
at 109 tbl.16 (50,515 persons charged with violent crimes, derived by summing four violent crime categories).
n50. See Monica A. Walker, Do We Need a Clear-Up Rate?, 2 Policing & Soc'y 293, 304 (1992)
(suggesting that in Britain conviction data are better than clearance data because they rest on known offenders as
opposed to suspects).
n51. See Cassell, supra note 13, at 396-98 (discussing why conviction rates cannot show how Miranda
hinders the investigative process).
n52. See Isaac Ehrlich & George D. Brower, On the Issue of Causality in the Economic Model of Crime and

Page 37
50 Stan. L. Rev. 1055, *1145

Law Enforcement: Some Theoretical Considerations and Experimental Evidence, 77 Am. Econ. Rev., May
1987, Papers & Proceedings, at 99, 104 (finding conviction rate data "highly questionable").
n53. See Letter from Bennie F. Brewer, Chief, Programs Support Section, Criminal Justice Information
Services Division, FBI, to Paul G. Cassell, Associate Professor of Law, University of Utah College of Law 1
(Feb. 7, 1995) [hereinafter Brewer Letter] (on file with the Stanford Law Review) (noting the discontinuation of
the FBI conviction rate series).
n54. See Leo, supra note 14, at 645-46 (concluding that the proposition that Miranda "has exercised only a
negligible effect on the ability of police to elicit confessions [and] solve crimes" has "become the conventional
wisdom among scholars").
n55. Schulhofer, Reconsidering Miranda, supra note 43, at 456.
n56. See note 78 infra and accompanying text.
n57. Kamisar, supra note 12, at 586 n.164; see also Janet E. Ainsworth, In a Different Register: The
Pragmatics of Powerlessness in Police Interrogation, 103 Yale L.J. 259, 299 n.200 (1993) (reporting that most
commentators believe "Miranda has had little negative effect on criminal prosecutions" (citing Schulhofer,
Reconsidering Miranda, supra note 43, at 455-58)); Louis Michael Seidman, Brown and Miranda, 80 Cal. L.
Rev. 673, 743 (1992) (concluding that Miranda "did not result in a huge decline in the number of confessions"
(citing Schulhofer, Reconsidering Miranda, supra note 43, at 455-61)); Marcy Strauss, Reinterrogation, 22
Hastings Const. L.Q. 359, 379 (1995) (arguing that "the most recent data is fairly consistent" that shortly after
Miranda ""clearance ... rates were thought to be returning to pre-Miranda levels'" (quoting Schulhofer,
Reconsidering Miranda, supra note 43, at 456)).
n58. See Schulhofer, Fifth Amendment, supra note 43, at 954 n.17; Schulhofer, Reconsidering Miranda,
supra note 43, at 456 n.52.
n59. One of the studies found exactly the opposite of Schulhofer's claim. See Neal A. Milner, The Court and
Local Law Enforcement: The Impact of Miranda 217 (1971) (finding substantial decreases in clearance rates in
early 1967 in three of four Wisconsin cities). The other study, the Pittsburgh study, offered mixed results, noting
at one point that "there has been a decline in the clearance rate from the first half of 1966. One of several
possible explanations for this is the imposition of the Miranda requirements on the Pittsburgh police." Seeburger
& Wettick, supra note 22, at 24.
n60. See Bernard Weinraub, Crime Reports up 72% Here in 1966; Actual Rise Is 6.5%, N.Y. Times, Feb.
21, 1967, at A1.
n61. Id. But cf. Daniel Nagin, General Deterrence: A Review of the Empirical Evidence, in Panel on
Research on Deterrent and Incapacitative Effects, National Research Council, Deterrence and Incapacitation:
Estimating the Effects of Criminal Sanctions on Crime Rates 95, 114-15 (Alfred Blumstein, Jacqueline Cohen &
Daniel Nagin eds., 1978) (suggesting that more accurate recordkeeping policies might explain the decline in the
New York clearance rate); Taping Police Interrogations, supra note 38, at 79-80 (questioning whether Miranda

Page 38
50 Stan. L. Rev. 1055, *1145

could have caused the reported changes in New York's clearance rate).
n62. See Witt, supra note 42, at 330-31. Witt cautioned that clearance rates actually rose in 1966 and then
fell in subsequent years. See id. at 331. Post-Miranda data also came from other jurisdictions, such as Omaha,
Nebraska, where clearance rates rose slightly after the decision, see Cyril D. Robinson, Police and Prosecutor
Practices and Attitudes Relating to Interrogation As Revealed by Pre- and Post-Miranda Questionnaires: A
Construct of Police Capacity to Comply, 1968 Duke L.J. 425, at 465 n.90 (citing statement of Detective
Inspector Richard R. Andersen of the Omaha Police Department to the Senate Judiciary Subcommittee on
Constitutional Amendments), and Austin, Texas, where clearance rates appeared to fall after Miranda, a drop
that is attributed in part to the decision, see Controlling Crime Through More Effective Law Enforcement:
Hearings Before the Subcomm. on Criminal Laws and Procedures of the Senate Comm. on the Judiciary, 90th
Cong. 726 (1967) [hereinafter Controlling Crime Hearings] (statement of Austin Chief of Police R.A. Miles).
n63. See note 384 infra and accompanying text for a description of grand total crimes. We do not consider
arson, which was only added to the Crime Index in the 1970s.
n64. UCR-1965, supra note 36, at 18 (citing 24.6% clearance for grand total crimes and 26.3% for the index
crimes). For reasons cited in Appendix.A.1, this article shall address only the seven index crime categories. See
also UCR-1964, supra note 36, at 2 (listing the seven Crime Index offenses and explaining that the FBI uses
them "to determine the trend of crime in the United States").
n65. See UCR-1966, supra note 36, at 27 ("Whereas police, nationally, cleared 26.3 percent of Crime Index
offenses in 1965, in 1966 this dropped to 24.3 percent. The decrease was noted in every Crime Index offense
with robbery solutions having the sharpest decline, down 14 percent." ).
n66. See id.
n67. Id.
n68. See UCR-1967, supra note 36, at 30 ("Whereas police nationally cleared 24.3 percent of these offenses
in 1966, this dropped to 22.4 percent in 1967. The decrease was noted in every Crime Index offense with auto
theft solutions having the sharpest decline ....")
n69. Id.
n70. See UCR-1968, supra note 36, at 30 ("Whereas police nationally cleared 22.4 percent of these offenses
in 1967, this dropped to 20.9 percent in 1968. The decrease was noted in every Crime Index offense.").
n71. See UCR-1969, supra note 36, at 28.
n72. See UCR-1970, supra note 36, at 30.
n73. See, e.g., UCR-1975, supra note 36, at 37; UCR-1980, supra note 36, at 180; UCR-1985, supra note

Page 39
50 Stan. L. Rev. 1055, *1145

36, at 154.
n74. To be clear, the point on the graph marked "Last Pre-Miranda Year" is 1965, the last pre-Miranda data
point. Because the FBI figures are reported annually, the 1966 number is the first to reflect Miranda's effects.
But cf. George C. Thomas III, Telling Half-Truths, Legal Times, Aug. 12, 1996, at 20, 24 (possibly
misunderstanding this point in claiming that clearance rates had already fallen substantially before the decision).
n75. See notes 165-178 infra and accompanying text.
n76. See Schulhofer, Reconsidering Miranda, supra note 43, at 456.
n77. See Schulhofer, Bashing Miranda, supra note 46, at 356; Schulhofer, Clearance Rates, supra note 46, at
279.
n78. See Schulhofer, Bashing Miranda, supra note 46, at 364-72; Schulhofer, Clearance Rates, supra note
46, at 280-85.
n79. See generally David McDowall, Richard McCleary, Errol E. Meidinger & Richard A. Hay, Jr.,
Interrupted Time Series Analysis (1980). Time series design permits "the analysis of the impact of interventions
upon a single time series of data." Id. at 5.
n80. See Gary A. Mauser & Richard A. Holmes, An Evaluation of the 1977 Canadian Firearms Legislation,
16 Evaluation Rev. 603, 604 (1992) ("Although unsophisticated, a visual inspection has the advantage of being
able to assess trends across time, so that possible links ... may be identified."). See generally William S.
Cleveland, Visualizing Data (1993) (discussing the importance of visualization to data analysis and the
application of various types of visualization tools). For illustrations of this technique, see David McDowall,
Colin Loftin & Brian Wiersema, Using Quasi-Experiments to Evaluate Firearm Laws: Comment on Britt et al.'s
Reassessment of the D.C. Gun Law, 30 L. & Soc'y Rev. 381, 390 (1996), and Franklin E. Zimring, Firearms and
Federal Law: The Gun Control Act of 1968, 4 J. Legal Stud. 133, 170-94 (1975).
n81. See notes 54-57 supra and accompanying text.
n82. See, e.g., Simon I. Singer & David McDowall, Criminalizing Delinquency: The Deterrent Effects of
the New York Juvenile Offender Law, 22 L. & Soc'y Rev. 521, 527 (1988) (discussing the use of control groups
for their case study).
n83. See Paul G. Cassell, The Costs of the Miranda Mandate: A Lesson in the Dangers of Inflexible,
"Prophylactic" Supreme Court Inventions, 28 Ariz. St. L.J. 299, 311 (1996) (noting the reluctance of police
agencies to explore alternatives to Miranda).
n84. It seems fair to number among Miranda's costs the fact that the decision has prevented controlled
experimentation in this area. See Cassell & Hayman, supra note 29, at 922 (concluding that Miranda prevented
research on important interrogation questions).

Page 40
50 Stan. L. Rev. 1055, *1145

n85. Schulhofer, Clearance Rates, supra note 46, at 291.
n86. See id. at 291 n.31 (erroneously citing as cross-sectional analysis a national time series study by Isaac
Ehrlich, The Deterrent Effect of Capital Punishment: A Question of Life and Death, 65 Am. Econ. Rev. 397,
406 (1975) (providing a national time series analysis of "aggregate crime statistics relating to the United States"
from 1933-1969)); see also Ehrlich & Brower, supra note 52, at 100, 104-06 (describing a national time series
analysis on questions similar to those discussed here and reaching similar conclusions). In addition, virtually all
of the data for Schulhofer's analysis of clearance rates come from a book by Professor James Alan Fox that is in
fact a national time series analysis. See Schulhofer, Clearance Rates, supra note 46, at 281 n.12 (noting that all
graphical data "are drawn from" Fox, supra note 36 (providing a national time series econometric equations
attempting to forecast crime)).
n87. See Schulhofer, Clearance Rates, supra note 46, at 291 n.31 (citing Raymond Atkins & Paul H. Rubin,
The Impact of Changing Criminal Procedure on Crime Rates (Oct. 28, 1995) (unpublished manuscript, on file
with the Stanford Law Review) (reporting a cross-sectional analysis of states that used the exclusionary rule
with those that did not before and after the Supreme Court's decision in Mapp v. Ohio), and John R. Lott, Jr. &
David B. Mustard, Crime, Deterrence, and Right-to-Carry Concealed Handguns, 26 J. Legal Stud. 1 (1997)
(reporting a cross-sectional analysis of jurisdictions that permit concealed handguns with those that do not)).
n88. In theory, such research might be possible if some jurisdictions followed Miranda rules before the
decision and some did not. Then it would be possible to compare changes in these differing groups. See, e.g.,
Atkins & Rubin, supra note 87, at 2; cf. Abraham N. Tennenbaum, The Influence of the Garner Decision on
Police Use of Deadly Force, 85 J. Crim. L. & Criminology 241, 247-48, 255-56 (1994) (supplementing national
interrupted time series data with state comparison data). However, before Escobedo v. Illinois, 378 U.S. 478
(1964), no state had adopted anything approaching the full set of the Miranda rules. Even during the 1964-1966
period, only a few states moved in the direction of Miranda, and they did not impose all the Miranda rules. See
Miranda v. Arizona, 384 U.S. 436, 519 n.17, 521 (Harlan, J., dissenting) (collecting cases and concluding that
"no State ... [has] chosen to go nearly so far on its own"); see also Cassell, supra note 13, at 493-96 (discussing
efforts to warn suspects after Escobedo and concluding that the largest drop in confession rates came from
Miranda's novel imposition of waiver requirements).
n89. It might be possible to compare American confession rates with overseas confession rates. The
available data on foreign confession rates support the conclusions reached in this article. See Cassell, supra note
13, at 418-22 (noting that post-Miranda confession rates are much lower in the United States than in Britain and
Canada). But cf. Thomas, supra note 29, at 942-43 (not contesting the lower foreign rates but arguing that
British and Canadian data are simply not comparable and therefore do not offer strong support in the Miranda
impact debate).
n90. In 1966, for example, the FBI-reported clearance rate data only for the nation and various regional
groupings (e.g., New England states, Middle Atlantic states, etc.). See UCR-1966, supra note 36, at 102 tbl.13.
State-by-state clearance data appear to be unpublished for the relevant time period. See Brewer Letter, supra
note 53, at 1 (indicating that state-by-state clearance rate data are available for purchase from 1979 to date). As a
result, the only feasible cross-sectional analysis of clearance rates is one based on these rather artificial regional
groupings, which seems likely to shed little light on Miranda questions. Moreover, even if clearance rate data
were available, other data limitations would also preclude effective analysis. See James Alan Fox, Reexamining
Some Perils of Quantification in the Econometric Study of Crime: A Reply to Decker and Kohfeld, 19 J. Res.
Crime & Delinq. 122, 123 (1982) (noting that many crime researchers have used national aggregates because

Page 41
50 Stan. L. Rev. 1055, *1145

"annual data on most exogenous variables (e.g., demographic composition and economic indicators) are
available only for a restricted range of units").
n91. See Nagin, supra note 61, at 113 (concluding that police departments that record fewer of the crimes
reported to them will have a higher clearance rate, generating a spurious correlation in cross-sectional deterrence
studies). Moving to smaller units would also inevitably introduce considerable "noise" into the data. See notes
102-105 infra and accompanying text (noting the advantages of national clearance rate data).
n92. See Donald T. Campbell, Reforms As Experiments, 24 Am. Psychologist 409, 417 (1969) ("The
interrupted time-series design ... is available for those settings in which no control group is possible, in which
the total governmental unit has received the experimental treatment, the social reform measure."); D.J. Pyle &
D.F. Deadman, Assessing the Impact of Legal Reform by Intervention Analysis, 13 Int'l Rev. L. & Econ. 193,
213 (1993) ("Legal reform should be an area in which intervention analysis is widely used."); H. Laurence Ross,
Donald T. Campbell & Gene V. Glass, Determining the Social Effects of a Legal Reform: The British
"Breathalyzer" Crackdown of 1967, 13 Am. Behav. Scientist 493, 495 (1970) ("Legal change is a subject for
which the Interrupted Time-Series design seems eminently suited.").
n93. See, e.g., H. Laurence Ross, Richard McCleary & Gary LaFree, Can Mandatory Jail Laws Deter Drunk
Driving? The Arizona Case, 81 J. Crim. L. & Criminology 156, 161 (1990); Singer & McDowall, supra note 82,
at 526-32; Zimring, supra note 80, at 171-94.
n94. See, e.g., Donald T. Campbell & Julian C. Stanley, Experimental and Quasi-Experimental Designs for
Research 37-42 (1966); Thomas D. Cook & Donald T. Campbell, Quasi-Experimentation: Design and Analysis
Issues for Field Settings 207-32 (1979). For particular studies that suggest using interrupted time series analysis
with care, see, for example, Chester L. Britt, Gary Kleck & David J. Bordua, A Reassessment of the D.C. Gun
Law: Some Cautionary Notes on the Use of Interrupted Time Series Designs for Policy Impact Assessment, 30
L. & Soc'y Rev. 361, 362-64 (1996) (recognizing the strengths but also identifying some problems with
interrupted time series research), and McDowall et al., supra note 80, at 381-90 (discussing the observations of
Britt et al., supra).
n95. See Charles R. Tittle, Crime Rates and Legal Sanctions, 16 Soc. Probs. 409, 411-12 (1969) ("The
unreliability of crime statistics is well-known, but the lack of other sources of data precludes alternative
approaches."). Victimization data has been used in preference to FBI-reported crime data in some other research,
but that is not possible here because victimization data was not collected on any significant scale until the 1970s,
well after Mirandawas decided.
n96. See Fox, supra note 36, at 8 ("Although there are inherent problems in using UCR data for research
purposes, these problems are minimized in analyses that employ aggregate time series data."); Scott H. Decker
& Carol W. Kohfeld, Certainty, Severity, and the Probability of Crime: A Logistic Analysis, 19 Pol'y Stud. J. 2,
6 (1990) (concluding that statistical analysis of "official crime statistics" can often be done "with considerable
confidence"); Ehrlich & Brower, supra note 52, at 104 (finding "no apparent technical reasons" to avoid using
FBI data for time series analysis); Walter R. Gove, Michael Hughes & Michael Geerken, Are Uniform Crime
Reports a Valid Indicator of the Index Crimes? An Affirmative Answer with Minor Qualifications, 23
Criminology 451 (1985) (defending UCR data generally); James A. Inciardi, The Uniform Crime Reports: Some
Considerations on Their Shortcomings and Utility, 6 Rev. Pub. Data Use 3, 14 (1978) (concluding that UCR
clearance rate trend data can be useful); Wesley G. Skogan, The Validity of Official Crime Statistics: An

Page 42
50 Stan. L. Rev. 1055, *1145

Empirical Investigation, 55 Soc. Sci. Q. 25, 37-38 (1974) (finding official crime statistics to be distributionally
accurate across cities); George D. Brower, The Supreme Court and the Growth of Crime 170 (1985)
(unpublished Ph.D. dissertation, State University of New York (Buffalo)) (on file with the Stanford Law
Review) (concluding that, "given appropriate care in the selection and preparation by researchers of particular
series," FBI annual data "are appropriate for use in empirical analysis").
n97. See James Q. Wilson & Barbara Boland, The Effect of the Police on Crime, 12 L. & Soc'y Rev. 367,
368 (1978).
n98. See, e.g., Jerome H. Skolnick, Justice Without Trial: Law Enforcement in Democratic Society 167-81
(2d ed. 1975) (providing examples of clearance rate manipulations); Wald et al., supra note 41, at 1580 n.161
(noting manipulation of clearance rate data); Yale Kamisar, How to Use, Abuse - and Fight Back With - Crime
Statistics, 25 Okla. L. Rev., 239, 252-53 (1972) (discussing problems with clearance rates); David Seidman &
Michael Couzens, Getting the Crime Rate Down: Political Pressure and Crime Reporting, 8 L. & Soc'y Rev.
457, 476 (1974) (describing political modification of crime data).
n99. See Controlling Crime Hearings, supra note 62 (statement of Robert W. Johnson, County Attorney,
Anoka County, Minnesota) ("Cleared cases can mean different things to different chiefs."); cf. Keith Bottomley
& Clive Coleman, The Police, in Interpreting Crime Statistics 44, 54-55 (Monica A. Walker ed., Royal
Statistical Soc'y Lecture Note Series No. 1, 1995) (criticizing British clearance statistics on the ground that
clearances are interpreted differently from jurisdiction to jurisdiction); Lawrence W. Sherman, Defining Arrest:
Practical Consequences of Agency Differences (Part I), 16 Crim. L. Bull. 376, 380 (1980) ("Clearance rates may
vary substantially among departments for reasons having nothing to do with the objective probability of getting
caught." (quoting Wilson & Boland, supra note 97, at 368)).
n100. See Peter W. Greenwood, An Analysis of the Apprehension Activities of the New York City Police
Department 18 (1970).
n101. See Bernard Greenberg, Oliver S. Yu & Karen I. Lang, Enhancement of the Investigative Function 10
(1972).
n102. Cf. Yehuda Grunfeld & Zvi Griliches, Is Aggregation Necessarily Bad?, 42 Rev. Econ. & Stat. 1, 10
(1960) ("Aggregation is not necessarily bad if one is interested in the aggregates.").
n103. See Fox, supra note 36, at 7 (concluding that the problem of data manipulation is "not overly
troublesome" for time series analysis that "does not involve cross-sectional data, but rather a time series from the
same population"); Charles R. Tittle & Alan R. Rowe, Certainty of Arrest and Crime Rates: A Further Test of
the Deterrence Hypothesis, 52 Soc. Forces 455, 456 (1974) (stating that although manipulations are possible,
"there are good reasons to doubt that they greatly contaminate the data ... Moreover, such biases would seem to
be distributed throughout the various police departments so that the validity of a study which examines internal
variations in the entire body of data ... would be unaffected.").
n104. See Gene Swimmer, The Relationship of Police and Crime: Some Methodological and Empirical
Results, 12 Criminology 293, 304 (1974) (claiming that the impact on the data by variability of reporting
techniques is minimized by grouping the types of offenses in larger "property" and "violence" categories); see

Page 43
50 Stan. L. Rev. 1055, *1145

also Fox, supra note 90, at 124 n.4 (noting that most econometric studies employ aggregated crime indicators).
n105. See Fox, supra note 36, at 127 n.11 ("Although the group of cities included in the FBI tabulations
does change annually, the extent of error resulting from these fluctuations is minimal relative to the aggregate
data."). The problem of changing jurisdictions in the UCR, like the other problems just discussed, reinforces the
desirability of focusing on nationaldata. National data tends to average out these fluctuations, allowing a view of
the "forest" rather than individual "trees."
n106. See Appendix.
n107. See Peggy S. Sullivan, Determinants of Crime and Clearance Rates for Seven Index Crimes 28 (1985)
(unpublished Ph.D. dissertation, Vanderbilt University) (on file with the Stanford Law Review) ("Social science
research has generally not focused on the clearance rate ... as a topic for study.").
n108. See, e.g., Thomas F. Pogue, Effect of Police Expenditures on Crime Rates: Some Evidence, 3 Pub.
Fin. Q. 14, 24 (1975) (finding that "clearance ratios are not well-explained by the variables included in the
regression equation"); Sullivan, supra note 90, at 174 (concluding after extensive modeling that "the models
used to predict the clearance rates in general were very poorly specified").
n109. See Fox, supra note 36, at 26 (advancing this hypothesis); Llad Phillips & Harold L. Votey, Jr., An
Economic Analysis of the Deterrent Effect of Law Enforcement on Criminal Activity, 63 J. Crim. L.
Criminology & Police Sci. 330, 335 (1972) (same); cf. Dale O. Cloninger & Lester C. Sartorius, Crime Rates,
Clearance Rates and Enforcement Effort: The Case of Houston, Texas, 38 Am. J. Econ. & Soc. 389, 396-97
(1979) (finding that the clearance rate in Houston, Texas, from 1960-1975 was inversely related to the crime
rate); Pogue, supra note 108, at 21 (suggesting that the number of clearances might be positively related to the
number of crimes, i.e., the number of potential clearances).
n110. Michael Geerken & Walter R. Gove, Deterrence, Overload, and Incapacitation: An Empirical
Evaluation, 56 Soc. Forces 424, 429-31 (1977).
n111. Nagin, supra note 61, at 119.
n112. Charles H. Logan, Arrest Rates and Deterrence, 56 Soc. Sci. Q. 376, 384 (1975).
n113. Compare Pogue, supra note 108, at 27 tbl.4 (finding an apparently statistically significant overload
effect for rape, larceny, and auto theft but not for other crimes), and Sullivan, supra note 90, at 171 (finding
support for overload hypothesis in cross-sectional study), with Richard R. Bennett, The Effect of Police
Personnel Levels on Crime Clearance Rates: A Cross-National Analysis, 6 Int'l J. Comp. & Applied Crim. Just.
177, 186 (1982) (finding, in a cross-national study, that an increase in the number of crimes increases police
ability to solve them); David F. Greenberg, Charles H. Logan & Ronald C. Kessler, A Panel Model of Crime
Rates and Arrest Rates, 44 Am. Soc. Rev. 843, 849 (1979) (finding no consistent, statistically significant relation
between either instantaneous or lagged effect of crime rates on arrest rates); and Eric Rasmusen, Stigma and
Self-Fulfilling Expectations of Criminality, 39 J.L. & Econ. 519, 520 (1996) (concluding that "the overload
theory ... cannot explain the U.S. pattern" of crime).

Page 44
50 Stan. L. Rev. 1055, *1145

n114. See Cassell, supra note 83, at 307-08 (noting that, although crimes rates rose throughout the 1960s
and early 1970s, the drop in clearance rates occurred in a shorter time span); Cassell, supra note 46, at 334-35
(noting that clearance capacity fell from1962-1965 while clearance rates rose and that a continued drop in
clearance capacity in the 1970s was accompanied by steady clearance rates).
n115. See notes 398-408 infra and accompanying text.
n116. See, e.g., Fox, supra note 36, at 26 (hypothesizing that clearance rates depend on the size of police
forces); Bennett, supra note 113, at 179 (hypothesizing that "personnel levels and police expenditures intuitively
relate to the ability of the police to clear crimes").
n117. See, e.g., Bennett, supra note 113, at 187 (reporting cross-national analysis suggesting that police
variables have minimal effect on clearance rates); Pogue, supra note 108, at 24 (reporting cross-sectional
analysis finding no support for effect of police expenditures or police manpower on clearance rates); Charles R.
Wellford, Crime and the Police: A Multivariate Analysis, 12 Criminology 195, 205-06 (1974) (reporting that
police variables did not account for much variation in clearance rates); Thomas F. Pogue, The Crime Prevention
Effects of Arrest and Imprisonment: Evidence from Multiple Cross-Section Analysis 31-35 (1983) (unpublished
manuscript, on file with the Stanford Law Review) (finding no statistically significant relationship between
criminal justice expenditures and clearance rates); Sullivan, supra note 90, at 169 (reporting cross-sectional
analysis suggesting that police variables have no effect on clearance rates for violent crime but positive effects
for property crimes); cf. Dale O. Cloninger, Enforcement Risk and Deterrence: A Re-examination, 23 J.
Socio-Econ. 273, 281 (1994) (finding that police presence is superior to clearance rates as a measure for
deterrence studies).
n118. Ideally, we should control for the number of police personnel on the streets. See Wilson & Boland,
supra note 97, at 377 (noting that the correlation between the number of officers and the number of patrol units
on the street is low). Unfortunately, national data on this subject are unavailable.
n119. See Schulhofer, Clearance Rates, supra note 46, at 291.
n120. Specifically, we took the previously defined variables POLICE PERSONNEL and POLICE
DOLLARS and divided them each by CRIMES (the total number of index crimes committed in a given year).
Because the numerator and denominators contain rates, the effect is to express the capacity variables in terms of
personnel/crimes and dollars/crimes - e.g., (personnel <div> population) <div> (crimes <div> population) =
personnel <div> crime. This is precisely as Schulhofer has suggested. See Schulhofer, Bashing Miranda, supra
note 46, at 365.
In an earlier, preliminary report of our findings, we generated the personnel capacity by dividing the rate of
employees by the number of crimes and the dollar capacity by using nominal dollars. These differences, along
with a data set that extended only to 1994, produced slightly different coefficients from those reported here. See
Cassell, supra note 46, at 339 tbl.1 (reporting preliminary regression results).
n121. . Another commonly cited demographic variable is racial composition of the population. We have not
controlled for this factor because, among other reasons, long-term racial changes are unlikely to explain
short-term clearance rate fluctuations and because the empirical support for an association with clearance rates is
thin. See Sullivan, supra note 90, at 30, 163-64 (concluding from the existing literature that the impact of

Page 45
50 Stan. L. Rev. 1055, *1145

minority groups on the clearance rate is "unknown" and then reporting her own cross-sectional analysis finding
no influence of minority groups on clearance rates).
n122. See generally Travis Hirschi & Michael Gottfredson, Age and the Explanation of Crime, 89 Am. J.
Soc. 552 (1983); Darrell Steffensmeier, Cathy Streifel & Miles D. Harer, Relative Cohort Size and Youth Crime
in the United States, 1953-1984, 52 Am. Soc. Rev. 702 (1987).
n123. See, e.g., Task Force on Assessment, President's Comm'n on Law Enforcement and Admin. of
Justice, Task Force Report: Crime and Its Impact - An Assessment app. D (1967) [hereinafter Crime and Its
Impact]; Darrel Steffensmeier & Miles D. Harer, Did Crime Rise or Fall During the Reagan Presidency? The
Effects of an "Aging" U.S. Population on the Nation's Crime Rate, 28 J. Res. Crime & Delinq. 330, 343 (1991).
n124. See Kamisar, supra note 98, at 246 (identifying the 15-24 age band as having the most influence on
crime rates); Steffensmeier & Harer, supra note 123, at 331 (same).
n125. An argument could be made for using only males because crimes are most often committed by males.
However, such a refinement would likely make little difference to our results, as the ratios of males and females
remain relatively constant in this age range.
n126. See, e.g., Isaac Ehrlich, Participation in Illegitimate Activities: A Theoretical and Empirical
Investigation, 81 J. Pol. Econ. 521, 544 (1973) (using percentage of males ages 14-24 as a variable); Tittle &
Rowe, supra note 103, at 457 (controlling for population aged 15-24 in a deterrence study). But cf. D. Wayne
Osgood, Patrick M. O'Malley, Jerald G. Bachman & Lloyd D. Johnston, Time Trends and Age Trends in Arrests
and Self-Reported Illegal Behavior, 27 Criminology 389, 412 (1989) (cautioning against the use of broad age
bands in some types of criminal justice research); Steffensmeier & Harer, supra note 123, at 336 (same).
n127. Compare M. Harvey Brenner, Subcomm. on Economic Goals and Intergovernmental Policy of the
Joint Economic Comm., 98th Cong., Estimating the Effects of Economic Change on National Health and Social
Well-Being, S. Prt. 98-198, at 54-56 (Comm. Print 1984) [hereinafter Effects of Economic Change] (reporting
associations between unemployment and crime rates); M. Harvey Brenner, Book Review, 70 J. Crim. L. &
Criminology 273, 274 (1979) [hereinafter Brenner, Book Review] (reviewing Fox, supra note 36) (arguing that
Fox improperly excluded unemployment from crime forecasting models); Wilson & Boland, supra note 97, at
374-75 (including unemployment rate as a control variable in a study of robbery rates); M. Harvey Brenner,
Economy, Society & Health 6 (Oct. 16, 1992) (unpublished manuscript, presented at the Conference on Society
& Health, on file with the Stanford Law Review) (same); and Pogue, supra note 117, at 41 (finding an
association between unemployment and crime), with Fox, supra note 36, at 29 (reviewing literature and
concluding that "the absence of an impact of the unemployment rate on the rate of crime appears at this time to
be unequivocal"); Ehrlich, supra note 126, at 555 (finding no relation between unemployment and crime rate but
wondering about this conclusion); Kenneth C. Land & Marcus Felson, A General Framework for Building
Dynamic Macro Social Indicator Models: Including an Analysis of Changes in Crime Rates and Police
Expenditures, 82 Am. J. Soc. 565, 566 (1976) (finding no relationship between crime and unemployment); and
Swimmer, supra note 104, at 309 (reporting that unemployment rate had little explanatory power in criminal
justice models).
n128. See Effects of Economic Change, supra note 127, at 54, 57 (noting that a decrease in per capita
income had an inverse effect on most serious crimes and that, surprisingly, there was a positive correlation

Page 46
50 Stan. L. Rev. 1055, *1145

between labor market participation and arrest rates for the 16-17 age group); Swimmer, supra note 104, at 309
(finding that "median income relates significantly with fewer violent crimes but not fewer property crimes").
n129. Cf. Sullivan, supra note 90, at 165 (finding a relationship between murder clearances and the
percentage of families with single mothers).
n130. See Kamisar, supra note 98, at 247.
n131. See Schulhofer, Bashing Miranda, supra note 46, at 366-68 (noting that "clearance rates are
consistently lower in the larger series" and suggesting that as a greater percentage of the population (and of the
crimes committed) are in large cities, the national clearance rate will be affected).
n132. See UCR-1995, supra note 36, at 199 tbl.25 (providing that clearance rates for index crimes were
16.9% for the nation's seven largest cities (1,000,000+ residents), 21.3% for medium-sized cities
(100,000-249,999 residents), and 25.9% for smaller cities (10,000-24,999 residents)).
n133. See Pogue, supra note 117, at 33 tbl.5 (finding urbanization not statistically significant in explaining
clearance rates and population density related only to robbery clearance rates); Sullivan, supra note 90, at 161-62
(finding urban variables "virtually insignificant" in explaining most clearance rates); cf. Pogue, supra note 108,
at 27 tbl.4 (finding that population density reduced clearance rates to a statistically significant degree for overall
crime rates, murder, and auto theft, but not for rape, robbery, aggravated assault, burglary, or larceny).
n134. See Schulhofer, Bashing Miranda, supra note 46, at 366-67 (suggesting this control for violent
crimes).
n135. See id. at 367.
n136. In theory, one should look at the percentage of all crimes committed in small cities, not just violent
crimes. However, using the larger number of index crimes might tend to smooth out the fluctuations that could
explain away the Miranda effect. Accordingly, to give maximum play to Schulhofer's theory, we use - as he has
suggested we should - the violent crime percentage.
n137. See, e.g., Charles R. Nelson & Heejoon Kang, Pitfalls in the Use of Time As an Explanatory Variable
in Regression, 2 J. Bus. & Econ. Stat. 73 (1984).
n138. See McDowall et al., supra note 80, at 387-88 (defending the use of an abrupt, permanent change
model and the law's effective date in gun control time series research).
n139. See text accompanying notes 165-194 infra.
n140. See, e.g., Richard A. Hay, Jr. & Richard McCleary, Box-Tiao Time Series Models for Impact
Assessment: A Comment on the Recent Work of Deutsch and Alt, 3 Evaluation Q. 277, 304-05 (1979)
(cautioning against "overmodeling").

Page 47
50 Stan. L. Rev. 1055, *1145

n141. See text accompanying notes 228-301 infra.
n142. See text accompanying notes 208-215 infra.
n143. All statistical significance tests reported in this article are two-tailed, although an argument could be
made for a one-tailed test. We use the standard 95% confidence level for interpreting our results. We also report
which of our results are significant at the 90% level to assist those who would use the lower confidence interval.
n144. See UCR-1995, supra note 36, at 199 tbl.25.
n145. See id. (listing the clearance rate for property crimes as 17.6%).
n146. Cf. Lott & Mustard, supra note 87, at 7 (raising a similar concern in a gun control study).
n147. In the disaggregated equations, we continue to use the crime rate for all index crimes as the
independent variable, rather than the crime rate for the particular index crime. We thought this was the superior
model because police overload is presumably a function of the entire police workload, rather than the workload
from a particular crime.
n148. In Part IV below, we discuss in detail why we attribute our results to a Miranda effect rather than
some other cause.
n149. Because our focus is on the effects of Miranda, we will not discuss other effects we found or failed to
find in specific equations - e.g., the effects of police and economic variables on clearance rates. Unlike our
Miranda effect, these other effects appear to be "fragile" in the sense that they depend on model specification to
produce them. See notes 216-221 infra and accompanying text (presenting extreme bounds analysis of the
equations).
n150. See Cassell, supra note 13, at 464-66 (citing studies that show that repeat offenders may be less likely
to confess than those without criminal records).
n151. See Greenwood, supra note 100, at 24 tbl.3.
n152. See id. The following ratios of clearance rate to arrest rate can be derived from Greenwood's data:
1.045 for homicide; 1.063 for rape; 1.660 for robbery; 1.073 for assault; 3.778 for burglary; 2.564 for grand
larceny; and 2.416 for grand larceny vehicle.
n153. For further discussion of this issue, see notes 325-331 infra and accompanying text.
n154. See David Simon, Homicide: A Year on the Killing Streets 17-19 (1991) (giving an example of this
for the Baltimore police).

Page 48
50 Stan. L. Rev. 1055, *1145

n155. See Kamisar, supra note 98, at 252; cf. Joan Petersilia, Allan Abrahamse & James Q. Wilson, Police
Performance and Case Attrition 40 (Rand No. R-3515-NIJ 1987) (noting that police work "hardest on the most
promising cases ... and, to a degree, on the most important ones").
n156. See note 117 supra (collecting conflicting evidence on this issue); cf. Tables I & II (finding sporadic
effect of police resources on clearances).
n157. See Sullivan, supra note 90, at 8 (suggesting this possibility).
n158. These figures are derived from subtracting the post-Miranda rates from the pre-Miranda rates listed in
Seeburger & Wettick, supra note 22, at 11 tbl.1.
n159. See id. at 20 (finding overall clearances in Pittsburgh higher after Miranda than before).
n160. See id. at 21 tbl.9.
n161. See id. This reading of the Pittsburgh study, however, does not explain why auto larceny and burglary
clearances rose slightly, see id., even in the face of declining confessions.
n162. UCR-1993, supra note 36, at 285. A recent edition of the UCR contained the following data on
homicides, estimating the percent of total murder victims attributable to each category:
[SEE TABLE IN ORIGINAL]
See id. at 285 tbls.5.3-5.5.
n163. FBI data before 1965 on offender-victim relationships in homicide cases do not appear to be regularly
published in a consistent format. See Letter from Bennie F. Brewer, Chief, Programs Support Section, Criminal
Justice Information Services Division, FBI, to Paul G. Cassell, Professor of Law, University of Utah College of
Law 1 (Jan. 28, 1997) (on file with the Stanford Law Review).
n164. See Lawrence E. Cohen & Marcus Felson, Social Change and Crime Rate Trends: A Routine Activity
Approach, 44 Am. Soc. Rev. 588, 600 tbl.5 (1979). The authors report offense trends for only four crimes:
robbery, burglary, larceny, and murder. Although their analysis is based on data apparently collected as part of
the UCR program, the data does not appear to be published in the annual crime reports and it is unclear whether
the data is readily available.
n165. See McDowall et al., supra note 79, at 75 (concluding that many social science impacts will be
gradual).
n166. Schulhofer, supra note 16, at 513.
n167. See Wald et al., supra note 41, at 1550-56 (discussed in Cassell, supra note 13, at 407-08).

Page 49
50 Stan. L. Rev. 1055, *1145

n168. See id. at 1550 tbl.3 (showing increasing compliance with Miranda from June through August, but
not full compliance in every interrogation).
n169. Richard J. Medalie, Leonard Zeitz & Paul Alexander, Custodial Police Interrogation in Our Nation's
Capital: The Attempt to Implement Miranda, 66 Mich. L. Rev. 1347, 1394 (1968). The study's data presentation
is criticized in Leo, supra note 14, at 639.
n170. See Albert J. Reiss, Jr. & Donald J. Black, Interrogation and the Criminal Process, 374 Annals Am.
Acad. Pol. & Soc. Sci. 47, 55 n.6 (1967) (reporting that few suspects received complete Miranda warnings in
field interrogations). The proper interpretation of Reiss and Black's findings is unclear, however, because they
do not specify which of the field interrogations involved custodial interrogation, the predicate for application of
Miranda.
n171. Taping Police Interrogations, supra note 38, at 75-76.
n172. See Lawrence S. Leiken, Police Interrogation in Colorado: The Implementation of Miranda, 47 Denv.
L.J. 1, 10, 14, 47 (1970) (finding "strict compliance with [Miranda's] formal requirements"). The study
suggested that police sometimes ignored suspects' refusals to waive rights. See id. at 30.
n173. See id. at 30 (reporting allegations by suspects that police continued interrogations after requests for
counsel).
n174. A study in Wisconsin also possibly followed quasi-observational methodology, but contains very
little specific information on police practices. See Neal Milner, Comparative Analysis of Patterns of Compliance
with Supreme Court Decisions: "Miranda" and the Police in Four Communities, 5 L. & Soc'y Rev. 119 (1970).
A 1969-1970 survey of police officers in Tennessee and Georgia found that 36% (18 of 50) said that they would
continue to question a suspect if the suspect first waived his right to presence of counsel and then later changed
his mind by attempting to invoke his Miranda rights. See Stephens, supra note 46, at 196 tbl.10. However, this
finding has limited significance because this scenario (of suspects initially waiving and subsequently invoking
their Miranda rights) occurs infrequently. See Cassell & Hayman, supra note 29, at 860 (finding that suspects
waived and then asserted rights in only 3.9% of interrogations in the sample); Leo, Inside the Interrogation
Room, supra note 31, at 275 tbl.2 (finding the percentage to be 1.1%). Other studies from around the time of
Miranda contain only conclusory information about police compliance with the decision. See Cassell, supra note
20, at 1088 n.26 (collecting citations).
n175. See Cassell & Hayman, supra note 29, at 848-49 (finding almost no studies on police interrogations
since 1970); Thomas, supra note 9, at 822 (same).
n176. See Cassell & Hayman, supra note 29, at 887-92 (collecting references and evidence to this effect).
n177. Van Kessel, supra note 18, at 102-03; see also Leo, supra note 14, at 645 (concluding after
comprehensive review of the studies that "in the years 1966-1969 after an initial adjustment period American
police began to comply regularly with the letter of the new Miranda requirements"); Thomas, supra note 29, at
953 (suggesting that the "starting point" for Miranda research should be "after the controversy had subsided and

Page 50
50 Stan. L. Rev. 1055, *1145

the police had adjusted to its strictures ... in the decade of the 1970s").
n178. See David Dixon, Keith Bottomley, Clive Coleman, Martin Gill & David Wall, Safeguarding the
Rights of Suspects in Police Custody, 1 Policing & Soc'y 115, 122 (1990) (finding a rapid increase in requests
for legal advice by juveniles in British city from 1984-1987 under new interrogation regime, due in part to the
spread of information). The claim is not that suspects responded to Miranda by changing the kinds of crime that
they were committing. Cf. Philip J. Cook, The Clearance Rate As a Measure of Criminal Justice System
Effectiveness, 2 Criminology Rev. Y.B. 669 (1980) (discussing this problem in deterrence studies). Instead, the
claim is that some suspects responded to Miranda by offering fewer incriminating statements, thereby making
crimes harder to solve. See Cassell, supra note 13, at 450-51 ("There is some support for the hypothesis that the
population has now, in effect, been "Mirandized.'").
n179. 378 U.S. 478 (1964).
n180. Yale Kamisar, Wayne R. LaFave & Jerold H. Israel, Modern Criminal Procedure 465 (8th ed. 1994).
n181. Id.
n182. The evidence is collected in Cassell, supra note 13, at 493-94. For example, the "clear consensus" of
the attendees of the May 1996 annual meeting of the National Association of Attorneys General was that
Escobedo had little effect on confession rates. See Sidney E. Zion, Prosecutors Say Confession Rule Has Not
Harmed Enforcement, N.Y. Times, May 18, 1966, at 27.
n183. See, e.g., United States ex rel. Russo v. New Jersey, 351 F.2d 429 (3d Cir. 1965) (requiring police to
advise suspects of a right to counsel), vacated, 384 U.S. 889 (1966); People v. Dorado, 398 P.2d 361 (Cal. 1965)
(en banc) (requiring California police officers to warn suspects of their rights to counsel and to remain silent).
n184. See Schulhofer, Clearance Rates, supra note 46, at 291-92 (noting the potential effects of an
inappropriate time period selection on regression analysis); see also Britt et al., supra note 94, at 368-70
(explaining that the specification of the model of the impact of a gun control law has significant influence on the
validity of the conclusions); McDowall et al., supra note 80, at 385-86 (same).
n185. See, e.g., McDowall et al., supra note 80, at 386.
n186. See Sir Maurice Kendall & J. Keith Ord, Time Series 222 (3rd ed. 1990).
n187. See id.
n188. An anticipation effect would occur if police changed their procedures before receiving an expected
legal mandate. Other time series analyses have considered the possibility of such an anticipation or
"announcement" effect preceding the actual imposition of a legal change. See, e.g., Glenn L. Pierce & William J.
Bowers, The Bartley-Fox Gun Law's Short-Term Impact on Crime in Boston, 455 Annals Am. Acad. Pol. &
Soc. Sci. 120, 124 (1981) (finding an announcement effect during the month prior to the actual effective date of

Page 51
50 Stan. L. Rev. 1055, *1145

a gun-control law). In the Miranda context, while it is possible that law enforcement agencies "anticipated" the
Miranda decision by adjusting some of their procedures, see Milner, supra note 174, at 128 (noting that police
may have "adjusted their education and interrogation procedures in anticipation of such a decision"), it does not
appear that they anticipated the most harmful features of Miranda, specifically the waiver and questioning
cut-off rules, by complying early. See Cassell, supra note 13, at 494-96.
n189. We also ran a more complex polynomial distributed lag over a several-year period. The results were
generally consistent with the results reported in Table IV.
n190. Cf. McDowall et al., supra note 80, at 385-86 (observing that incorrect placement of the effective date
of a legal change can result in underestimating the magnitude of the effect of the change).
n191. Schulhofer, supra note 16, at 507.
n192. See Kendall & Ord, supra note 186, at 222.
n193. See id.
n194. Cf. Britt et al., supra note 94, at 369 (arguing that hypothesized effects of a legal change should be
tested and, if the model does not fit, the hypothesis should be rejected).
n195. See id. at 370-71; David Cantor & Lawrence E. Cohen, Comparing Measures of Homicide Trends:
Methodological and Substantive Differences in the Vital Statistics and Uniform Crime Report Time Series
(1933-1975), 9 Soc. Sci. Res. 121, 135 (1980); McDowall et al., supra note 80, at 389-90.
n196. See, e.g., McDowall et al., supra note 80, at 389.
n197. William J. Bowers & Glenn L. Pierce, The Illusion of Deterrence in Isaac Ehrlich's Research on
Capital Punishment, 85 Yale L.J. 187, 197 (1975). But see Isaac Ehrlich, Deterrence: Evidence and Inference, 85
Yale L.J. 209, 214-17 (1975) (disputing this conclusion).
n198. See notes 398-408 infra and accompanying text.
n199. See Crime and Its Impact, supra note 123, at 20 ("Figures prior to 1958 ... must be viewed as neither
fully comparable with nor nearly so reliable as later figures."); UCR-1958, supra note 36, at 2-3 (noting
recommended changes in the UCR program). For present purposes, the most significant change seems to be
limiting the figures for rape to forcible rape. See UCR-1958, supra note 36, at 2. It has also been reported that
the definition of larceny changed in 1960. See Cohen & Felson, supra note 164, at 601.
n200. See note 433 infra and accompanying text (noting this problem).
n201. Schulhofer, Clearance Rates, supra note 46, at 292.

Page 52
50 Stan. L. Rev. 1055, *1145

n202. Id.
n203. Id. at 293.
n204. One measure of fit is provided by the Durbin-Watson statistic, which is a measure of autocorrelation.
In a nutshell, autocorrelation is a violation of an assumption that the error terms in the regression model will
exhibit no discernable pattern.Incorrect model specification (for example, fitting a linear function to a nonlinear
function) can produce autocorrelation. See Damodar N. Gujarati, Basic Econometrics 353-59 (2d ed. 1988)
(illustrating this point). In some of our equations, the Durbin-Watson statistic was "inconclusive" on the
presence of autocorrelation. However, in none of our equations did the Durbin-Watson statistic conclusively
indicate autocorrelation.Furthermore, in a number of our equations, notably the robbery equations, the
Durbin-Watson statistic conclusively rejected the presence of autocorrelation, suggesting that our linear model is
appropriate.
n205. See, e.g., Fox, supra note 36; see also Peter Passell & John B. Taylor, The Deterrent Effect of Capital
Punishment: Another View, 67 Am. Econ. Rev. 445, 448 (1977) (criticizing nonlinear criminal justice models
and noting that "forms are usually chosen which are linear in the parameters to make it easier to interpret the
statistical properties of the estimators").
n206. See Schulhofer, Clearance Rates, supra note 46, at 293.
n207. See UCR-1965, supra note 36, at 18.
n208. See generally Gary S. Becker, Crime and Punishment: An Economic Approach, 78 J. Pol. Econ. 169
(1968); cf. Atkins & Rubin, supra note 87, at 20 (finding the creation of the exclusionary rule had an effect on
crime rates starting in 1961).
n209. For good introductions to the problem, see Nagin, supra note 61, at 117-29, and Swimmer, supra note
104, at 293-303.
n210. For an illustrative sample of this voluminous literature, compare, for example, Morgan O. Reynolds,
Crime and Punishment in America 23 (National Center for Policy Analysis, Policy Report No. 193, June 1995)
(finding a deterrent effect); Cloninger & Sartorius, supra note 109, at 398-99 (providing step regression
equations finding that rates for some crimes were responsive to clearance rates); Decker & Kohfeld, supra note
96, at 15-18 (finding increased clearance rates deter robbery, but not burglary or larceny); Ehrlich & Brower,
supra note 52, at 104-06 (reporting simultaneous models with results generally consistent with a deterrent
effect); Steven D. Levitt, Why Do Increased Arrest Rates Appear to Reduce Crime: Deterrence, Incapacitation,
or Measurement Error?, 35 Econ. Inquiry (forthcoming 1998) (finding a general deterrent effect from increased
arrest rates); Lott & Mustard, supra note 87, at 57 (finding that higher conviction rates resulted in lower crime
rates); Stephen J. Markman & Paul G. Cassell, Protecting the Innocent: A Response to the Bedau-Radelet Study,
41 Stan. L. Rev. 121, 154-56 (1988) (collecting evidence of a deterrent effect from capital punishment); Wilson
& Boland, supra note 97, at 367-68, 375-76 (collecting evidence that increased police resources reduce crime
and reporting result that aggressive police patrols reduce robbery rates); Ann Dryden Witte, Estimating the
Economic Model of Crime with Individual Data, 94 Q.J. Econ. 57, 74-78 (1980) (finding a deterrent effect of
higher conviction rates for certain crimes in individual data); and Kenneth I. Wolpin, A Time Series-Cross

Page 53
50 Stan. L. Rev. 1055, *1145

Section Analysis of International Variation in Crime and Punishment, 62 Rev. Econ. & Stat. 417, 421 (1980)
(finding increased clearance rates associated with decreased level of robbery), with, for example, David H.
Bayley, Police for the Future 7 (1994) ("In all probability, the relationship does not run from clearances to crime
but from crime to clearances."); David F. Greenberg & Ronald C. Kessler, The Effect of Arrests on Crime: A
Multivariate Panel Analysis, 60 Soc. Forces 771, 784 (1982) (using multiwave panel models to deal with
simultaneity issues and finding "no consistent evidence for the proposition that higher arrest clearance rates
result in substantially lower index crime rates"); Greenberg et al., supra note 113, at 849 (using panel models to
deal with simultaneity issues and concluding that "aggregate criminal activity for the F.B.I. index offenses is not
substantially influenced by marginal variations in arrest clearance rates within the range found in our sample of
American cities"); Herbert Jacob & Michael J. Rich, The Effects of the Police on Crime: A Second Look, 15 L.
& Soc'y. Rev. 109, 110, 120 (1980-1981) (collecting evidence that police effect on crime is mixed and reporting
the result that in some cities police activities increase robbery rates); Thomas F. Pogue, Offender Expectations
and Identification of Crime Supply Functions, 10 Evaluation Rev. 455, 478-79 (1986) (finding little effect of
arrest rates on crime); and Pogue, supra note 117, at 40 (finding clearance rates are a minor source of
cross-sectional variation in crime rates). Also, see generally John J. Donohue III & Peter Siegelman, Is the
United States at the Optimal Rate of Crime? (ABF Working Paper Preliminary Draft, Feb. 13, 1995) (reviewing
the literature on the police effect on crime and concluding that the literature "does not speak with an
authoritative voice").
n211. See Ehrlich & Brower, supra note 52, at 104-06.
n212. See notes 216-221 infra and accompanying text.
n213. Cf. Philip J. Cook, The Clearance Rate As a Measure of Criminal Justice System Effectiveness, in 2
Criminology Review Yearbook 669, 676 (Egon Bittner & Sheldon L. Messinger eds., 1980) (stating, in a
deterrence study, that even though the reduced form equation could not distinguish between two hypotheses, it is
still relevant for policy analysis); Sullivan, supra note 90, at 171 (suggesting that certain models "can be used to
predict even if they cannot be used to establish the causal relationship").
n214. See, e.g., Cook, supra note 213, at 676; Nagin, supra note 61, at 129 (hoping that information for
dealing with simultaneity issues in deterrence will develop).
n215. With the capable assistance of Katarzyna Celinska, we have run some preliminary simultaneous
equations. For example, we have created a three-stage model in which, in addition to the clearance rates
equation in this paper, equations also exist for crime rates affected by clearance rates and for police personnel
levels affected by lagged crime rates. All the models we have tried so far continue to show the same strong
Miranda effect found in our reduced form model. We hope to be able to publish a paper on this issue in the
future.
n216. For more technical descriptions of the problem in various contexts, see Richard Fowles & Peter D.
Loeb, Speeding, Coordination, and the 55-MPH Limit: Comment, 79 Am. Econ. Rev. 916, 920 (1989); Richard
Fowles & Mary Merva, Wage Inequality and Criminal Activity: An Extreme Bounds Analysis for the United
States, 1975-1990, 34 Criminology 163, 168 (1996); Edward E. Leamer, Let's Take the Con Out of
Econometrics, 73 Am. Econ. Rev. 31, 41-42 (1983).
n217. For technical explanations and examples of EBA, see Richard Fowles, Micro EBA: Extreme Bounds

Page 54
50 Stan. L. Rev. 1055, *1145

Analysis on Microcomputers, 42 Am. Statistician 274 (1985); Edward E. Leamer, Sets of Posterior Means with
Bounded Variance Priors, 50 Econometrica 725 (1982).
n218. For another example of this approach, see Fowles & Merva, supra note 216, at 168.
n219. We included in our "doubtful" set all variables for which we had data over the relevant time period.
n220. Here we need to make a technical argument. While the extreme bounds on burglary do straddle zero
(ranging from +0.999 to -4.413), the bounds that are "favored" by the data do not. In Bayesian estimation, the
posterior mean is pulled between the prior location (zero) and the maximum likelihood estimate (the
data-favored location). Dogmatic priors are characterized by high prior precision that the prior location is
correct. For example, by dropping a variable from a regression equation, the researcher is expressing an opinion
that the associated parameter is zero with probability one. Diffuse priors place zero precision on the prior mean
so the data are completely favored. With completely diffuse priors, the posterior mean and the maximum
likelihood estimate are the same. Posterior bounds for burglary associated with low prior precision range from
-0.875 to -3.752, a nonfragile result.
n221. See Fowles & Merva, supra note 216, at 167-68.
n222. We caution against drawing firm conclusions from the aggregate category of "violent" crimes for the
reasons noted at note 146 supra and accompanying text.
n223. See Table VI.
n224. 24.2% + (1.7-7.2%)/24.2%.
n225. See note 347 infra and accompanying text (noting the Supreme Court's interest in absolute numbers of
crimes affected by the exclusionary rule).
n226. This number is derived by multiplying the range of the Miranda effect by the number of offenses
known in 1995. We have used offenses known to all law enforcement agencies rather than offenses known to
city law enforcement agencies to provide a nationwide estimate. See note 396 infra and accompanying text
(noting the general parallel between nationwide and city data).
n227. See notes 47-49 supra and accompanying text.
n228. See Edward R. Tufte, Data Analysis for Politics and Policy 139 (1974) ("Whether or not there really
is a causal relationship ... depends on having a theory, consistent with the data, that links the variables.").
n229. See Britt et al., supra note 94, at 362-63 (expounding this point); cf. McDowall et al., supra note 80, at
382-84 (explaining that controls are useful but not required for time series inferences).

Page 55
50 Stan. L. Rev. 1055, *1145

n230. One of the present authors (Cassell) has presented a preliminary version of this paper to and sought
comments from audiences at the ABA's Annual National Conference, see Miranda Decision's Legitimacy,
Effects on Law Enforcement Debated at ABA Meeting, 59 Crim. L. Rep. (BNA) 1465, 1465-66 (1996), at
Chicago-Kent, Loyola, Northwestern, San Diego, Stanford, UCLA, and Utah law schools, at the Goldwater
Institute, the Heritage Foundation, and elsewhere. In addition, he has sought suggestions in widely available
fora. See Cassell, supra note 20, at 1091 n.33 (seeking suggestions as to "other variables that should be
included" in the regression analysis); Paul Cassell, Harmful Effects of Miranda on Law Enforcement (June 19,
1996) <crimprof@chicagokent.kentlaw.edu> (reciting regressions and asking, "Am I overlooking anything?").
We remain open, of course, to receiving other suggestions.
n231. See Ross et al., supra note 92, at 504 ("When a set of hypotheses cannot be ruled out mechanically
through design, the researcher bears the burden of seeking out the reasonable hypotheses included therein and
ruling them out or allowing for them individually.").
n232. UCR-1967, supra note 36, at 30.
n233. Cf. Sullivan, supra note 90, at 161 (finding that travel time to work was negatively associated with
clearance rates for auto theft).
n234. See 2 Bureau of the Census, U.S. Dep't of Commerce, Historical Statistics of the United States:
Colonial Times to 1970, at 717 ser.Q:175-86 (1975) (reporting that the percentage of families owning
automobiles increased slowly from 1960-1970).
n235. See Otis H. Stephens, Jr., Robert L. Flanders & J. Lewis Cannon, Law Enforcement and the Supreme
Court: Police Perceptions of the Miranda Requirements, 39 Tenn. L. Rev. 407 (1972) [hereinafter Stephens et
al., Police Perceptions]; see also Stephens, supra note 46, at 179-99 (recounting the survey results first published
in Police Perceptions, supra).
n236. See Stephens et al., Police Perceptions, supra note 235, at 420 tbl.IV (finding 92% believed the Court
decisions created negative effects, 58% attributing this first and foremost to Miranda). The percentage fingering
Miranda as responsible might have been higher but for a legal memorandum on legal restrictions governing
search and seizure that circulated shortly before some of the interviews. See id. at 421. Seventy-four percent said
that advice of defendants' rights had an adverse effect on investigations. See id. at 424 tbl.VIII. In individual
interviews, the officers surveyed generally gave negative assessments of Miranda. See id. at 426-29. In light of
these findings, Stephens' conclusion that his survey showed little impact from Miranda, see id. at 430-31, is hard
to understand.
n237. See Wald et al., supra note 41, at 1528.
n238. Id. at 1611.
n239. Id. at 1612 n.265.
n240. See Gary L. Wolfstone, Miranda - A Survey of Its Impact, 7 Prosecutor 26, 27 (1971).

Page 56
50 Stan. L. Rev. 1055, *1145

n241. Witt's article was published in 1973. See Witt, supra note 42, at 334. He appears to have begun
collecting his data sometime after 1968.
n242. Id. at 325.
n243. Id. at 330.
n244. See Cassell, supra note 20, at 1108-10 (discussing 1987 and 1988 studies on police views of
Miranda); see also Stephen J. Schulhofer, Confessions and the Court, 79 Mich. L. Rev. 865, 873 (1981) ("Law
enforcement authorities continue to press vigorously for the overruling of Miranda ...."). But see Peter Arenella,
Miranda Stories, 20 Harv. J.L. & Pub. Pol'y 375, 381 (1996) ("Police fighting in the trenches do not seem eager
to abandon Miranda[.]"); Schulhofer, supra note 16, at 507 (claiming that testimony of police officers that
Miranda did not have any long-term negative impact is "widespread and consistent" since the mid-1970s).
n245. Marco Leavitt, Law Professor Says It Is Time to Throw Off Shackles of Miranda Ruling, Ogden
Standard Examiner, June 12, 1996, at 1 (quoting Salt Lake City Police Detective Mike Welch).
n246. See Schulhofer, Bashing Miranda, supra note 46, at 369-70 (suggesting that other Warren Court
rulings should be taken into account).
n247. See, e.g., Kamisar, supra note 98, at 245-57 (denying that the Warren Court decisions were
responsible for sinking clearance rates in the 1960s). But see Reynolds, supra note 210, at 8 (suggesting that
"punishment philosophy softened so much" during the Warren Court years that crime control became more
difficult). See generally Robert Weisberg, Supreme Court Review Foreword: Criminal Procedure Doctrine:
Some Versions of the Skeptical, 76 J. Crim. L. & Criminology 832 (1985) (discussing skepticism about whether
Supreme Court decisions significantly affect the real world of criminal justice).
n248. See Ehrlich & Brower, supra note 52, at 103 tbl.2.
n249. 388 U.S. 218 (1967).
n250. See Stovall v. Denno, 388 U.S. 293, 296 (1967); Gilbert v. California, 388 U.S. 263, 272 (1967);
Wade, 388 U.S. at 235.
n251. See Kirby v. Illinois, 406 U.S. 682, 690-91 (1972) (refusing to extend counsel requirements to lineup
cases where "a person has not been formally charged with a criminal offense").
n252. Joshua Dressler, Understanding Criminal Procedure 166[B][2], at 352 (1991).
n253. See Figures 1-5 (finding no noticeable rise in clearance rates in 1973 and 1974).
n254. 389 U.S. 347 (1967).

Page 57
50 Stan. L. Rev. 1055, *1145

n255. Id. at 351.
n256. See id. at 354.
n257. 385 U.S. 293 (1966).
n258. See id. at 309-10.
n259. 387 U.S. 1 (1967).
n260. See id. at 58.
n261. 386 U.S. 18 (1967).
n262. Id. at 24.
n263. But cf. Kamisar, supra note 98, at 242 n.5 (concluding that the Warren Court decisions "caused
relatively little furor until applied to the "police practice' phases of the criminal process in the late 1960's").
n264. 367 U.S. 643, 655-56 (1961) (excluding evidence in a state criminal trial that was the result of an
illegal search).
n265. 372 U.S. 335, 343-45 (1963) (establishing a right to counsel for indigent defendants).
n266. See Schulhofer, supra note 16, at 512.
n267. Note that robbery clearance rates changed little from 1962-1965, even rising slightly from 1964-1965.
See Figure 4.
n268. See James D. Cameron & Richard Lustiger, The Exclusionary Rule: A Cost-Benefit Analysis, 101
F.R.D. 109, 128 (1984) (stating that the exclusionary rule "does not deter police misconduct"); Dallin H. Oaks,
Studying the Exclusionary Rule in Search and Seizure, 37 U. Chi. L. Rev. 665, 755 (1970) (arguing that the
"exclusionary rule is a failure" in terms of "deterring illegal searches and seizures by the police"); see also Paul
G. Cassell, The Mysterious Creation of Search and Seizure Exclusionary Rules Under State Constitutions: The
Utah Example, 1993 Utah L. Rev. 751, 842-46 (collecting evidence that state exclusionary rules are unlikely to
have an incremental deterrent effect beyond the federal rules). The Court's ruling in United States v. Leon, 468
U.S. 897 (1984) (establishing a good faith exception to the warrant requirement), appears to have "practical
day-to-day effects" that are "minimal." Craig D. Uchida, Timothy S. Bynum, Dennis Rogan & Donna M.
Murasky, The Effects of United States v. Leon on Police Search Warrant-Policies and Practices 39 (1986).
n269. See Stuart S. Nagel, Testing the Effects of Excluding Illegally Seized Evidence, 1965 Wis. L. Rev.

Page 58
50 Stan. L. Rev. 1055, *1145

283, 285-88 (finding changes in police search and seizure practices due to the exclusionary rule from
1960-1963); Atkins & Rubin, supra note 87, at 10 (finding Mapp had an effect on crime rates starting in 1961);
cf. Bradley C. Canon, Is the Exclusionary Rule in Failing Health? Some New Data and a Plea Against a
Precipitous Conclusion, 62 Ky. L.J. 681, 711-16 (1974) (arguing that the Supreme Court's refinement of the
exclusionary rule in its 1969 decision Chimel v. California, 395 U.S. 752, produced an increase in use of search
warrants from 1968-1973).
n270. See William J. Stuntz, Implicit Bargains, Government Power, and the Fourth Amendment, 44 Stan. L.
Rev. 553, 559 (1992) (stating that, prior to the late 1960s, "police could, and did, keep public areas "clean' by
stopping or arresting whomever they wished"); William J. Stuntz, Privacy's Problem and the Law of Criminal
Procedure, 93 Mich. L. Rev. 1016, 1076 (1995) (noting that pre-1960s "vagrancy and loitering laws applied to
almost any public behavior").
n271. See Robert C. Ellickson, Controlling Chronic Misconduct in City Spaces: Of Panhandlers, Skid
Rows, and Public-Space Zoning, 105 Yale L.J. 1165, 1209-14 (1996) (detailing court decisions in the 1960s and
1970s that struck down regulations permitting arrest for minor forms of public disobedience).
n272. 405 U.S. 156, 171 (1972).
n273. Arrest rates per 100,000 inhabitants from 1960-1975 were:
[SEE TABLE IN ORIGINAL]
See UCR-1960, supra note 36, at 90 tbl.15 (tables for arrest rates by population groups); UCR-1965, at 108
tbl.18 (same); UCR-1966, at 110 tbl.22 (same); UCR-1967, at 116 tbl.23 (same); UCR-1968, at 110 tbl.22
(same); UCR-1969, at 108 tbl.23 (same); UCR-1970, at 120 tbl.23 (same); UCR-1975, at 180 tbl.29 (same).
n274. See Steven B. Duke & Albert C. Gross, America's Longest War: Rethinking Our Tragic Crusade
Against Drugs 100-02 (1993) (identifying 1964-1978 as "the Age of Aquarius," when drug use expanded within
several cultural subgroups, and 1978 to the present as "the Age of Narco Glitz," when commerce in cocaine and
crack has posed a threat to the social order).
n275. See Bureau of Justice Statistics, U.S. Dep't of Justice, Sourcebook of Criminal Justice Statistics 1994, at 413 tbl.4.33 (1995) [hereinafter Sourcebook] (reprinting an FBI compilation of 1993 UCR data).
n276. See, e.g., John Kaplan, The Hardest Drug: Heroin and Public Policy 57 (1983) (stating that experts
cannot "relate the contribution of heroin addiction to the total volume of property crime in the United States except to say that is far less than the popular literature has portrayed").
n277. See W.S. Wilson Huang & Michael S. Vaughn, Support and Confidence: Public Attitudes Toward the
Police, in Americans View Crime and Justice: A National Public Opinion Survey 31, 31 (Timothy J. Flanagan &
Dennis R. Longmire eds., 1996) (noting the need for citizen cooperation if a police force is to be effective).
n278. See 3 George H. Gallup, Gallup Poll: Public Opinion 1935-1971, at 1935, 2077 (1972).

Page 59
50 Stan. L. Rev. 1055, *1145

n279. See National Ctr. on Police and Community Relations, Michigan St. Univ., A National Survey of
Police and Community Relations 10 (1967) (reprinting a national Harris Poll finding a "good-excellent" rating of
76% for federal law enforcement, 70% for state law enforcement, and 65% for local law enforcement around
1967); note 278 supra and accompanying text.Polling data also suggest that racial minorities had lower
confidence in the police. See A National Survey of Police and Community Relations, supra, at 11-13. However,
this low confidence existed well before 1966. See id. at 16 (citing California Advisory Comm. to the U.S.
Commission on Civil Rights, Report on California: Police-Minority Group Relations (1963), which described
poor police relations with African Americans).
n280. See Sourcebook, supra note 275, at 166 tbl.2.31 (reporting that 49% of Harris Poll respondents felt
more uneasy on the streets in 1966 than in the previous year.This number increased to 53% in 1968 and to 55%
in 1969).
n281. See notes 65-69 supra and accompanying text.
n282. See Dennis R. Longmire, America's Attitudes About the Ultimate Weapon, in Americans View Crime
and Justice, supra note 277, at 93, 99 (reporting polls showing that public support for the death penalty has
generally risen since 1966); cf. Sourcebook, supra note 275, at 140 tbl.2.1 (reporting polls showing that public
concern about crime rose dramatically in the mid-1990s).
n283. See Leslie McAneny, The Gallup Poll: Pharmacists Again Most Trusted; Police, Federal Lawmakers
Images Improve (visited Aug. 26, 1997) <http://www.gallup.com/poll/news/9713.html>.
n284. See Caroline Wolf Harlow, Bureau of Justice Statistics, Special Report: Motor Vehicle Theft 4 tbl.10
(1988) (noting 87% of completed vehicle thefts reported to the police and 36% of attempted thefts, for 68%
overall); Cloninger & Sartorius, supra note 109, at 392 (reporting that auto theft showed no appreciable
reporting errors in national surveys) (citing Morris Silver, Punishment, Deterrence, and Police Effectiveness: A
Survey and Critical Interpretation of the Recent Econometric Literature (unpublished report for the Crime
Deterrence and Offender Career Project, New York, 1974)); Scott H. Decker, Alternate Measures of Police
Output, 1 Am. J. Police 23, 27, 30 (1981) (finding little difference between victimizations and reported vehicle
thefts); Samuel Nunn, Computers in the Cop Car: Impact of the Mobile Digital Terminal Technology on Motor
Vehicle Theft Clearance and Recovery Rates in a Texas City, 17 Evaluation Rev. 182, 187 (1993) ("Unlike
burglary, robbery, and assault, for example, that are subject to both the uncertainty of reporting and nonreporting
by victims and changes in the definition of the crimes, motor vehicle thefts are generally reported and are not
particularly subject to changes in definition.").
n285. See Schulhofer, Bashing Miranda, supra note 46, at 369.
n286. See, e.g., Anthony V. Bouza, The Police Mystique: An Insider's Look at Cops, Crime, and the
Criminal Justice System 134 (1990) ("Citizens report crime more faithfully when confidence in the police is
highest ....").
n287. See, e.g., Alfred Blumstein, Jacqueline Cohen & Richard Rosenfeld, Trend and Deviation in Crime
Rates: A Comparison of UCR and NCS Data for Burglary and Robbery, 29 Criminology 237, 246-48 (1991)
(finding, from 1973-1985, strong consistency between reported crimes, as measured by the UCR, and

Page 60
50 Stan. L. Rev. 1055, *1145

victimizations from crimes, as reported in the National Crime Victimization Survey).
n288. See note 94 supra. The incentive to report auto theft likely stems from insurance requirements.
n289. See Ross et al., supra note 292, at 494 (noting that in interrupted time series analysis, changes in the
means of measuring a variable can pose difficulties in interpreting before-and-after conditions of that variable);
cf. Campbell, supra note 292, at 415 (noting the larceny "crime wave" created in Chicago, when a
reform-minded administrator improved crime bookkeeping); Walker, supra note 50, at 295 (noting that the 1980
change in definition of "clear-up" rates in England means that "research conducted on data recorded before 1980
may require a different interpretation from later research").
n290. The 1962 UCR Handbook defines a "cleared" crime and provides examples. See Federal Bureau of
Investigation, U.S. Dep't of Justice, Uniform Crime Reporting Handbook: How to Prepare Uniform Crime
Reports 48-49 (1962) (defining "offenses cleared by arrest" and "exceptional clearances"). The 1965 and 1966
handbooks contain virtually identical definitions. See, e.g., Federal Bureau of Investigation, U.S. Dep't of Justice
Uniform Crime Reporting Handbook: How to Prepare Uniform Crime Reports 50 (1966) [hereinafter 1966 UCR
Handbook]. Later versions appear to be substantively indistinguishable. See, e.g., Federal Bureau of
Investigation, U.S. Dep't of Justice, Uniform Crime Reporting Handbook 41-42 (1984). It appears that the
definition of a "cleared" crime was slightly broadened some time before the 1960s by allowing police to count as
cleared crimes cases in which the offender was "turned over to the court for prosecution" rather than actually
requiring the prosecution to take place. See Sherman, supra note 99, at 380.
n291. Schulhofer, Bashing Miranda, supra note 46, at 369.
n292. See generally Bouza, supra note 286; Skolnick, supra note 98; Jerome H. Skolnick & James J. Fyfe,
Above the Law: Police and the Excessive Use of Force (1993).
n293. See Schulhofer, Bashing Miranda, supra note 46, at 368 (citing President's Comm'n on Law
Enforcement and Admin. of Justice, The Challenge of Crime in a Free Society 106-15 (1967)).
n294. See Robert Fogelson, Big-City Police 219-42 (1977) (discussing the wave of police reforms from the
1930s to the 1970s); Richard A. Leo, From Coercion to Deception: The Changing Nature of Police Interrogation
in America, 18 Crime L. & Soc. Change 35, 49-52 (1992) (discussing the growth and gradual entrenchment of
police professionalism from the 1930s to the mid-1960s).
n295. See notes 43 & 55 supra and accompanying text (noting Schulhofer's claim to this effect).
n296. See notes 63-76 supra and accompanying text.
n297. Schulhofer, Clearance Rates, supra note 46, at 280 (emphasis added); see id. ("Soaring rates of violent
crime and stagnant levels of police resources easily explain the observed clearance-rate trends.") (emphasis
added); Stephen J. Schulhofer, Pointing in the Wrong Direction, Legal Times, Aug. 12, 1996, at 24 ("There is no
reason - none - to blame Miranda, rather than precipitously shrinking resources, for the decline in clearance rates
during the late 1960s.").

Page 61
50 Stan. L. Rev. 1055, *1145

n298. See Cassell, supra note 83, at 307, 308 fig.2 ( "Crime rates were increasing well before Miranda was
handed down and continued to climb afterwards.").
n299. Schulhofer, Bashing Miranda, supra note 46, at 357-58.
n300. See id. at 365-67; Schulhofer, Clearance Rates, supra note 46, at 291-92.
n301. See, e.g., Tables II-IV (finding a statistically significant Miranda effect on violent and property
crimes, but no statistically significant effect from crime rate, police variables, capacity variables, or crime in
small cities - urbanization was significant only for violent crimes).
n302. See notes 54-76 supra and accompanying text.
n303. See text accompanying notes 77-227 supra.
n304. See id.
n305. See text accompanying notes 232-245 supra.
n306. See notes 20-24 supra and accompanying text.
n307. See notes 29-33 supra and accompanying text.
n308. See Cassell & Hayman, supra note 29, at 876-78 (noting confession rates of over 60% in Great
Britain and Canada); cf. Richard S. Frase, Comparative Criminal Justices As a Guide to American Law Reform:
How Do the French Do It, How Can We Find Out, and Why Should We Care?, 78 Cal. L. Rev. 542, 590-94
(1990) (reporting French clearance rates higher than American rates in three of four categories but concluding
that differences are relatively modest).
n309. See Miranda v. Arizona, 384 U.S. 436, 516 (Harlan, J., dissenting) ("There can be little doubt that the
Court's new code would markedly decrease the number of confessions."); id. at 541 (White, J., dissenting)
(discussing clearance rates and concluding that "the rule announced today will measurably weaken the ability of
the criminal law to perform these tasks").
n310. See id. at 481 (claiming that the decision "should not constitute an undue interference with a proper
system of law enforcement" (emphasis added)).
n311. See, e.g., Transcript: The Goldwater Institute and the Federalist Society: Federalism and Judicial
Mandates, 28 Ariz. St. L.J. 17, 160 (1996) (remarks of Professor David Rudovsky) [hereinafter Rudovsky]
(raising this possibility).

Page 62
50 Stan. L. Rev. 1055, *1145

n312. To be clear, we believe that even isolated instances of coerced confessions should be strongly
condemned.
n313. Cassell, supra note 13, at 473-75 (describing police reform efforts in the twentieth century); Leo,
supra note 294, at 47-53 (same).
n314. Leo, supra note 294, at 51.
n315. Fred P. Graham, The Self-Inflicted Wound 22 (1969); see also Fred E. Inbau & James P. Manak,
Miranda v. Arizona - Is It Worth the Cost?, Prosecutor, Spring 1988, at 31, 36 ("Prior to Miranda, the movement
was under way to improve the quality of police interrogation practices.").
n316. Miranda v. Arizona, 384 U.S. 436, 447-48; see also id. at 499-500 (Clark, J., dissenting). However,
the majority continued on to state that police abuses "are [still] sufficiently widespread to be the object of
concern." Id. at 447-48.
n317. For a discussion of this empirical evidence, see Cassell, supra note 13, at 474-75.
n318. Gerald N. Rosenberg, The Hollow Hope: Can Courts Bring About Social Change? 326 (1991).
n319. Professor Wayne R. LaFave reported the year before the Miranda decision that "in the great majority
of in-custody interrogations observed, the possibility of coercion appeared slight." Wayne R. LaFave, Arrest:
The Decision to Take a Suspect into Custody 386 (1965); see also Edward L. Barrett, Jr., Police Practices and
the Law - From Arrest to Release or Charge, 50 Cal. L. Rev. 11, 42 (1962) (reporting California data in 1960
that most interrogations lasted under two hours). Similarly, the student observers in New Haven in 1966,
assessing all forms of police "tactics," found "a low level of coerciveness in most questioning." Wald et al.,
supra note 41 , at 1558.
n320. Such challenges are rarely made today and even more rarely granted. See Cassell, supra note 13, at
392-93 (collecting the studies on this point).
n321. See id. at 476 (citing studies showing few rejections of defendants' statements by prosecutors or by
judges at preliminary hearings).
n322. Miranda v. Arizona, 384 U.S. 436, 505 (Harlan, J., dissenting).
n323. See Evelle J. Younger, Prosecution Problems, 53 A.B.A. J. 695, 698 (1967) ("Miranda will not affect
the brutal or perjurious policeman - he will continue to extract confessions without reference to the intonations
of the Supreme Court; and when he testifies, he will simply conform his perjury to the latest ground rules."); cf.
Leiken, supra note 172, at 22 (finding defendants claimed, after Miranda, that police had made promises or
threats to get confessions). Shortly after Miranda, Neal Milner studied the police response in Wisconsin and
found that "generally most interrogations continued to operate under rules formalized prior to the Miranda
decision." Milner, supra note 59, at 228; cf. George C. Thomas III & Marshall D. Bilder, Aristotle's Paradox and

Page 63
50 Stan. L. Rev. 1055, *1145

the Self-Incrimination Puzzle, 82 J. Crim. L. & Criminology 243, 278 (1991) (noting the paradoxical quality of
the Miranda debate in which the decision's defenders frequently argue it has had little effect.)
n324. Note that we are only arguing that the clearance rate decline found from 1966-1968 is not explained
by a sudden, contemporaneous reduction in coerciveness. "One could believe that police interrogation has
generally become less coercive over the last several decades and still accept this claim." Cassell, supra note 13,
at 478 n.533.
n325. Joe Costanzo, Has Miranda Handcuffed the Cops?, Deseret News, June 12-13, 1996, at A1 (quoting
Kamisar). Kamisar has discussed the issue of the reliability of clearance rates at greater length in Kamisar, supra
note 98, at 252-53 (citing Skolnick, supra note 98, at 178-79, and suggesting the unreliability of clearance rate
statistics); Yale Kamisar, On the Tactics of Police-Prosecution Oriented Critics of the Courts, 49 Cornell L.Q.
436, 466 n.173 (1964) (discussing clearances through confessions to multiple crimes).
n326. See, e.g., Skolnick, supra note 98, at 178-79 (describing a burglar giving multiple clearances to
police); Wald et al., supra note 41, at 1593 n.197, 1595-96 (finding that solving other crimes is the most
important reason for questioning other than gathering evidence for trial and giving examples of questioning to
clear crimes); Rudovsky, supra note 311, at 160 (noting that the impact on the defendant from confessing to
additional crimes is minimal, while it significantly benefits the police because they are able to "clear the books"
of many unsolved cases); Weinraub, supra note 60 (describing police use of questioning to clear other crimes);
cf. Pauline Morris, Royal Comm'n on Criminal Procedure, Police Interrogation: Review of Literature 11 (1980)
(stating that, in Britain, interrogation may lead to the clearance of crimes other than the one being investigated);
Bottomley & Coleman, supra note 99, at 50-52 (attributing part of drop in British clearance rates to the
introduction of new confession procedures reducing confessions to multiple crimes).
n327. Cf. Greenwood, supra note 100, at 24 (discussing differences between clearance rates and arrest
rates).
n328. As an illustration, assume that the violent crime clearance rate before Miranda was 60% because, of
every 100 crimes, the police arrested 40 suspects and cleared 1.5 crimes per arrest. After Miranda, if police
arrested 40 suspects and cleared only one violent crime per arrest, then the clearance rate would fall to 40%.
n329. The limited quantitative empirical evidence on whether such multiple confessions were sufficiently
prevalent to make Kamisar's speculation even theoretically possible is mixed. Compare Taping Police
Interrogations, supra note 38 (finding that, after Miranda, only 12 unsolved crimes were solved through 1460
interrogations), with Wald et al., supra note 41, at 1595 n.203 (finding, after Miranda, about 10-15% of all
unsolved crimes during observation period were solved through interrogations). See generally Weinraub, supra
note 60 (reporting that a police spokesperson attributed the 10% decline in clearance rates from 1965-1966 in
New York City in part to Supreme Court decisions restricting the ability to get multiple confessions).
n330. The FBI has published a consistent revision of its arrest data that extends back only to 1965. The
revised arrest data are used in preference to unadjusted data for reasons discussed in the Appendix. See notes
398-400 infra and accompanying text. Violent crime data is used because Professor Kamisar was discussing a
presentation of our thesis that used violent crimes and because an aggregate grouping tended to smooth some
puzzling fluctuations in the year-to-year data. See note 331 infra.

Page 64
50 Stan. L. Rev. 1055, *1145

n331. Figure 6 shows large and apparently unpredictable fluctuations in both clearances per arrest and
arrests per crime, fluctuations that were even larger in some individual crime categories. This suggests that the
data on these variables may not be consistent over time. Also, from about 1977 onward, arrests per crime
modestly decline while clearances per arrest modestly rise. Because the two figures are reciprocally related, see
notes 437-438 infra and accompanying text, these movements may be, in part, a statistical artifact rather than a
reflection of some real-world trend. However, declining arrest rates from 1974-1986 for various crimes were
also found in Patrick A. Langan, America's Soaring Prison Population, 251 Science 1568, 1572 tbl.4 (1991).
n332. See Richard A. Leo, Police Interrogation and Social Control, 3 Soc. & Legal Stud. 93, 99 (1994)
(reporting police views that "getting a confession makes the investigator's job a lot easier ... If he gets a
confession ... he doesn't have to spend hours tracking down witnesses, running fingerprints, putting together
line-ups, etc.").
n333. This conclusion assumes that police might be able to solve other crimes through additional
investigation.But cf. Table II (showing police resources generally had little effect on clearances); Wald et al.,
supra note 41, at 1596 (questioning whether investigation of uncleared crimes is effective).
n334. See Greenberg et al., supra note 101, at 45-46 (reporting detectives' views to this effect).
n335. See Leslie Sebba, Third Parties: Victims and the Criminal Justice System 80-81 (1996) (reaching this
conclusion after a review of several studies).
n336. See President's Task Force on Victims of Crime, Final Report 4 (1982) (describing fear of
"composite" victim of crime); see also id. at 28 (noting that victims suffer when police fail to solve crimes).
n337. See id. at 5.
n338. See Greenberg et al., supra note 101, at 45 (reporting detectives' views that the possibility of returning
property to burglary victims justified attempts to obtain multiple clearances).
n339. Cf. Morris v. Slappy, 461 U.S. 1, 14 (1983) ("In the administration of criminal justice, courts may
not ignore the concerns of victims.").
n340. See Cassell & Hayman, supra note 29, at 909-16 (finding that whether police obtain an incriminating
statement makes a significant difference in case outcome).
n341. See Cassell, supra note 13, at 494-96 (presenting data on what percentage of suspects do not waive
rights and thus cannot be interrogated).
n342. See Baker, supra note 13, at 200-01 (collecting examples of such responses to Miranda).
n343. See Table VIII.

Page 65
50 Stan. L. Rev. 1055, *1145

n344. See id.
n345. See generally Akhil Reed Amar, The Constitution and Criminal Procedure: First Principles 20-31
(1997) (discussing the Fourth Amendment exclusionary rule).
n346. See United States v. Leon, 468 U.S. 897, 907 n.6 (1984) (citing Thomas Y. Davies, A Hard Look at
What We Know (and Still Need to Learn) About the "Costs" of the Exclusionary Rule: The NIJ Study and Other
Studies of "Lost" Arrests, 1983 Am. B. Found. Res. J. 611, 621, 667 tbl.6).
n347. Id. at 908 n.6.
n348. See note 47 supra and accompanying text.
n349. Compare note 343 supra and accompanying text with note 346 supra .
n350. See notes 151-153 supra and accompanying text.
n351. Cf. Davies, supra note 346, at 656-68 (comparing estimates of the effects of illegal searches to
exclusionary rule studies).
n352. See Dallin H. Oaks, Studying the Exclusionary Rule in Search and Seizure, 37 U. Chi. L. Rev. 665,
681 (1970) (noting that suppression motions are most frequently filed in cases involving narcotics and weapons).
n353. See generally Paul G. Cassell, Balancing the Scales of Justice: The Case for and the Effects of Utah's
Victims' Rights Amendment, 1994 Utah L. Rev. 1373 (considering the crime victim's perspective); Douglas E.
Beloof, Victims in Criminal Procedure (forthcoming 1998) (discussing victims and the exclusionary rule).
n354. For an example, see State v. Oldham, 618 S.W.2d 647 (Mo. 1981). See also 136 Cong. Rec. S9027
(daily ed. June 28, 1990) (statement of Sen. Hatch describing this case).
n355. Gerald M. Caplan, Miranda Revisited, 93 Yale L.J. 1375, 1384-85 (1984) (reviewing Baker, supra
note 13).
n356. Cf. Charles Murray, Losing Ground: American Social Policy, 1950-1980, at 117 (1984) (reviewing
crime statistics and concluding: "Put simply, it was much more dangerous to be black in 1972 than it was in
1965, whereas it was not much more dangerous to be white.").
n357. Cf. Supplemental Brief for the United States as Amicus Curiae Supporting Reversal at 46, Illinois v.
Gates, 462 U.S. 213 (1983) (No. 81-430) ("The freeing of even one guilty defendant by virtue of an irrational
rule may exact a greater cost than society should be expected to bear ....").

Page 66
50 Stan. L. Rev. 1055, *1145

n358. See Grano, supra note 9, at 173-98 (discussing how the prophylactic nature of Miranda differs from
most constitutional law decisions).
n359. 417 U.S. 433 (1974).
n360. Id. at 443-44.
n361. Id. at 446; see also Withrow v. Williams, 507 U.S. 680, 690-91 (1993) (collecting numerous cases
describing Miranda rights as ""prophylactic' in nature").
n362. See Grano, supra note 9, at 173-98; Office of Legal Policy, U.S. Dep't of Justice, Report to the
Attorney General on the Law of Pre-Trial Interrogation 3-41 (1986) [hereinafter OLP Pre-Trial Interrogation
Report].
n363. Schulhofer, supra note 16, at 552 n.214.
n364. Moran v. Burbine, 475 U.S. 412, 433 n.4. (1986).
n365. See Cassell & Hayman, supra note 29, at 921-22; cf. Davies, supra note 29, at 626 (noting that "there
is no empirical content to this "balancing' approach to considering the costs and benefits" of the search and
seizure exclusionary rule).
n366. See note 11 supra and accompanying text.
n367. See 18 U.S.C. 3501 (1994) (a federal statute rejecting Miranda in favor of voluntariness test); Grano,
supra note 9, at 199-222 (advocating overruling of Miranda on purely doctrinal grounds).
n368. See Miranda v. Arizona, 384 U.S. 436, 523 (Harlan, J., dissenting) (noting the "ironic untimeliness"
of the Court's new confession rules in view of ongoing "massive re-examination of criminal law enforcement
procedures on a scale never before witnessed"); OLP Pre-Trial Interrogation Report, supra note 362, at 40-41
(discussing American Law Institute efforts to draft a Model Code of Pre-arraignment Procedure).
n369. Miranda, 384 U.S. at 467.
n370. OLP Pre-Trial Interrogation Report, supra note 362, at 96.
n371. See Judge Harold J. Rothwax, Guilty: The Collapse of Criminal Justice 237 (1996) (urging
abandonment of Miranda and replacement with videotaping); Cassell, supra note 13, at 486-92 (advocating
videotaping and modified warnings as a replacement for Miranda); Paul G. Cassell, Protecting the Innocent from
False Confessions and Lost Confessions - and from Miranda, 88 J. Crim. L. & Criminology 497 (1998)
(suggesting that videotaping as a replacement for Miranda is a superior solution to the problem of false
confessions and lost confessions); Paul G. Cassell, Balanced Approaches to the False Confessions Problem: A

Page 67
50 Stan. L. Rev. 1055, *1145

Brief Comment on Ofshe, Leo, and Alschuler," 74 Denv. U. L. Rev. 1123 (1997) (raising videotaping as a
replacement for Miranda); Richard J. Ofshe & Richard A. Leo, The Social Psychology of Police Interrogation:
The Theory and Classification of True and False Confessions, 16 Stud. Law, Pol. & Soc. 189, 238 (Austin Sarat
& Susan S. Sibley eds., 1997) (suggesting a "taping" of interrogatories as one of the safeguards against false
confessions).
n372. See Cassell, supra note 13, at 489-92.
n373. Miranda, 384 U.S. at 467.
n374. See Paul G. Kauper, Judicial Examination of the Accused - A Remedy for the Third Degree, 30 Mich.
L. Rev. 1224 (1932).
n375. See Walter V. Schaefer, The Suspect and Society: Criminal Procedure and Converging Constitutional
Doctrines 76-81 (1967) (arguing for an interpretation of the Fifth Amendment that would permit magistrates to
question suspects and prosecutors to use resulting nontestimonial fruits at trial); Henry J. Friendly, The Fifth
Amendment Tomorrow: The Case for Constitutional Change, 37 U. Cin. L. Rev. 671, 721-25 (1968) (similar);
Akhil Reed Amar, The Constitution and Criminal Procedure 76-77 (1997) (similar); Akhil Reed Amar & Renee
B. Lettow, Fifth Amendment First Principles: The Self-Incrimination Clause, 93 Mich. L. Rev. 857, 898-928
(1995) (similar).
n376. See Yale Kamisar, Kauper's "Judicial Examination of the Accused" Forty Years Later - Some
Comments on a Remarkable Article, 73 Mich. L. Rev. 15, 33-35 n.70 (1974) (discussing constitutional problems
with Kauper's proposal in light of contemporary doctrine).
n377. See, e.g., Ronald J. Allen, Richard B. Kuhns & William J. Stuntz, Constitutional Criminal Procedure:
An Examination of the Fourth, Fifth, and Sixth Amendments and Related Areas 1229 (3d ed. 1995) (raising the
Kauper proposal as a possible alternative to Miranda).
n378. One way to accomplish this result is to "disincorporate" the Miranda doctrine and allow the states to
determine whether to follow it. See Barry Latzer, Toward the Decentralization of Criminal Procedure: State
Constitutional Law and Selective Disincorporation, 87 J. Crim. L. & Criminology 63, 101-11 (1996) (suggesting
this approach).
n379. Cf. Paul G. Cassell, The Rodney King Trials and the Double Jeopardy Clause: Some Observations on
Original Meaning and the ACLU's Schizophrenic Views of the Dual Sovereign Doctrine, 41 UCLA L. Rev. 693,
717-19 (1994) (discussing role of original meaning in constitutional criminal procedure).
n380. See 18 U.S.C. 3501 (1994); see also Davis v. United States, 512 U.S. 452, 465 (1994) (Scalia, J.,
concurring) (arguing that section 3501 "reflects the people's assessment of the proper balance to be struck
between concern for persons interrogated in custody and the needs of effective law enforcement"); United States
v. Crocker, 510 F.2d 1129 (10th Cir. 1975) (concluding that section 3501 is constitutional); United States v.
Rivas-Lopez, 988 F. Supp. 1424 (D. Utah 1997) (same); Eric D. Miller, Comment, Should Courts Consider 18
U.S.C. 3501 Sua Sponte?, 665 U. Chi. L. Rev. (forthcoming 1998) (concluding that federal courts must follow

Page 68
50 Stan. L. Rev. 1055, *1145

section 3501).
n381. Miranda v. Arizona, 384 U.S. 436, 504 (Harlan, J., dissenting).
n382. See notes 302-310 supra and accompanying text.
n383. Pyle & Deadman, supra note 292, at 207.
n384. See UCR-1958, supra note 36, at 76 tbl.12; Brewer Letter, supra note 53, at 1.
n385. See UCR-1973, supra note 36, at 1; Brewer Letter, supra note 53, at 1.
n386. Fox appears to ignore this discontinuity. See Fox, supra note 36, at 6.
n387. Id. at 86 (based on UCR data).
n388. Compare id. with UCR-1974, supra note 36, at 166.
n389. See Fox, supra note 36, at 7.
n390. See UCR-1958, supra note 36, at 2 (limiting rape figures to those rapes involving force).
n391. See notes 198-201 supra and accompanying text.
n392. See, e.g., UCR-1995, supra note 36, at 181 tbl.12 (showing that 80.1% of offenses known to police
were reported by cities (percentage derived from id.)).
n393. Cf. Fox, supra note 90, at 123 n.3 (concluding that since city data is weighed in proportion to
population in the UCR data, inclusion of small cities should contribute little to the analytic results).
n394. See UCR-1995, supra note 36, at 199 tbl.25 (reporting clearance rate data for cities of 43.5% for
violent crimes and 17.7% for property crimes and data for all agencies of 45.4% for violent crimes and 17.6%
for property crimes).
n395. See id. at 187 tbl.16 (reporting crime rate data, per 100,000 population, for cities of 919.9 for violent
crimes and 5793.0 for property crimes and data for all agencies of 746.5 for violent crimes and 4877.8 for
property crimes).
n396. It is possible that use of city data in our regression equations may have slightly increased the
magnitude of the Miranda effect that we detected. The available empirical evidence suggests that Miranda

Page 69
50 Stan. L. Rev. 1055, *1145

depressed confession rates more in bigger cities. See Cassell, supra note 13, at 447-50 (analyzing the evidence of
post-Miranda reduction of such rates).
n397. See notes 191-194 supra and accompanying text.
n398. The modification is published in UCR-1975, supra note 36, at 49 tbl.2, and in some, but not all,
subsequent UCRs.
n399. See, e.g., Jacob & Rich, supra note 210, at 122 (using UCR city data on robberies for nine cities
without any adjustment).
n400. See Cassell, supra note 46, at 334-38 (discussing Schulhofer's reliance upon unrevised crime rate data
to construct "clearance capacity" charts, which do not track clearance rate changes).
n401. See Brower, supra note 96, at 172 tbl.47, 175 tbl.48, 178 tbl.49, 181 tbl.50, 184 tbl.51, 187 tbl.52 &
190 tbl.53.
n402. See note 403 infra (providing ratios of official revised figures to unofficial revised figures).
n403. The mean ratio of the unofficial revision to official revision for the seven index crimes was: 0.991 for
murder; 0.991 for rape; 0.995 for robbery; 0.989 for aggravated assault; 0.987 for burglary; and 0.933 for auto
theft. The ratios did not vary much during the overlap period. For instance, for assault the ratios ranged from a
low of 0.9887 to a high of 0.9890.
n404. Compare Brower, supra note 96, at 187 tbl.52 (reporting that unpublished revision of larceny for 1960
was 507,300), with UCR-1975, supra note 36, at 49 tbl.2 (reporting that official revision of larceny for 1960 was
1,855,400).
n405. The 1960 ratio was 507,300 ) 1,855,400 = 27.3%; the 1972 ratio was 1,837,800 ) 4,151,200 = 44.3%.
Compare Brower, supra note 96, at 187 tbl.52 (providing unpublished revised data), with UCR-1975, supra note
36, at 49 tbl.2 (providing official revised data).
n406. See UCR-1958, supra note 36, at 1-3 (noting a change in estimation techniques and larceny
definition).
n407. See UCR-1975, supra note 36, at 49 tbl.2 (reporting total larcenies for 1960 as 1,855,400).
n408. See UCR-1960, supra note 36, at 79 tbl.5 (reporting an 11.7% increase in total larcenies in 1960)
(percentage derived from id. by combining larceny under and over $ 50 and calculating the percent increase);
UCR-1959, supra note 36, at 75 tbl.5 (reporting a 0.8% decrease in total larcenies in 1959) (percentage derived
from id. by combining larceny under and over $ 50 and calculating the percent increase); UCR-1958, supra note
36, at 69 tbl.5 (reporting a 9.4% increase in total larcenies in 1958) (percentage derived from id. by combining
larceny under and over $ 50 and calculating the percent increase); UCR-1957, supra note 36, at 79 tbl.26

Page 70
50 Stan. L. Rev. 1055, *1145

(reporting an 8.0% increase in total larcenies in 1957). Data for cities rather than the entire country were used
because they were consistently reported for total larcenies, not just larcenies over $ 50, in urban data. Using
national estimates rather than city data would produce only a modest change in numbers used here. See
UCR-1960, supra note 36, at 35 tbl.2 (reporting a 14.2% increase in larcenies over $ 50 nationwide in 1960);
UCR-1959, supra note 36, at 4 (reporting a 3% increase in larcenies over $ 50 nationwide in 1959); UCR-1958,
supra note 36, at 1 (reporting a 10.3% increase in larcenies over $ 50 nationwide in 1958); UCR-1957, supra
note 36, at 71 tbl.24 (reporting an 8.4% increase in total larcenies nationwide in 1957).
n409. For the early 1950s, this information appears in a table labeled "Number of police department
employees." See, e.g., Semiannual Bulletin, supra note 36, at 21 tbl.11. For the years from 1958-1971, this
information appears in a table labeled "Full-Time Police Department Employees." See, e.g., UCR-1971, supra
note 36, at 158 tbl.51. For the years after 1971, the table is labeled "Full-Time Law Enforcement Employees."
See, e.g., UCR-1972, supra note 36, at 161 tbl.54.
n410. See, e.g., Jacob & Rich, supra note 210, at 122 app. (using the UCR's figures for number of sworn
officers).
n411. See Fox, supra note 36, at 6 (using the "number of full-time law enforcement employees per 1000
population").
n412. For instance, the reported number of police department employees went from 186,539 in 1959 to
195,109 in 1960 to 189,093 in 1961. However, the rate of police employees per thousand inhabitants during
those years remained relatively constant, moving from 1.8 in 1959 to 1.9 in 1960 to 1.9 in 1961. See UCR-1959,
supra note 36, at 105 tbl.23; UCR-1960, supra note 36, at 105 tbl.31; UCR-1961, supra note 36, at 108 tbl.33.
n413. See UCR-1980, supra note 36, at 264 tbl.68.
n414. In 1980, for instance, total agency population was 211 million; total city population was 143 million.
See id. at 262, 264.
n415. See, e.g., id. at 264 tbl.68; UCR-19985, supra note 36, at 248 tbl.74; UCR-1990, supra note 36, at 242
tbl.69.
n416. For 1950-1970, the data are available in 1 Bureau of the Census, supra note 234, at 416
ser.H:1012-1027. For later years, the data are available in Bureau of the Census, U.S. Dep't of Commerce,
Government Finances, Series GF, No. 5, which is in turn sometimes reprinted in the annual Bureau of the
Census, U.S. Dep't of Commerce, Statistical Abstract of the United States (hereinafter Statistical
Abstract-[year]). Where revisions of earlier data appeared, we used the revisions. Since 1962, the data are for
government fiscal years, which generally end June 30 (e.g., the data for 1994 would be for the fiscal year ending
in June 1994). This means that the data do not align precisely with the crime variables in our equations, which
rest on calendar year compilations.
n417. See, e.g., Fox, supra note 36, at 5, 127 n.5; Jacob & Rich, supra note 210, at 122 app.; Pogue, supra
note 108, at 44 app..

Page 71
50 Stan. L. Rev. 1055, *1145

n418. See U.S. Dep't of Commerce, Government Finances in 1962, G-GF62 No. 2, at 19 n.1 (1963).
n419. See Advisory Comm'n to Study the Consumer Price Index, Final Report to the Senate Finance
Comm.: Toward a More Accurate Measure of the Cost of Living (1996); cf. Scott H. Decker & Carol W.
Kohfeld, Fox Reexamined: A Research Note Examining the Perils of Quantification, 19 J. Res. Crime & Delinq.
110, 117-20 (1982) (debating whether the CPI is a useful predictor of crime rates); Fox, supra note 90, at 126
(debating same in response).
n420. See Fox, supra note 36, at 127 n.6 (finding the CPI to be a logical adjuster, but using an alternative
measure because the CPI was to be used as a predictor in that particular regression model).
n421. See Bureau of Labor Statistics, U.S. Dep't of Labor, Employment and Earnings (January issues,
published annually); see also Statistical Abstract, supra note 416.
n422. See, e.g., Statistical Abstract-1995, supra note 416, at 399 tbl.627.
n423. See generally Bureau of Econ. Analysis, U.S. Dep't of Commerce, Survey of Current Business; see
also Statistical Abstract, supra note 416.
n424. See e.g., Fowles & Merva, supra note 216 (discussing the correlation between wage inequality and
crime and reviewing relevant literature).
n425. See, e.g., id. at 77 tbl.94 (reprinting National Center for Health Statistics data).
n426. See Statistical Abstract-1975, supra note 416, at 19 tbl.20 (reporting that the urban population of the
U.S. was 64.0% in 1950, 69.9% in 1960, and 73.5% in 1970); Statistical Abstract-1995, supra note 416, at 43
tbl.44 (reporting that urban population of the U.S. was 73.7% in 1980 and 75.2% in 1990). The function was of
the form y = 0.0003x[su'3'] - 0.0719x[su'2'] + 5.81x - 84.449.
n427. See note 134 supra and accompanying text.
n428. Schulhofer, Bashing Miranda, supra note 46, at 367.
n429. We extended Schulhofer's series to 1995 by calculating data for that year. See UCR-1995, supra note
36, at 187 tbl.16 (providing relevant data in table entitled "Crime Rates, Offense Known to the Police,
Population Group").
n430. See Statistical Abstract, supra note 416.
n431. See UCR-1995, supra note 36, at 58 tbl.1, n.1 (indicating that the population data was supplied by the
Bureau of the Census).

Page 72
50 Stan. L. Rev. 1055, *1145

n432. See Statistical Abstract, supra note 416.
n433. In 1970, the ratio was 1.1 (40.3 million to 35.50 million). In 1980, the ratio was 1.09 (46.5 million to
42.5 million). See Statistical Abstract-1981, supra note 416, at 26 tbl.29 & 27 tbl.30. We assumed that the ratio
closed constantly from 1970-1980.
n434. See notes 218-221 supra and accompanying text.
n435. See John Neter, Michael H. Kutner, Christopher J. Nachtsheim & William Wasserman, Applied
Linear Regression Models 347-52 (1996) (discussing the merits of the procedure and explaining its mechanics).
n436. See id. at 348-52.
n437. See note 327 supra and accompanying text.
n438. A complicating factor is the possibility that police can claim a clearance "by exception" when some
element beyond police control precludes arrest and prosecution (e.g., a victim's failure to cooperate). See 1966
UCR Handbook, supra note 290, at 50. Such circumstances will be captured in the clearances per arrest
component; ideally they should be broken out separately. Unfortunately, no regularly published national data on
exceptional clearances are available. Data from Chicago for 1966 suggest that exceptional clearances are more
likely for crimes for which we failed to find a Miranda effect (e.g., rape, aggravated assault) than for crimes for
which we found a Miranda effect (e.g., robbery, burglary). See Albert J. Reiss, Jr., The Police and the Public 81
tbl.2.3 (1971). Therefore, we do not expect that the inclusion of exceptional clearances has had any influence on
our finding a Miranda effect.
n439. The arrest rate for violent crimes inexplicably fluctuates wildly from 1959-1963. The violent crime
arrest rate per 100,000 population is 92.2 in 1959, 112.8 in 1960, 117.8 in 1961, 112.8 in 1962, and 97.0 in
1963. See Fox, supra note 36, at 86 tbl.A-1, col.EV.
n440. See id.
n441. See UCR-1966, supra note 36, at 4.
n442. See Fox, supra note 36, at 127 n.9.
n443. See Uniform Crime Reporting Program, U.S. Dep't of Justice, Age-Specific Arrest Rates and
Race-Specific Arrest Rates for Selected Offenses 1965-92, at 173 (1993) [hereinafter Age-Specific Arrest Rates
and Race-Specific Arrest Rates]. The number of agencies reporting in each of the various years does vary, but
the extent of variation appears to be relatively small. See id. at 205-07.
n444. We used the FBI's revised series of crimes for this calculation. See notes 398-400 supra and
accompanying text (discussing the revised series).

Page 73
50 Stan. L. Rev. 1055, *1145

n445. See notes 392-396 supra and accompanying text (discussing the relative soundness of this
assumption).
n446. Compare, e.g., UCR-1966, supra note 36, at 100 tbl.12 (reporting 1966 clearance rate data based on
2857 cities with a population of 99.4 million), with Age-Specific Arrest Rates and Race-Specific Arrest Rates,
supra note 443, at 205 (reporting 1966 arrest rate based on 4048 cities with population of 132.4 million).