This article investigates whether the death penalty encourages defendants charged with potentially capital crimes to plead guilty in exchange for lesser sentences. I exploit a natural experiment in New York State: the 1995 reinstatement of capital punishment, coupled with the public refusal of some prosecutors to pursue death sentences (N.Y. Penal Law § 125.25 [McKinney 1975]). Using individual-level data on all felony arrests in the state between 1985 and 1998, I find the death penalty leads defendants to accept plea bargains with harsher terms, but does not increase defendants’ overall propensity to plead guilty. A differences-in-differences analysis of a national cross-section of homicide defendants confirms these results.

That the death penalty would encourage plea bargains is perhaps more plausible than that it would deter murders.

1. Among the many articles that examine the deterrent effect of capital punishment

are Ehrlich (1975), Ehrlich (1977), Cover and Thistle (1988), Avio (1988),

Chressanthis (1989), Grogger (1990), and Katz, Levitt, and Shustrovich (2003).

Cameron (1994) provides a good survey of the literature.

Does the Death Penalty Affect Plea Bargaining? 117

Second, plea bargaining raises constitutional issues. Plea bargains entail

that defendants—perhaps even innocent ones—relinquish their constitutional

right to a criminal trial. Civil libertarians may well view a positive

correlation between the death penalty and plea bargaining as evidence that

risk-averse defendants are willing to surrender an important individual

right when they are faced with the threat of execution.

Finally, determining the connection between sentence structure and

plea bargaining augments our understanding of the guilty plea—a fundamental

element of the criminal justice process. As over 90% of convictions

in criminal cases are obtained by guilty pleas (Grossman and Katz, 1983),

studying the incentives of prosecutors and defendants in pretrail negotiations

is essential to understanding the operation of the criminal justice

This article utilizes both time-series and cross-sectional data. I first

examine suspects arrested on felony charges between 1985 and 1998 in

New York, which changed its state law in 1995 to allow executions for

those convicted of first-degree murder. However, DAs in some counties

decided against pursuing the new sentencing option. The reinstatement of

the death penalty, coupled with the refusal of certain prosecutors to pursue

death sentences, provides a natural experiment to estimate the effect of the

death penalty on plea bargaining. Those defendants arrested for murder

after the statute was passed in counties where DAs pursue death sentences

serve as the treatment group in a differences-in-differences-in-differences

(DDD) estimate of the effect of the death penalty on plea bargaining.

I augment these results with analysis of a cross-section of murder cases

in 33 large, urban United States counties in 1988. The data set includes

defendants arrested for first-degree murder, as well as a variety of other

homicides. Since defendants charged with second-degree murder or lesser

charges do not face the death penalty, I use these defendants as a control

group in a differences-in-differences estimate.

Section 1 reviews the features of the 1995 New York statute relevant to

my empirical work. Section 2 describes the New York data set and the

empirical strategy. Section 3 discusses the results. Section 4 provides results

2. Only a few studies have examined the economics of plea bargaining. See

Rhodes (1976), Adelstein (1984), and Ostram (1988).

118 American Law and Economics Review V8 N1 2006 (116–142)

of several robustness checks. Section 5 analyzes the 1988 cross-sectional

data. Section 6 concludes.

1. A Brief Review of the New York Statute

In New York State the capital-case process begins with a first-degree

murder arraignment. Once an arrest is made, law enforcement officials

have 24 hours to arraign a suspect; otherwise, he is released. At the

arraignment the suspect is officially charged in court, bail is set, and a

trial date is discussed. The arraignment charge is usually similar to the

arrest charge, but might add a level of specificity. For example, a suspect

might be arrested for second-degree murder (§ 125.25 in the New York

criminal code) and arraigned for second-degree murder by a depraved

indifference toward human life (§ 125.25.2 in the New York criminal

code). Before the death penalty statute became law, prosecutors had a

sentence range of 25 years to life imprisonment in first-degree murder

cases. The statute gave prosecutors full discretion in choosing whether to

seek the death penalty in cases involving a defendant arraigned for firstdegree

murder, but they were not permitted to seek the death penalty in

cases involving any other murder charge.

After the arraignment of a defendant on first-degree murder charges,

the prosecutor has 120 days to issue a death notice. A death notice signifies

the prosecutor’s intention to seek the death penalty instead of life imprisonment.

After the arraignment the prosecutor must first convince a grand

jury that the evidence against the suspect justifies a criminal trial. If he

does so, the defendant is officially indicted. During the period of time

between the arraignment and indictment, plea bargains can be made. Even

after indictment, most prosecutors continue to offer and accept plea bargains.

One important peculiarity of the New York law is that any defendant

charged with capital murder has the option to plead guilty and receive

If no plea bargain can be brokered between the prosecutor and the

defendant, the case proceeds to a two-part trial. In the first phase the only

question before the jury is the defendant’s guilt. The jury can find the

3. This provision was later challenged in court but was in effect throughout my

entire sample.

Does the Death Penalty Affect Plea Bargaining? 119

defendant guilty of first-degree murder or a lesser charge. If the jury finds

the defendant guilty of first-degree murder, the same jury proceeds to the

sentencing phase of the trial. The jury’s decision must be unanimous for

the defendant to receive a death sentence. In the case of gridlock, the

sentence reverts to life imprisonment. Because each juror knows that he

can vote to convict and then later veto a death sentence, he should not feel

compelled to vote ‘‘not guilty’’ for fear that he would send a defendant to

First-degree murder in New York has been narrowly defined for decades.

Before the death penalty statute passed, only the willful killing of a

law enforcement official was considered first-degree murder. After the law

changed, several new categories were added. Today, contract murder,

serial murder, murder involving torture, murder of witnesses, murder

while serving a life sentence or while escaped from prison, and intentional

change in definition will impact the empirical work in this article, because

the 1995 law did not simply introduce capital punishment, but may also

have changed the pool of defendants charged with first-degree murder.

In New York, as in other states, a prosecutor has several alternatives

when he is confronting a defendant charged with a certain crime. First, he

than a single punishment associated with a given crime. For example, in

many states those convicted of manslaughter can receive prison sentences

of anywhere from three to 12 years. The range of sentences permits the

prosecutor to adjust favorably the sentence within its legal range in return

case, he offers to lower the charge to which the defendant pleads, and the

defendant thus receives a sentence within a range strictly less than that of

the original charge. Clearly, charge bargains benefit the defendant more

4. See Andreoni (1991) for a discussion of how sentence severity may affect the

probability of conviction.

5. For a good summary of New York’s capital punishment statute, see Dao

(1995).

6. There has been very little research distinguishing the two types of bargains,

though McCarthy (1985) finds that a 1975 ban on sentence bargaining in Alaska

substantially lowered the number of guilty pleas.

120 American Law and Economics Review V8 N1 2006 (116–142)

cases in which sentence bargaining took place from cases in which charge

bargaining took place.

2. New York Data and Empirical Strategy

The Capital Defenders’ Office, a state agency charged with representing

defendants in New York capital cases, provides information on the propensity

of DAs to seek the death penalty in each of New York’s 62 counties. As

explained in section 1, if a New York DA intends to seek the death penalty,

he must issue a ‘‘death notice’’ to ensure that the defense attorney has

sufficient time to prepare for a capital case. New York DAs issued 36

death notices between 1995 and 2000. Table 1 shows the number of death

notices and death sentences by county during this period. Brooklyn DAs

have issued the most of any county, while Manhattan DAs have issued

none. By 2000 only six men sat on the state’s death row, because the

majority of the death notices were withdrawn after plea bargains were

brokered between DAs and the defendants. Because of the recent timing

of the New York death penalty legislation and the numerous appeals most

From the State of New York Division of Criminal Justice Services

(DCJS), I obtained individual-level data on all felony arrests in the state

between 1985 and 1998, including the approximately 500 first- and seconddegree

murder cases each year in the state. This data set includes information

on each suspect’s demographic characteristics, prior record, arrest

charge, the final disposition of his case (acquittal, trial conviction, or

plea), and the final disposition charge. The data set also includes a variable

that identifies the county in which the case was tried. One of the key

advantages of the data is that the universe is all suspects arrested, not all

defendants indicted or convicted. Thus, there are no selection issues to

address. Summary statistics appear in Table 2.

The first part of the empirical work employs a differences-in-differences

estimation strategy. My treatment group consists of those arrested for

first- or second-degree murder. As mentioned in section 1, New York

changed its definition of first-degree murder when it reinstated the death

7. This is unlikely to change in the near future, as the state supreme court

Does the Death Penalty Affect Plea Bargaining? 121

penalty. To keep the treatment group consistent across time, I aggregate

first- and second-degree murder defendants, because only this larger group

arrested for burglary, forcible rape, or armed robbery. Like murder defendants,

these felony defendants, if convicted, face long sentences. However,

they are not affected by the change in the death penalty law.

I follow the typical differences-in-differences strategy of comparing the

treatment and control groups before and after the policy change. The

County Death Notices

Since April 1995

Death Notices

Since April 1995

(as % of First-Degree

Murder

Indictments)

Death Sentences

Since April 1995

Albany 2 29% 0

Bronx 1 2% 0

Cortland 1 100% 0

Dutchess 2 50% 0

Erie 1 11% 0

Essex 1 50% 0

Genessee 1 50% 0

Jefferson 1 100% 0

Kings (Brooklyn) 5 15% 1

Monroe (Rochester) 3 50% 1

Onandaga 2 33% 1

Queens 4 33% 0

Rensselaer 2 50% 0

Schoharie 3 100% 0

Suffolk 3 30% 2

Sullivan 2 67% 0

Ulster 1 100% 0

Wyoming 1 100% 0

Total, counties that issued

at least one death sentence

36 18.6% 5

Total, all counties 36 7.16% 5

in that county had served a defendant a death notice. In 44 of the 62 counties, no death notices had been

issued.

8. I more carefully explore the implications of this change in classification in

section 4.

122 American Law and Economics Review V8 N1 2006 (116–142)

following equation yields an estimate of the effect of the policy change on

plea bargaining:

variable coded as one if the defendant was arrested after April 1995 (when

variable coded as one if the defendant was charged with first- or seconddegree

either sentence or charge bargains. The detailed nature of theNew York data

set allows me to distinguish between the two types of bargains. Thus, I define

Variable Mean SD

Plead guilty to original charge .0961 .295

Plead guilty to lesser charge .431 .495

Plead guilty to any charge .527 .499

Male (dummy variable) .914 .281

Black (dummy variable) .548 .498

Hispanic (dummy variable) .304 .460

Age 27.3 8.19

Prior convictions .721 1.11

New York City (dummy variable) .784 .411

MurderNew York City .0456 .209

After (dummy variable) .376 .484

Murder (dummy variable) .0532 .224

Death notice (as % of first-degree

murder indictments)

.121 .148

MurderAfter .00608 .0777

NoticeAfter .0519 .119

MurderNotice .00232 .0260

MurderNoticeAfter .000852 .0165

Observations 145,894

Does the Death Penalty Affect Plea Bargaining? 123

the dependant variables as ‘‘plead guilty to original arraignment charge’’ (i.e.,

facing murder charges, regardless of whether the arrest took place before or

variable picks up any effect of the new law—or other changes occurring

simultaneously—on plea bargaining in felony cases in general. The coefficient

on the interaction term is the differences-in-differences estimate of the effect

of the new law on plea bargaining in murder cases. Two different scenarios

could account for a statistically significant coefficient on the interaction term:

there could be a true effect of the death penalty on plea bargaining, or New

York state merely could have become relatively tougher on murder defendants

vis-a` -vis other defendants at the same time it passed the death penalty

law. There is some reason to suspect the second scenario. Governor George

Pataki, a Republican who ran againstMario Cuomo in part on a ‘‘get-toughon-

crime’’ platform, took office in 1995, promising tougher treatment of

murder defendants. Separating any ‘‘Pataki effect’’ from the effect of the

death penalty is difficult if one uses only variation across time, because only

a couple of months separate the swearing in of the new governor and his

signing into law the new death penalty act.

Using county-level variation in the percentage of first-degree murder

indictments resulting in death notices allows me to distinguish between

these two scenarios. Such a test can be accomplished by estimating the

following equation:

calculated by dividing, for each county, the total number of death notices

issued since the new law became effective, by the total number of firstdegree

murder indictments in that same period (recall that this ratio is

124 American Law and Economics Review V8 N1 2006 (116–142)

listed by county in Table 1). This variable serves as a measure of how

aggressive each county is in pursuing the death penalty. Section 4 examines

whether this statistic is indeed an appropriate measure of county-level

death-penalty policy and experiments with alternative measures.

differences-in-differences-in-differences (DDD) estimate of the effect of the

new statute on plea bargains in murder cases, as distinguished from other

state-level policy changes that might have occurred simultaneously. If the

effect is only at the state level and does not respond to the varying degrees to

which counties used the new the death penalty statute, then we would observe

on the triple interaction term. Such a result would not allow us to distinguish

between the effect of the death penalty and the ‘‘Pataki effect.’’ If, however,

estimating equation (2) yields a significant coefficient on the triple interaction

term, there is reason to believe that the threat of the death penalty is driving

the change in defendants’ behavior. A credible alternative hypothesis would

murder cases but not plea bargaining in other felony cases, that changed at

roughly the same time as the death penalty law, and that changed to a

have to correspond roughly to the county-level variation in death notice

frequency. The difficulty in identifying such a factor inspires confidence in

the DDD estimates, though I later test several alternative hypotheses.

3. Results

Differences-in-Differences Results

Before turning to the regression results, I present graphical evidence of

trends in the raw data. Figure 1 tracks the share of defendants pleading guilty

to their original arraignment charge. The trend is virtually flat for the control

crimes (rape, robbery, burglary). The largest positive shock occurs the year the

legislation was enacted, 1995 (though, as the plea bargain share in 1993 was

quite low, therewas also a substantial gain between 1993 and 1994).Except for

1997, each year after the death-penalty lawwas passed saw higher plea bargain

rates than did any year before the law. Though obviously noisy, the trend for

plea bargains inmurder cases suggests that the probability of pleading guilty to

the original arraignment charge changed substantially in 1995.

Does the Death Penalty Affect Plea Bargaining? 125

The results reported in Table 3 confirm this finding. The first column

estimates the effect of the reinstatement of capital punishment on a defendant’s

probability of pleading guilty to the original arraignment charge. The

defendants) and the control group (rape, robbery, and burglary defendants)

did not have significantly different propensities to plead guilty before the law

was not significantly more likely to plead guilty after the law was passed.

However, the positive and significant coefficient on the differences-in-differences

were more likely to plead guilty to the original arraignment change

after the law changed. Specifically, their probability of pleading guilty rose by

2.8 percentage points, or 26% relative to the baseline probability of 10.6%.

The regression reported in the second column is identical to that in the

first, with the exception that the dependent variable indicates whether the

defendant pleads guilty to a lesser charge (i.e., received a charge bargain).

Recall that defendants prefer charge bargains, because the charge is actually

lowered and they thus face less severe punishments. The negative coefficient

bargains both before and after the law was enacted. The negative coefficient

offences in the period after the law passed (perhaps suggesting a general gettough

mindset of the Pataki administration). The negative and significant

0.04

0.06

0.08

0.1

0.12

1990 1991 1992 1993 1994 1995 1996 1997 1998

Year

Share of defendents pleading guilty to

original charge

Rape, robbery

and burglary

Murder

126 American Law and Economics Review V8 N1 2006 (116–142)

coefficient on the interaction term suggests that the probability a murder

defendant would receive a charge bargain differentially fell after the law

changed. As predicted, the death penalty seemed to decrease the chance that

a murder defendant would receive an especially favorable deal. The point

estimates suggest that the policy decreased the probability that a murder

defendant would receive a charge bargain by about 4 percentage points.

Finally, the regression reported in the last column is identical to the

previous two, with the exception that the dependent variable indicates

whether the defendant pleaded guilty, regardless of the charge. Given the

Variable Plead Guilty

to Original

Arraignment

Charge

Plead Guilty to

a Lesser Charge

Plead Guilty,

Regardless of

Charge

(.028) (.074) (.0908)

(.0028) (.0086) (.0113)

(.015) (.0229) (.0238)

(.0023) (.0036) (.00407)

(.0028) (.0040) (.00696)

(.0034) (.0033) (.00386)

(.00009) (.00021) (.000264)

(.0013) (.0029) (.00203)

(.0077) (.0087) (.0126)

(.010) (.019) (.0199)

Observations 145,894 145,894 145,894

Mean of dependent variable .106 .460 .566

and quadratic time trends, allowing separate trends for murders. Coefficients (and adjusted standard errors)

this table, are .0258 (.0169), –.0445 (.0149), and .0229 (.0166). Probit results are reported as changes in

probability (adjusted standard errors in parentheses).

Does the Death Penalty Affect Plea Bargaining? 127

results in the first column, it is not surprising that the coefficient on the

differences-in-differences estimator is insignificant. Although the death

penalty is associated with an increase in the probability that a defendant

agrees to a sentence bargain, it is associated with a decrease in the probability

that he will receive a charge bargain, and the effect on plea bargains

The coefficients on the interaction terms suggest that the reinstatement

of the death penalty made murder defendants more likely to sentence

bargain and less likely to charge bargain, but did not significantly affect

their tendency to plea bargain in general. These results indicate that DAs

gained bargaining power in murder cases after April 1995, because defendants

seem more willing to accept harsher deals and less likely to receive

generous ones. However, it is unclear if this effect is merely statewide (a

‘‘Pataki effect’’) or sensitive to county-level death-penalty policy.

DDD Results

Table 4 displays the DDD results. The first column, which uses ‘‘plead

guilty to original arraignment charge’’ as the dependent variable, suggests

that the positive effect of the reinstatement of the death penalty on sentence

results, the significance of the coefficient on the interaction term between

the coefficient on the triple interaction term is positive and significant. The

results in the differences-in-differences analysis thus seem to be driven only

by those counties that actually do pursue death sentences.

TheDDDresults reported in columns 2 and 3 suggest that, unlike sentence

bargaining, charge bargaining and overall plea bargaining were not significantly

affected by the new law. Though the differences-in-differences results

reported in Table 3 suggested a negative effect of the death penalty on charge

9. The results in Table 3 are robust to the inclusion of county and year fixed

effects. See the notes to Table 3 for the fixed-effects estimates of the coefficient on

the interaction term.

10. Though not reported here, the results in Table 4 are robust to adding

indicators of the party affiliation of each county’s DA. Thus, the results do not

seem to be driven by Republican DAs’ prosecuting defendants more aggressively

after Pataki’s election or to changes in DA party affiliation correlated with the law

change.

128 American Law and Economics Review V8 N1 2006 (116–142)

Variable Plead Guilty to

Original

Arraignment

Charge

Plead Guilty

to a Lesser

Charge

Plead Guilty,

Regardless of

Charge

After (dummy variable) .00461 –.0231 –.0178

(.00851) (.0343) (.0372)

Murder (dummy variable) –.0627 –.0520 –.133

(.0310) (.112) (.111)

Death notice (as % of first-degree –.0902 .183 .0722

murder indictments) (.0606) (.171) (.154)

MurderAfter –.00381 –.0178 –.0245

(.0179) (.0504) (.0504)

NoticeAfter –.0400 .00149 –.0479

(.0452) (.123) (.120)

MurderNotice .366 –.809 –.0479

(.295) (.508) (.120)

(.115) (.244) (.232)

(.00302) (.00617) (.00655)

(.00273) (.00742) (.00854)

(.00322) (.00611) (.00694)

(.000132) (.000200) (.000232)

(.00106) (.00236) (.00179)

(.00651) (.0174) (.0202)

(.0119) (.0294) (.0277)

Observations 145,894 145,894 145,894

Mean of dependent variable .106 .460 .566

estimates, reported in the same order as the columns on this table, are .228 (.108), –.148 (.248), and .117

(.221). Probit results are reported as changes in probability (Adjusted standard errors in parentheses).

Does the Death Penalty Affect Plea Bargaining? 129

bargaining, this result is not borne out by the DDD estimates. Table 4

indicates that, all else equal, a defendant in a county where 10% of firstdegree

murder defendants are issued death notices is 2.3 percentage points (or

21%) more likely to plead guilty to his original arraignment charge than is a

the differences-in-differences results, theDDD results suggest that defendants

exposed to the threat of capital punishment are more likely to accept plea

bargains with harsher terms.

4. Robustness Checks

One possible objection to the results in the first column of Table 4 concerns

the potential endogeneity of the share of first-degree murder indictments that

result in death notices. This variable is meant to proxy for the DA’s aggressiveness

in pursuing death sentences. However, if defendants know that they

are up against a tough DA, they may plead guilty before a death notice can

even be issued. Such behavior may bias the estimate of the effect of death

notices on plea bargains toward zero, so Iwould expect amore perfect measure

of prosecutorial zeal to yield even larger estimates of the effect in question.

One way of testing the direction of the bias is to use a less continuous

measure of county-level death-penalty policy. First, I substitute for the share

of first-degree murder indictments resulting in a death notice a dummy

variable coded as one if the county has ever issued a death notice, and as

zero otherwise. The logic behind this estimation is that the strategic behavior

of defendants might contaminate the exact share of murder cases resulting in

death notices, but is less likely to contaminate whether a county has ever

issued a death notice.Where DAs unequivocally oppose capital punishment,

the dummy variable will equal zero. Where DAs support the death penalty,

there is likely to have been at least one case in which a death notice was

issued. Measurement error will persist in those counties whose DAs support

the death penalty but, perhaps because of small populations, have yet to

process a case in which a death notice could be issued.

The first column of Table 5 reports the results of a regression identical to

that in the first column of Table 4, with the exception that I substitute the

of the coefficient on the triple interaction term.

130 American Law and Economics Review V8 N1 2006 (116–142)

above-mentioned dummy variable for the death notice percentage. The results

are very similar to those inTable 4. Though the triple interaction term using the

dummy variable loses some significance, it is still positive and significant at the

.1 level. The results indicate that defendants facingmurder charges in a county

49.8%) more likely to plead guilty to the original arraignment charge than

How does this compare to the results in Table 4? As Table 1 reports,

conditional on a county’s ever issuing a death sentence, the average (weighted

by the number of defendants) share ofmurder cases resulting in deathnotices is

0.186. Hence, the coefficients in Table 4 indicate that a defendant in a county

that has ever issued a death sentence is 0.1860.227 = 4.2 percentage points

more likely to plead guilty to his original arraignment charge. Comparing this

County Death Penalty Policy, New York Felony Data

Variable Plead Guilty to Original Arraignment Charge

(.00113) (.00575)

(.0112) (.0162)

(.015) (.0191)

MurderAfter –.00621 –.00984

(.00810) (.0146)

(.00476) (.00689)

MurderNotice dummy .0702 –.0142

(.0363) (.0311)

(.0203) (.0296)

Sample restrictions All counties New York City

Observations 145,894 114,458

Mean of dependent variable .106 .0998

Each regression includes linear and quadratic time trends, allowing separate trends for murders, and separate

(adjusted standard errors in parentheses).

Does the Death Penalty Affect Plea Bargaining? 131

result to the 5.3 percentage-point effect found when the death notice dummy

variable is used suggests that, as suspected, the strategic behavior of defendants

downwardly biases the estimated effect of the policy when the exact share of

murder cases resulting in a death notice is used as the explanatory variable.

As mentioned, it is possible that some counties with zero death notices

have prosecutors who are actually zealous proponents of capital punishment.

To address this concern, I limit the sample to New York City, where

the DA in each of the five counties made public his position on the new law

(those in Manhattan, the Bronx, and Staten Island were opposed, whereas

those in Brooklyn and Queens supported it). Thus, there is no need to glean

from county death notice statistics the DAs’ probable positions. The second

column of Table 5 reports the results when only the most conservative

measure of death penalty policy is used: a simple for-or-against dummy

variable for each of the five boroughs. The coefficient is slightly larger than

in the second column and highly significant, consistent with the results in the

Another possible objection to the results in Table 4 relates to the fact

that the same 1995 act that legalized the death penalty in New York also

expanded the definition of first-degree murder. Before the 1995 act firstdegree

murder was limited to those who willfully killed a law enforcement

official, while second-degree murder was a catchall category including

murders ranging from contract killings to actions displaying a ‘‘depraved

indifference toward human life.’’ As discussed in section 1, the death

penalty legislation added several new categories to the first-degree murder

classification. As a result of this redefinition, my ‘‘murder’’ category

includes both first- and second-degree murder defendants, since only this

aggregate group did not change over time.

The change in murder definitions raises two issues. First, since the firstdegree

share of murder arrests rose after the death penalty law, the law’s

effect on plea bargains could be merely a ‘‘first-degree’’ effect—that is,

defendants may be more likely to plead guilty to first-degree murder than

to second-degree murder. However, such a first-degree effect would bias

12. When restricted to cases outside New York City, the coefficients on the

regressions in Tables 3–5 do not change sign but usually lose their significance

because of, perhaps, the much smaller sample size (New York City accounts for

more than three-fourths of the data).

132 American Law and Economics Review V8 N1 2006 (116–142)

the DDD results only if the increase in the first-degree share of murder

arrests were disproportionately higher in counties that tended to use the

death penalty.

I examine the data to see whether this is in fact the case. In counties with

DAs who have issued at least one death notice, the percentage of first- and

second-degree murder arrests that are for first-degree murder rose from 1.0

before the death penalty law was enacted, to 2.4 after the enactment. In

counties with DAs who have not sought the death penalty, that same

figure rose from 2.1 to 4.5. Thus, the first-degree share of first- and

second-degree murder arrests rose at roughly the same rate in the two

sets of counties. Additionally, it is not clear that there is any first-degree

effect with respect to plea bargaining. Before the law changed—when the

death penalty did not affect the decisions of first-degree murder defendants—

the percentage of first-degree murder defendants who pleaded

guilty to their original charge is nearly identical to (actually, slightly less

than) the percentage of second-degree murder defendants who pleaded

guilty to their original arraignment charge.

The second issue relates to the interpretation of the results in Tables 4

and 5. The coefficient on the triple interaction term estimates the increased

propensity of those arrested for first- and second-degree murder to plead

guilty after the law changed. However, for most of those defendants, the

death penalty is not relevant, because it applies only to those arraigned on

first-degree murder charges, suggesting that the true effect on first-degree

murder defendants is underestimated by the results in Tables 4 and 5.

These estimates are roughly analogous to an ‘‘intent to treat’’ measure, in

the parlance of program evaluation.

To determine the effect of the ‘‘treatment on the treated,’’ one first needs

to determine who exactly is in the treatment group. As Table 1 shows, only

36 defendants have been issued death notices in New York. However, the

death penalty’s effect may not be limited to these 36 defendants, because its

specter may have encouraged some defendants to secure plea bargains

after the DA merely mentioned a death sentence as a possibility but before

he actually issued a death notice.

As a rough estimate of the effect of the death penalty on those who

actually face it, I scale the effect measured in the DDD estimate by the

inverse of the ratio of first-degree murder indictments to all murder arrests.

Does the Death Penalty Affect Plea Bargaining? 133

Table 1 reports that this ratio is 0.0716. Taking the coefficient from the

first column of Table 4, 0.227, as the preferred estimate of the effect of the

law, I calculate that a murder suspect in a county where 5% of the firstdegree

murder defendants are issued a death notice is roughly 16 percentage

points (0.2270.05(1/0.0716) = 15.9) more likely to plead guilty to his

original arraignment charge than is a first-degree murder defendant in a

county where the DA does not pursue the death penalty.

Note that this calculation implicitly assumes that the positive coefficient

of those who actually face the possibility of capital punishment to

plead guilty. If this is not the case, then the scale factor is too large and the

effect on the treatment group is overestimated. Although providing an

exact value is difficult, the true measure of the treatment on the treated

probably lies somewhere between the value of coefficient on the triple

interaction term in Table 4 and the scaled estimate just reported.

I address one final issue related to the death notice share variable.

As Table 1 reports, one death notice was issued in Bronx County. However,

Robert Johnson, the county’s longtime DA, did not issue it. When Johnson

announced that he would not seek the death penalty against first-degree

murder suspect, Angel Diaz, Governor Pataki removed Johnson from the

case and replaced him with a prosecutor from the state’s Attorney General’s

Although this move has a number of interesting jurisdictional and legal

implications, only the rather mundane question of how to code the death

notice variable for Bronx County will be addressed here. The share of firstdegree

murder indictments resulting in death notices ostensibly represents

the tendency of DAs to seek the death penalty. However, the death notice

opponent of capital punishment, to seek the death penalty. Though not

reported, all of the results in Tables 4 are robust to recoding the explanatory

variables as if the Diaz death notice had never been issued, and to

excluding the Bronx from the sample altogether.

Although the New York results appear to be quite robust, there are a

number of issues surrounding the state’s statute that limit my ability to

13. Johnson challenged Pataki’s action in court and lost. Because Diaz committed

suicide while awaiting trial, Pataki’s prosecutor never actually tried the case.

134 American Law and Economics Review V8 N1 2006 (116–142)

generalize these results to the rest of the nation. First, New York’s law is

prosecutors wielded a more expansive statute. Second, New York’s unusual

provision allowing defendants the right to plead guilty and receive life

imprisonment makes generalization risky. Finally, as already mentioned,

the state initiated a Capital Defenders Office (CDO) as part of the capital

punishment statute. Because the CDO attracts many well-trained attorneys

who oppose capital punishment, murder defendants tried after the new

statute was passed may have been represented by better attorneys than

those tried before 1995. Thus, in the absence of the CDO, we might have

seen an even greater increase in the likelihood that defendants would

accept plea bargains with relatively harsher terms. In order to address

the potential idiosyncrasies of the New York experiment, I turn to a

nationally representative sample of murder defendants.

5. National Cross-Section of Murder Defendants in 1988

This section examines murder defendants across the country and

exploits state-level variation in death penalty policy. I draw primarily

33 counties in 1988, gathered by the U.S. Bureau of Justice Statistics

(1996). The data set provides information on each case’s disposition,

including whether the defendant was acquitted or convicted, and whether

the defendant pleaded guilty to the original arraignment charge or pleaded

guilty to a lesser offense. The data set also includes demographic characteristics

of the defendant and the victim in each case. I supplement this

data on city and county demographic and crime rate information provided

14. Unlike most other states, New York does not permit the execution of the

mentally retarded or of defendants who committed their capital offense before their

eighteenth birthday. In addition, felony murder, of which the majority of U.S.

death row inmates were convicted, is not subject to the death penalty in New York

unless the prosecutor can show that the murder itself was intentional.

Does the Death Penalty Affect Plea Bargaining? 135

from the U.S. Bureau of Justice Statistics (1999), which provides the

number of death sentences and executions per year per state. Summary

statistics are reported in Table 6.

The obvious problem with comparing murder defendants in states

where death sentences are commonly issued with those in states where

they are rarely or never issued is the possibility that other state-level

difference are correlated both with states’ death penalty use and with

defendants’ tendencies to plead guilty. If, for example, states that send a

large share of their murder defendants to death row also underfinance

public defenders’ budgets (increasing the caseload of defense attorneys and

making them more willing to accept plea bargains), then regressing a guilty

plea dummy variable on the percentage of murder defendants sent to death

row in each state would yield a positively biased coefficient.

To address this omitted-variables problem, I employ, once again, a

differences-in-differences estimation strategy. In this case, the treatment

group includes defendants arrested for first-degree murder, and the control

group includes murder defendants arrested for lesser charges. Because all

states that have reinstated the death penalty restrict its application to

Variable Mean SD

Plea to original charge .157 .364

Plea to a lesser charge .438 .496

Plea to any charge .281 .450

State has death penalty (dummy variable) .883 .322

First degree (dummy variable) .832 .374

Death penaltyFirst degree .758 .428

Death sentences / homicide arrests .0160 .0116

Death sentences / homicide arrestsFirst degree .0132 .0116

Male (dummy variable) .884 .320

Black (dummy variable) .593 .491

Hispanic (dummy variable) .166 .372

Prior convictions .940 2.18

White female victim (dummy variable) .104 .305

Multiple murder (dummy variable) .045 .208

County population 2,288,513 2,155,316

Black share of county population .232 .153

County arrest rate .0673 .019

South (dummy variable) .365 .482

SouthBlack .226 .418

Observations 2,322

136 American Law and Economics Review V8 N1 2006 (116–142)

defendants who have been convicted of first-degree murder, those defendants

facing second-degree murder or lesser charges serve as a natural

on this group of defendants would control for any general states’ criminal

justice system differences that are correlated with their death penalty policies.

As Table 6 indicates, approximately one-fifth of the defendants in the

The differences-in-differences equation that I estimate takes the following

form:

variable coded as one if the defendant was arrested on first-degree murder

I use the same three variations on the dependent variable as in Tables 3

and 4: ‘‘plead guilty to the original arraignment charge,’’ ‘‘plead guilty to a

lesser charge,’’ and ‘‘plead guilty, regardless of charge.’’ Thus, analysis

parallel to that performed on the New York data with respect to charge

and sentence bargaining can be carried out here, as well.

of a death sentence, given arrest for murder, which I approximate by the

ratio of death sentences in 1987 to homicide arrests in 1986 per state. I lag

the numerator by one year before the year in which the cases were disposed

(1988) in order to eliminate possible simultaneity between the dependent

15. Georgia—where any homicide is potentially a capital offense—is an exception.

which accounts for their large share of the sample.

Does the Death Penalty Affect Plea Bargaining? 137

and explanatory variables. I lag the denominator in this ratio by one

additional year because murder cases generally take about one year to

dispose (so, a defendant arrested in 1986 would likely be sentenced in

1987). However, all of the results in this section are robust to ignoring

either or both of these lags. This ratio varies widely across states, with

Maryland sentencing about 0.2% of those arrested for murder to death,

If states in which defendants are frequently sentenced to death also

classify more homicides as first-degree murder, then the coefficient on the

interaction term will pick up this difference. There does not appear to be a

connection between the frequency of death sentences in a state and the

first-degree share of murder arrests; the correlation between the 1987-

death-sentences-to-1986-murder-arrests ratio and the share of murder

In the first column of Table 7, the coefficient on the interaction term

dummy variable is positive and significant. The point-estimate suggests that,

all else equal, a first-degree murder defendant in a state (such as Oklahoma)

where the death-sentence-to-homicide ratio is 0.06 is 12.4 percentage points

(or 79%, as the baseline probability is .157) more likely to plead guilty to his

original charge than is a defendant in a state (such as Michigan) where that

ratio is zero. The second column reports that the death penalty has the

opposite effect on charge bargaining. The negative and highly significant

coefficient on the interaction suggests that first-degree murder defendants in

states where the death-sentences-to-homicide-arrests ratio is 0.06 are 9.5

17. Ideally, I would have liked to use other proxies for state death penalty

policy, such as executions per year per state. However, only three of the states in

my sample (Texas, Florida, and Louisiana) executed anyone between 1985 and

1988. Similarly, a dummy variable for whether the state had reinstated the death

penalty is of limited use, because only three states in my sample (Michigan,

Massachusetts, and New York) had not done so by 1988.

18. The exception is New York. Recall that, until 1995, New York had an

unusually narrow definition of first-degree murder. Thus, although first-degree

murder defendants comprise about 80% of the entire sample, less than 2% of the

New York defendants faced first-degree murder charges. Although first-degree

murder definitions vary somewhat across states, the statutes of the remaining states

in the sample are reasonably comparable. For this reason, I exclude defendants

from New York.

138 American Law and Economics Review V8 N1 2006 (116–142)

a similarly situated defendant in a state where that ratio is zero. As should be

expected, the third column reports that there is little effect of the death

The results in this section generally confirm those from the New York

State analysis, although the national analysis suggests a stronger effect of

the death penalty on charge bargaining. It is difficult to compare the exact

Variable Plead Guilty to

Original Arraignment

Charge

Plead Guilty to a

Lesser Charge

Plead Guilty,

Regardless of

Charge

(.0552) (.0168) (.0328)

Death sentences to homicide

arrestsFirst degree

(1.15) (.512) (1.15)

(.0219) (.0269) (.0416)

Black .0345 .0264 .0639

(.0214) (.0362) (.0398)

Hispanic .00171 .0386 .0475

(.0190) (.0326) (.0333)

Prior convictions –.00184 –.000273 –.00339

(.00421) (.00472) (.00630)

White female victim –.00556 .00823 .00408

(.0174) (.0226) (.0339)

(.00839) (.127) (.0992)

SouthBlack –.0200 –.0200 .0384

(.0415) (.0415) (.0497)

Mean of dependent variable .157 .281 .438

Observations 2,322 2,322 2,322

are included in all regressions. Probit results are reported as changes in probability (standard errors in

parentheses).

19. Though not reported here, the results in Table 7 are robust to replacing

county fixed effects with county-level death penalty variables, to including a

dummy variable coded as one if the state had reinstated the death penalty by

1988 and as zero otherwise, or to excluding states that had not yet reinstated the

death penalty.

Does the Death Penalty Affect Plea Bargaining? 139

magnitudes of the point estimates, because the proxies used for death

penalty policy are different. But both sets of results indicate that the

death penalty makes defendants more likely to plead guilty to their original

charge, yet no more likely to plead guilty in general—suggesting that the

threat of capital punishment encourages defendants to accept deals that

they otherwise would have rejected.

6. Conclusion

The findings here suggest that the threat of the death penalty leads more defendants to plead guilty to their original arraignment charges. The positive effect of the reinstatement of capital punishment on New York murder defendants’ propensity to plead guilty to their original charges is robust to several tests and does not appear to be the result of simultaneous changes in the state’s criminal code or political climate. Results from a national cross-section of murder defendants generally support this conclusion.

This article also demonstrates the importance of distinguishing between charge bargains and sentence bargains in empirical work on guilty pleas.

While the death penalty appears to encourage sentence bargains, it does not encourage charge bargains. In fact, in all regression results reported here, the death penalty has a negative (sometimes significant) effect on charge bargains. The offsetting effects of the death penalty on the two types of plea bargains can explain why I found no significant effect of the death penalty on guilty pleas in general. By separating the two types of plea bargains, this study was able to uncover a positive connection between the death penalty and the DA’s bargaining power.

These results are potentially relevant in at least two areas of criminal justice policy. First, since the death penalty affects only the terms of the deals defendants make but does not seem to increase defendants’ propensity to plea bargain in general, the threat of the death penalty does not seem to reduce the total number of cases that proceed to trial. Thus, the well-documented costs of capital trials do not appear to be offset by reducing total trial costs through plea bargains. Second, while the death penalty does not discourage defendants from going to trial, it does increase the bargaining position of DAs. Opponents of capital punishment might argue that, based on this finding, the threat of the death penalty is powerful enough to compel an innocent person to accept a plea bargain.

This work is, to the best of my knowledge, the first to examine the effect of the death penalty on plea bargaining. It remains to be seen whether defendants and DAs react only to particularly dramatic shifts of sentence structure, such as the introduction of capital punishment, or whether shifts in bargaining power result after sentencing reforms for lesser crimes.

Addressing this question would be a natural direction for future research.

Given that the vast majority of criminal cases are resolved by guilty plea, researchers and policy makers would be wise to consider the effects of sentencing laws on this aspect of the criminal justice system.

References

Andreoni, James. 1991. ‘‘Reasonable Doubt and the Optimal Magnitude of Fines:

Cameron, Samuel. 1994. ‘‘A Review of the Econometric Evidence on the Effects of

Chressanthis, George A. 1989. ‘‘Capital Punishment and the Deterrent Effect

http://www-pps.aas.duke.edu/people/faculty/cook/comnc.pdf (accessed March 8, 2006).

Cover, James Peery, and Paul D. Thistle. 1988. ‘‘Time Series, Homicide, and the

Ehrlich, Isaac. 1975. ‘‘The Deterrent Effect of Capital Punishment: A Question of

———. 1977. ‘‘Capital Punishment and Deterrence: Some Further Thoughts and

Grogger, Jeffrey. 1990. ‘‘The Deterrent Effect of Capital Punishment: An Analysis

Does the Death Penalty Affect Plea Bargaining? 141

Grossman, Gene, and Michael Katz. 1983. ‘‘Plea Bargaining and Social Welfare,’’

Katz, Lawrence, Steven D. Levitt, and Ellen Shustorovich. 2003. ‘‘Prison Conditions,

McCarthy, L. 1985. ‘‘The Effect on Plea Bargaining of the Form of Concessions:

A Theoretical and Empirical Analysis.’’ Ph.D. dissertation, Purdue University.

N.Y. Penal Law, § 125.25 (McKinney 1975).

Ostram, B. 1988. ‘‘An Economic Analysis of the Plea Bargaining Process.’’ Ph.D. dissertation, University of Washington.

People v. LaValle, 3 NY3d 88, 132 (2004).

Rhodes, William. 1976. ‘‘The Economics of the Criminal Courts: A Theoretical

Consortium for Political and Social Research.

ed. Ann Arbor, MI: Inter-university Consortium for Political and Social Research.

DC: US Government Printing Office.

Reporting Program Data [United States]: County-level Detailed Arrest and

and Social Research.

142 American Law and Economics Review V8 N1 2006 (116–142)