Article

Judicial Politics, Death Penalty Appeals, and Case Selection: An Empirical Study

John Blume & Theodore Eisenberg

I. Introduction

Several studies try to explain case outcomes based on the politics of judicial selection methods.[1] Scholars usually hypothesize that judges selected by partisan popular elections are subject to greater political pressure in deciding cases than are other judges. No class of cases seems more amenable to such analysis than death penalty cases. Indeed, strong anecdotal evidence suggests a relation between the politics of selection methods and death penalty appeal outcomes. For example, electoral removals of Chief Justice Rose Bird in California and Justice Penny White in Tennessee were directly linked to decisions in capital cases.[2] A 1980s study, moreover, suggests that states’ judicial selection methods correlate with capital appeal outcomes.[3] Thus, one expects judges who must face voters to be more reluctant to overturn death sentences.

No study, however, accounts both for judicial politics and case selection, the process through which cases are selected for death penalty litigation. Yet, the case selection process cannot be ignored because it yields a set of cases for adjudication that is far from a random selection of cases.[4] Effects based on judicial selection politics can only reliably be detected if one accounts for this filtering of cases judges adjudicate.[5]

In death penalty litigation, the case filtering process begins with the prosecutor’s decision to seek the death penalty. Some prosecutors are said to do so in virtually every case that arguably satisfies their state’s criteria for possible imposition of the death penalty.[6] Others exercise substantially more discretion.[7] These differences should lead to interstate variation in the death-worthiness of cases trial courts adjudicate. Consider two prosecutors: One prosecutor only seeks the death penalty in cases involving defendants who have been previously convicted of murder, were released, and subsequently murder again. The second prosecutor seeks death in every possible death-eligible case. While acknowledging the difficulty attending any formal definition of "death-worthiness," we expect the first prosecutor’s collection of capital cases to be, on average, more death-worthy than the second prosecutor’s collection of capital cases.

The case filtering process affecting reviewing courts’ case mix continues at the adjudicatory stage. The trial adjudicator, either the judge or the jury, may cushion the effect of extreme prosecutorial death-seeking behavior. Prosecutors may seek death, but the adjudicator need not impose it. So the most extreme variations in death-seeking behavior likely fade by the time reviewing courts address the residue of appealed death penalty cases. Indeed, the adjudicators themselves may skew the reviewing courts’ case mix. Given prosecutors with the same death-seeking behavior, more death-prone adjudicators will present reviewing courts with a different mix of death penalty cases than less death-prone adjudicators. The combined effects of prosecutors’ and adjudicators’ behavior need not lead to imposing the death penalty in equally death-worthy cases across states. And, in fact, we find that states obtain the death penalty at widely varying rates.

Variation in the rates at which prosecutors seek death penalties and adjudicators impose them thus ought to influence the rate at which reviewing courts overturn capital sentences. In the extreme, a state in which prosecutors seek and obtain the death penalty in every possible case will, on average, submit for judicial review a class of cases that is less death-worthy than a state in which prosecutors seek or obtain the death penalty only in the most egregious of the eligible cases. The group of less death-worthy cases ought to be subject to more frequent reversal than a more selectively filtered group of cases.

In theory, case selection could frustrate efforts to detect political effects. Life-tenured, fully independent judges may challenge death sentences infrequently because they serve in a jurisdiction in which prosecutors and adjudicators are highly selective about the cases in which they seek and impose death. Conversely, judges selected in highly partisan elections may reverse a significantly higher number of death sentences because prosecutors and adjudicators in their jurisdiction show little restraint in seeking and imposing the death penalty. Detecting the effect of judicial selection politics occurs against this background of case selection and must account for it.

Two aspects of the impact of judicial selection politics are worth separating. The first focuses on interstate differences in death penalty case processing. Methods of state judicial selection are believed to correlate with death penalty outcomes. The more independent the state’s judiciary, the more likely the judiciary is to scrutinize death sentences without fear of reprisal. A second aspect of judicial selection politics relates to possible differences between state and federal judges. Unlike most state judges, federal judges are independent of the electoral process. Whether or not interstate differences exist in capital case processing, we might expect federal judges to be more willing to question death sentences than state judges.

This Article uses two databases to explore factors affecting grants of relief from death penalties. The first consists of approximately 800 appeals of death sentences decided from 1995 to 1997. It consists solely of direct appeals in capital cases and is limited to state courts. The second database is the Bureau of Justice Statistics’ ("BJS") database of all persons sentenced to death from 1973 to 1995.[8] It includes information about whether a defendant obtained relief from a death sentence and includes post-conviction relief. The source of relief could be federal or state court but the database does not reveal which court system acted to grant relief.

Accounting for case selection, as measured by states’ rates of obtaining death penalties in murder cases, helps explain the pattern of relief from death sentences in the BJS data. The rate at which states impose sentences strongly correlates with the rate at which relief was obtained from those sentences. But case selection does not help explain the pattern of reversals in state court death penalty appeals from 1995 to 1997. That is, we find no correlation between states’ rates of obtaining death penalties and state courts’ reversing capital convictions or sentences.

Furthermore, we find little evidence that states’ methods of selecting judges correlates with outcomes of death penalty cases. The first aspect of judicial selection politics, interstate variation in relief from death sentences, does not depend on mode of judicial selection. The absence of such a correlation holds whether or not one accounts for case selection.[9] And the absence of correlation holds for both the BJS data and the state court appellate case data.

Although we find no system-wide evidence of the effect of state judicial election methods on case outcomes, developments in individual states do confirm the death penalty’s politically charged character. But the pattern of challenges to individual state justices and the responses manifested in subsequent death penalty adjudication are not readily explained by the conventional belief that justices are most vulnerable in states with partisan judicial elections. For example, California is generally not classified as a state with partisan judicial elections. Yet death penalty politics played a major role in reconstituting the California Supreme Court.[10] So interstate differences in death penalty case processing are not readily explicable by judicial selection methods.

We find indirect evidence that the federal judiciary processes death penalty cases differently than state courts. The correlation in the BJS data between states’ death-obtaining behavior and relief from death penalties suggests that death penalty decisionmaking that includes the federal judiciary reflects differences in the underlying death-worthiness of cases. The absence of such a correlation in the state court appellate data could be interpreted as evidence of pressure to affirm death penalty cases with less attention to their death-worthiness.

Studying states’ varying death-obtaining behavior also offers a new perspective on states’ overall capital case behavior. In particular, states with large death rows are not necessarily the states that most vigorously pursue capital punishment. California and Texas have large death rows. But California obtains death penalties at a lower rate per murder than any other major death penalty state. And Texas’ death-obtaining rate is not noticeably different from that in other capital punishment states.

One caveat is in order: Appellate judges of course are not always free to reverse capital case convictions or death sentences based on personal notions of death-worthiness. They are constrained by the requirement that there be legal error—although many states allow for proportionality review.[11] We assume that the tendency to detect legal error is related to some underlying notion of death-worthiness, though not solely determined by perceived death-worthiness.

Part II of this Article documents recent campaigns to oust judges based on their decisions in death penalty cases. These campaigns suggest that judges subject to partisan elections will be reluctant to question death sentences. Part III explains our methodology for exploring the relationship among judicial elections, states’ death-obtaining rates, and judicial review of capital cases. Finally, Part IV reports our empirical results.

II. Campaigns To Oust Judges Based on
Death Penalty Decisions

Several initiatives have sought to unseat state supreme court judges because of their votes in capital cases.[12] The most prominent recent campaign involved former Tennessee Supreme Court Justice Penny White. White was appointed to the state’s highest court in 1995. She participated in only one capital case during her eighteen-month tenure on the court: State v. Odom,[13] a case in which the defendant’s death sentence was unanimously reversed.[14] White did not author the court’s opinion; a review of the Odom decision revealed that reversal of the defendant’s death sentence was mandated by previously existing state law which White had no role in creating.[15]

Justice White’s retention election came soon after the Odom decision. The new Republican Governor of Tennessee—Justice White was a Democratic appointee—in conjunction with several victim’s rights groups, orchestrated a high profile campaign to unseat White because of her concurring vote in Odom. The campaign attacked White as putting "‘the rights of criminals before the rights of victims,’" and as believing that repeatedly raping and stabbing to death a seventy-eight year old woman "wasn’t heinous enough for the death penalty."[16] It worked and White was not retained. Since then, other members of the Tennessee Supreme Court have announced that they will not stand for reelection.[17] When the Governor of Tennessee was asked after White’s defeat if a judge should "look over his shoulder" when making decisions, he responded: "I hope so."[18]

Texas went through a similar experience in 1994. A Texas Court of Criminal Appeals decision reversing a high profile capital case led the Chairman of the state Republican Party to call for a takeover of the court.[19] Voters answered the call as Republican candidates won every position they sought on the court. One of the new judges, Stephen W. Mansfield, campaigned for the court on promises of expansion of the death penalty, greater use of the harmless-error doctrine, and sanctions for attorneys who file "frivolous appeals especially in death penalty cases."[20]

In 1992, Mississippi Supreme Court Justice James Robertson was defeated in his reelection bid. The central issue in the campaign was the death penalty. One advertisement used by his opponents urged the citizens of Mississippi to "vote against Robertson because he’s opposed to the death penalty and he wants to let all these people go."[21] His opponent in the Democratic primary ran as a "law and order candidate" with the endorsement of the Mississippi Prosecutors’ Association. He declared that Robertson thought that "a defendant who ‘shot an unarmed pizza delivery boy in cold-blood’ had not committed a crime serious enough to warrant the death penalty."[22]

Chief Justice Exum of the North Carolina Supreme Court was opposed because of his antideath penalty views.[23] He was forced to fight back with statistics outlining the number of cases in which he had voted to affirm the death sentence. Although he was reelected once, he announced that he would not run again when his current term expired in 1998.[24]

Perhaps the best known of these incidents occurred in California. In the mid-1980s three members of the California Supreme Court, including Chief Justice Rose Bird, lost their jobs in a retention election dominated by the death penalty.[25] The Governor was an active participant in the ultimately successful attempt to replace the judges, and went so far as to indicate to two members of the court that he may change his mind about opposing them if they voted to uphold more death sentences.[26] After the court reversed several other capital cases, the governor carried out his threat[27] and successfully opposed the retention of all three members of the California Supreme Court.[28] Today, the California Supreme Court has a high affirmance rate in capital cases.[29]

Pressure on the judiciary can be more subtle. In South Carolina, for example, judicial elections had never been politicized. No sitting justice of the state supreme court had been challenged since 1893.[30] However, when Justice Toal—the supreme court’s first female justice—stood for reelection in 1995, she was attacked as a liberal judge who was "soft on crime."[31] Republicans, who had recently gained control of the state House of Representatives, were reported to have encouraged a state trial judge to run against Toal.[32] Although Toal was ultimately successful in her reelection bid, the process had become politicized.[33]

In 1994, South Carolina’s new state Attorney General, Charlie Condon, was elected on the basis of his strong pro-death penalty views. As a prosecutor, Condon had sent eleven men to death row. Both in his campaign and following his election, he attacked the state and federal courts for the lack of executions in South Carolina, making the widely publicized comment that South Carolina’s death row was one of the safest places in the state because of the courts’ lenient treatment of inmates serving death sentences.[34] He was also instrumental in eliminating federal funding for death penalty resource centers, organizations that represented indigent inmates sentenced to death.[35] Furthermore, when a federal district court judge granted habeas corpus relief in a capital case, Condon was integrally involved in a well publicized effort to have the judge impeached.[36] The Attorney General remarked that "[j]ustice has been mocked"[37] and used the case as an example of the need for "dramatic reform"[38] of the death penalty system. The message to South Carolina’s state judges from this series of events was unmistakable: Reverse capital cases at your own peril.[39]

III. Methodology and Descriptive Statistics

Capital punishment-related campaigns may lead those judges subject to partisan elections to be more reluctant than other judges to question capital convictions. By partisan elections, we mean elections in which political parties are directly involved in the selection of candidates.[40] We rely on traditional classifications of states as having partisan or nonpartisan elections.[41]

Although we report in detail only the differences between states with partisan elections and all states, other categorizations of state judicial selection methods are plausible. One could, for example, distinguish between states with merit selection and other states, or between states with elections (partisan or nonpartisan) and other states. The alternative categorizations we explored do not yield results materially different from those reported here.[42]

To test hypotheses about the relation between judicial selection methods and capital case outcomes, one needs a measure of the outcome of capital case appeals or other efforts employed by prisoners to obtain relief from death sentences. But that measure should be evaluated in light of the process by which cases are selected for capital treatment. Consequently, one requires not only a measure of capital case outcomes, but also a measure of how cases are selected for capital punishment treatment.

A. Measuring States’ Death-Obtaining Behavior

To assess selection of cases for capital treatment, one would ideally like to know the death-worthiness of each potential capital case considered by prosecutors, as well as the prosecutors’ decisions whether to seek death. However, we lack case-level knowledge of individual murder cases; we do not know the details of each case that a prosecutor may have considered for capital treatment. We therefore cannot independently assess whether cases warranted seeking death sentences. Second-best methods of measuring thus have to suffice. To construct indices of states’ death-obtaining behavior, we estimate the number of potential death penalty cases and the number of death penalties actually imposed. We use these estimates to construct a measure of death-obtaining rates which allows for interstate comparisons.

1. The Number of Potential Death Penalty Cases in Each State

We use the number of murders in each state as a proxy for the number of possible death penalty cases. A crucial assumption is that the number of death-eligible murders correlates with the number of murders. The correlation need not be perfect, but it ought to be substantial. The FBI’s Uniform Crime Reports indicate the number of murders for each state for each year.[43] The murder data are considered to be among the most reliable data in the Uniform Crime Reports.[44]

2. The Number of Capital Sentences in Each State

Given each state’s number of murders, we need measures of each state’s number of death penalties.[45] The BJS data supply the number of persons sentenced to death in each state in each year. Portions of those data covering different time periods are used for our two different databases, the BJS data and appellate opinions. For the BJS data covering relief from death sentences, we cumulate the number of persons sentenced to death for the years 1985 to 1994. This yields the total number of death sentences obtained by each state for those years. For the appellate opinion data, we use the number of BJS-reported death sentences for the period 1994 to 1996. The time periods were chosen to coordinate with the periods covered by the BJS and appellate opinion databases, described in more detail below.

3. States’ Death-Obtaining Indices

We combine our measures of the number of possible death cases (as measured by the number of murders) and the number of death penalties to compute states’ death-obtaining rates. For the BJS data, we divide the number of persons reported by the BJS to have been sentenced to death in each state from 1985 through 1994 by the number of murders in the state, based on the 1985 to 1994 FBI Uniform Crime Reports. For the appellate opinion data, we divided the number of persons reported to have been sentenced to death from 1994 to 1996 by the number of murders in the state, based on 1994 to 1996 FBI Uniform Crime Reports.

The utility of these death-obtaining rates rests on the assumption that different rates of obtaining death sentences correlate with differences in trial level behavior and not solely with different underlying distributions of the death-worthiness of murder cases. In other words, we assume that groups of murders were approximately equally death-worthy across states. The trial-level process involves decisionmaking by prosecutors, judges, and juries. If interstate variation exists in these groups’ behavior, appellate courts will review sets of cases with varying levels of death-worthiness from state to state. We suspect that the greatest variant in trial level death-obtaining behavior is the prosecutors’ differing propensities to seek the death penalty.[46]

State law could also affect death-obtaining rates. For example, some states are regarded as having death penalty laws that make it easier to obtain death sentences than in other states.[47] States with laws that facilitate obtaining death penalties might be viewed as generating a less death-worthy set of cases for appellate review. But this need not be the prediction based on laws with varying severity. To illustrate, assume that State A’s law favors death more than State B’s. Holding constant prosecutor and adjudicator behavior, the potential number of death penalties per 100 murders in State A is ten and the potential number in State B is five. Assume that all potential death sentence cases in fact lead to death sentences. Reviewing courts, applying the same laws as the prosecutors and adjudicators, may affirm these sentences at the same rates. Different death-obtaining rates need not lead to different reversal rates because the states have, in effect, generated different criteria of death-worthiness.

Intrastate differences among laws could confound interpretation of the relation between death-obtaining rates and reversal rates. We might observe constant reversal rates for different death-obtaining rates simply because of different legal rules. We do not address this issue in detail for two reasons. First, it is our judgment that state death penalty laws, despite differing structures, are much more alike than different.[48] The basic structure is, in one way or another, to define a class of aggravated murders. The circumstances of aggravation probably do not differ dramatically. And nominal difference in statutes may fade in application. Second, we applied one knowledgeable source’s categorization of states as having more or less broad death penalty statutes.[49] We found no material difference in states’ death-obtaining behavior based on that categorization.[50]

B. Measuring Relief from Death Sentences

We use two measures of relief from death sentences. One is based on the outcomes of capital sentence appeals reported in our database of state court appellate opinions. The other is based on relief rates reported in the BJS capital punishment database.

1. The BJS Capital Punishment Database

The BJS tracks every person who is or has been under a sentence of death since 1973. The available data are current through 1995 and contain 6,228 observations.[51] To avoid the effects of early twists and turns in the post-Furman[52] death penalty era, we limit the sample to those defendants sentenced after 1984.[53] Thus, our BJS data cover defendants sentenced to death from 1985 to 1995. Defendants being considered for potential relief from death sentences have to be matched with death-obtaining rates from an appropriate time period. Death-obtaining rates used, as discussed above, cover the years 1985 to 1994. This is a reasonable period to use for potential death penalty cases because it substantially overlaps with the sentencing year period, 1985-1995. The one-year difference in ending years allows at least some 1994 and earlier murders to proceed to trial and lead to sentencing through 1995. We further limit the sample to the twenty-one states in our appellate opinion database. These selection criteria yield 3,046 observations in the BJS data.

The BJS data come from the prisons, which collect information under the National Prisoner Statistics Program.[54] The data include information on inmates whose death sentences were removed, including information about the reason for an inmate’s removal from death sentence classification. The BJS data include among the reasons for removal: whether a capital sentence was declared unconstitutional, whether a capital sentence was overturned, and whether both a conviction and capital sentence were overturned.[55] We define a variable, "relief," meaning relief from a death sentence, to equal one when any of these reasons for removal were present. We define relief to equal zero when a prisoner remains on death row and is executed, or dies from other causes, while on death row.

2. The Appellate Case Sample

Our appellate case sample consists of every direct appeal of a death penalty case available on Westlaw for the three-year period 1995 to 1997 for twenty-six states. The states are: Alabama, Arizona, Arkansas, California, Connecticut, Delaware, Florida, Georgia, Illinois, Indiana, Kentucky, Louisiana, Mississippi, Missouri, Nevada, New Hampshire, New Jersey, North Carolina, Ohio, Oklahoma, Pennsylvania, South Carolina, Tennessee, Texas, Utah, and Virginia. These states accounted for 3,112 of the 3,208 state prisoners (97%) under sentence of death as of December 31, 1996.[56] We do not include Connecticut, Delaware, New Hampshire, New Jersey, and Utah cases in our appellate case analysis because these states had too few death penalty cases to permit meaningful analysis. This leaves twenty-one states in the state appellate case sample.

We used search terms broad enough to identify capital cases in each state. With the help of research assistants, we recorded information about each case, including whether the case involved a resentencing.

To compute a state’s reversal rate in capital cases, we divide the number of appellate cases in which the defendant achieved some success by the number of appellate cases found. Success is not a self-defining term in assessing litigation.[57] It is possible to have a death sentence vacated on appeal, have the case remanded to a trial court, and have a death sentence imposed again. At the end of such a process, the defendant may be no better off than at the beginning, except for the delay, which is itself a nontrivial benefit to a prisoner under sentence of death. For defendants with more than one appeal in the database, we use the most recent appeal. Thus, for example, if a defendant’s conviction was affirmed by an intermediate appellate court and a state supreme court, only the latter case is retained. We do not otherwise trace each case’s subsequent history. We in effect take a snapshot of the process as of a substantial (three-year) period of time.

We define success (a reversal) to exist when the defendant obtains a ruling that precludes imposition of a death sentence unless further action is taken by some court. Therefore, reversals of convictions, remands for hearings on specific issues, vacation of death sentences, and remands for resentencing are defendant successes. Complications arise when both parties appealed or when the state appealed. We limit our analysis to cases in which the defendant’s appeal seeks relief from a death sentence. Thus, appeals by the state are not included.

The 1995 to 1997 appellate opinion data match the defendants’ appeals being considered for reversal with the time span covered by the Uniform Crime Reports 1994 to 1996 murder data, with a one year lag in the measure of murders. The one year lag in the measure of murders allowed the cases to become ripe for adjudication during the time period covered by the appellate data.

3. Differences Between the Measures of Relief from Death Sentences

Important differences exist between the BJS-based measure of obtaining relief from death sentences and the measure based on our reported appellate opinions. First, the reported opinions cover the years 1995 to 1997. The BJS data try to include all persons on death row, and any relief they obtained through 1995. Differences in the two measures of capital defendant success rates may result from events relating to a state’s capital defendant population for only one of the time periods. For example, a single Supreme Court case may have required reversal of many convictions within a state before 1995.[58] This, in turn, may have led to a burst of death penalty reversals before 1995, followed by a period of fewer reversals after 1995.

Second, our opinion data are limited to state courts. We are primarily interested in the effect of judicial selection methods across states and therefore do not include federal cases among opinions analyzed. Although the BJS data include relief from death sentences by both state and federal courts, they do not allow us to distinguish between relief granted by the two court systems.

Third, the BJS data include both direct appellate relief and post-conviction relief. The appellate data cover only direct appeals. We also have data on about 500 state post-conviction appeals. Including them would not change the results reported here. We exclude state post-conviction activity only because not all states issue opinions in such cases.[59]

We thus have measures of both trial level death-obtaining behavior in a state, and measures of defendants’ success in obtaining relief from death sentences—appellate and otherwise. These measures should be positively correlated. The appellate rate of reversing death penalty cases should be a function of how liberally the state sought and obtained death in murder cases. In a state that obtains death sentences relatively often—as measured by a high ratio of death penalty direct appeals to murders—there should be an increased chance of reversal. A high death-obtaining rate should mean that weaker, less death-worthy cases were brought. Reversals should thus come at a higher rate.

A caveat about the use of appellate opinions is in order: Opinions, both trial level and appellate, may not be representative of all underlying cases.[60] For example, published opinions in cases involving punitive damages awards have systematically higher awards than the mass of trial level punitive damages cases that do not result in published opinions.[61] But this concern is less of a problem in death penalty cases than in other classes of cases. Common sense, as well as some evidence, indicate that little selection of cases is at work in the decision to appeal a death sentence. A person condemned to death has little to lose by appealing, thus rates of appeal in death penalty habeas corpus proceedings are by far the highest of any area of law for which the Administrative Office of U.S. Courts maintains data.[62] The great majority of death penalty appeals wind up in state supreme courts,[63] which one expects to issue predominantly published opinions.

Table 1 summarizes the sources of data used in this study, as described in Parts III.A and III.B. It separates the sources by our two different studies of death penalty adjudication, the BJS data and the appellate opinion data.

Table 1. Data Sources for Numbers of Murders, Death Penalties, Death-Obtaining Rates, and Relief Rates

Study based on 

BJS data 

appellate opinion data 

Numbers of murders 

FBI data 1985-1994 

FBI data 1994-1996 

Number of death penalties 

BJS data 1985-1994 

BJS data 1994-1996 

Death-obtaining rates 

death penalties/murders 

death penalties/murders 

Relief rates 

BJS data 1985-1995 

appellate data
1995-1997
 

C. Descriptive Statistics

Table 2 reports descriptive statistics for the key variables in this study. Four pairs of columns should be distinguished: the number of murders in columns one and two, the number of death penalties in columns three and four, states’ death-obtaining rates in columns five and six, and rates of obtaining relief from death sentences in columns seven and eight. Column nine reports the number of direct death penalty appeals reported for each state from 1995 to 1997. Column ten reports whether the state has partisan judicial elections.

The first two columns show the number of murders in each state from 1994 through 1996, and from 1985 through 1994, respectively.[64] For the appellate opinion data covering 1995 to 1997, we use the shorter period of murders to compute a death-obtaining rate. For the BJS-based measure of the number of death penalties, the longer period is appropriate because the data on relief for capital prisoners extend from 1985 through 1995.

The third and fourth columns in Table 2 report our two measures of the number of death penalties. The BJS data in the third column are the number of death penalties imposed from 1985 through 1994. The data in the fourth column are the number of death penalties imposed from 1994 through 1996.[65]

The two death-obtaining rates reported in columns five and six are computed from the number-of-murder and number-of-death-penalty columns. The first death-obtaining rate column is the percent of murders from 1994 to 1996 that led to a death sentence in that period. The second rate column is the percent of murders from 1985 through 1994 that led to a death sentence from 1985 through 1994, according to BJS. These two measures of death-obtaining behavior are best viewed as indices facilitating interstate comparison rather than as measures of absolute activity levels. Thus, the relations among the states’ numbers in the death-obtaining rate columns are important, not their absolute levels. If one wants to relate a death-obtaining rate to real world events, the first row of Table 2 suggests, for example, that in Alabama 2.93% of the murders from 1985 through 1994 led to imposition of the death penalty. In Nevada, that rate was over 5%.

The death penalty relief columns show defendants’ rate of success in contesting death sentences. The first relief rate column is the percent of death sentences imposed from 1985 through 1995 for which BJS reports a defendant receiving relief, either through reversal of conviction, reduction of sentence, or both. The second relief rate column shows the percent of 1995 to 1997 death penalty appellate opinions in which the defendant received some relief.

The two death-obtaining rate columns in Table 2, columns five and six, show substantial interstate variation in death-obtaining behavior. For example, both columns show that Oklahoma, Nevada, North Carolina, Mississippi, Arizona, Alabama, and Florida have high death-obtaining rates. Both columns indicate that California has a low death-obtaining rate.

The penultimate column of Table 2 reports the number of direct capital case appeals from 1995 through 1997. These form the basis for computing the appellate reversal rates reported in column eight.

Anecdotal evidence supports the death-obtaining rates reported in columns five and six. The fairly high rate for Pennsylvania is consistent with the Philadelphia district attorney’s dedication to seeking the death penalty.[66] But Pennsylvania’s death-obtaining rate is kept from being extremely high by a much lower rate of death-seeking behavior by Pittsburgh’s district attorney.[67] Texas’ more moderate rate of death-obtaining

behavior is consistent with reported levels of death-seeking behavior by Dallas and Houston district attorneys.[68] California’s low rate of death-obtaining behavior is consistent with reported low death-seeking and death-obtaining rates in Los Angeles County.[69] Louisiana’s low rate of death-obtaining behavior is consistent with the reported lack of success of the New Orleans district attorney in obtaining death sentences.[70] Illinois’ moderate rate of death-obtaining behavior is consistent with relatively few capital sentences being imposed in Chicago.[71]

The relief columns in Table 2 show substantial interstate variation in relief rates. In the appellate opinion relief column, column eight, Arkansas, Florida, Louisiana, Mississippi, and Tennessee have the highest reversal rates. Virginia has the lowest. In the BJS-based relief column, the highest relief rates are found in Kentucky, North Carolina, and Mississippi. We thus find not only interstate variation in relief rates but some variation in relief rates across our two measures of relief.

IV. Explaining the Pattern of Relief from
Death Sentences

States have a broad range of death penalty reversal rates. The interstate differences in each of Table 2’s relief columns are statistically significant (p<.0001).[72] But the differences between the two relief columns, and the different contents of their underlying data, suggest exploring the two relief rate patterns separately.

Table 2’s appellate opinion relief column, column eight, is more appropriate to assess the effect of differences in judicial selection methods on interstate case outcomes than the BJS-based column seven, because only the opinion column is limited to decisions by state court judges. By contrast, the BJS-based relief rates include all possible modes of obtaining relief, including habeas corpus cases decided by federal judges.

A. The Absence of a Correlation Between Selection Method
and Reversals

In the appellate data, the hypothesis that partisan election of judges correlates with death penalty affirmance holds up for North Carolina, Pennsylvania, and Texas. All have partisan elections and low reversal rates. South Carolina and Virginia could also be viewed as fitting the pattern because their legislatures elect supreme court justices. But column eight also shows that Alabama, Mississippi, and Illinois, which have partisan elections, have high capital sentence reversal rates. Therefore, partisan elections, standing alone, cannot explain the pattern of appellate results. States without partisan elections also show varying reversal rates. Oklahoma and Florida have high reversal rates. Indiana has a low reversal rate. Overall, we did not find a statistically significant relation between judicial selection method and reversal rate. A test of the significance of their relation is significant at the .727 level.[73] If we count South Carolina and Virginia as partisan election states because their legislatures play a role in selecting judges, the significance level is .296.

Part of the reason why judicial selection methods may not explain the data well is that the classification of judicial selection methods is crude. For example, in the customary classification of state judicial selection methods, both New York and Texas are classified as partisan election states.[74] Yet, one hears much less complaining about the influence of campaign funding and contributions on justice with respect to New York than with respect to Texas.[75] And nominally nonpartisan judicial elections can take on the characteristics of partisan contests.[76] Indeed, if we take the liberty of reclassifying Ohio, which officially has nonpartisan judicial elections but in fact seems to have partisan elections,[77] a significant relation between partisan election status and reversal rates does emerge (p=.041).

Therefore, a statistically significant relationship between judicial selection methods and death penalty case outcomes depends, in part, on how strictly one adheres to conventional classification of judicial election methods. But the relation depends largely on how one state, Ohio, is classified and does not hold for the mass of states. Whatever the reason, the conventional wisdom about partisan judicial elections at best modestly explains the observed state-level pattern of appellate reversal rates in capital cases.

B. Accounting for Case Selection: Appellate Opinion
Reversal Rates

Prospects for explaining the pattern might improve if we account for case selection. Figure 1 shows the relation between the appellate reversal rate in capital cases (Table 2, column eight) and the rate at which states obtain the death penalty (Table 2, column five). We expect the relation to flow from the Figure’s lower left to its upper right. The expected relation between death-obtaining rates and reversal rates does not emerge.

Perhaps, however, a discernible pattern would emerge if we simultaneously accounted for judicial selection methods and death-obtaining rates. This requires multivariate analysis of the data. Multiple regression is a statistical technique that quantifies the influence of each of several factors (independent variables) on the phenomenon being studied (dependent variable). Multiple regression works to segregate the effects of various factors such as, for our purposes, case selection and judicial selection.[78] Since our cases have dichotomous outcomes—death sentences are either reversed or not—we use logistic regression.[79] We also account for the fact that the data are sampled by state.[80] For this study, the dependent variable is whether the case resulted in a reversal and the primary

Figure 1. Relation Between States’ Death Obtaining Rates and Death Penalty Reversal Rates[81]

The death-obtaining rate is an index allowing comparison of death-obtaining rates across states. It is based on the number of death sentences from 1994 to 1996 and the state’s murder rate from 1994 to 1996. The reversal rate is the proportion of 1995-1997 capital case opinions that reverse death sentences.

independent variables are the death-obtaining (case-selection) rate and a dummy variable that equals one if a state had partisan judicial elections.

Regression analysis allows us to control for another factor. Defendants appealing from the reimposition of previously imposed, but reversed, death sentences are likely, on average, to be weak candidates for reversal. These defendants have been previously sentenced to death. The subsequent resentencing to death, the subject of appeal in our data, could be taken as some evidence that these cases are death-worthy or that the arguments against death were not forcefully marshalled. Whatever the reason for the prior death sentence, it seems appropriate to account for whether an appeal involves a second death sentence. Simple statistics confirmed the possible need to control for this factor. Nine of fifty-seven resentencings (16%) led to reversals compared to 154 of 754 of cases (20%) not involving resentencings.[82] To control for the kind of appeal, we include a dummy variable equal to one for resentencing cases and zero otherwise.

Table 3 summarizes the independent variables in our model of appellate reversals. It includes a second partisan election variable, which treats South Carolina and Virginia as states with partisan judicial elections. Table 3’s next-to-last column shows that no variable is significantly correlated with appellate reversal rates of death sentences (Table 2’s column eight).

Table 3. Descriptive Statistics, Death Sentences in States with Nine or More Direct Capital Case Appeals from 1995-1997

 

mean 

range 

significance*  

n 

Reversal rate 

.202 

0-1 

- 

812 

State’s death-obtaining rate, 1994-96 

.028 

.008-.061 

.444 

813 

Partisan election state 

.464 

0-1 

.727 

813 

Partisan election state, including SC & VA 

.512 

0-1 

.296 

813 

Resentencing 

.070 

0-1 

.494 

812 

*In relation to reversal rate

Regression analysis confirms the absence of such significant correlations, but it does show that the correlations are in the expected directions.[83] Table 4 reports the regression results for a model using the variables in Table 3. The negative sign on the partisan election dummy variable indicates that appellate win rates correlate negatively with judicial selection by partisan elections. But this correlation is not statistically significant. Nor does the death obtaining rate appreciably help in explaining reversals of death sentences. These results provide little support for the influence of case selection or the influence of judicial selection methods on death penalty case appellate outcomes. We do note, however, that reclassifying Ohio as having partisan elections would yield a larger and more significant (p=.044) coefficient for the partisan election variable.

Table 4. Logistic Regression Analysis of Appellate Reversal Rates in Capital Cases, 1995-1997

dependent variable = reversal of death sentence 

Death-obtaining rate, 1994-1996 

1.678

(.784) 

Partisan election state (including SC and VA) 

-.166

(.329) 

Resentencing 

-.308

(.411) 

Constant 

-1.324

(.000) 

F-test significance of model 

.616 

Number of cases 

811 

Number of states 

21 

Significance levels in parentheses

C. Accounting for Case Selection: BJS Data

Relief rates based on the BJS data can be compared to the appellate opinion-based results. As in the case of the appellate data, we control for each state’s death-obtaining rate. We also use the partisan election dummy variable to control for judicial selection method.

The BJS data allow us to control for other characteristics that might relate to the likelihood of obtaining relief. We control for the defendant’s race by using dummy variables for black and hispanic status. We control for the defendant’s marital status by dummy variables for married and for whether a defendant was divorced or separated. We also control for the defendant’s age and education.[84] We control for the defendants’ legal status at the time of the capital offense through two variables. A dummy variable, "charges pending," has a value of one if the defendant was not under sentence at the time of the capital offense but charges were pending. A second dummy variable, "under sentence," has a value of one if the defendant was on probation, on parole, had escaped, or was imprisoned at the time of the capital offense. Finally, we control for whether the defendant had been convicted of a prior murder through the dummy variable, "prior murder."

For North Carolina, it is appropriate to account for one case that led to an unusual number of grants of relief during the period covered by the BJS data. In McKoy v. North Carolina,[85] the Supreme Court held that North Carolina’s requirement that jurors agree unanimously on mitigating factors impermissibly limited jurors’ consideration of mitigating evidence. A sharp decrease in relief from North Carolina death sentences occurred for defendants sentenced after McKoy.[86] We therefore include a dummy variable equal to one for North Carolina defendants sentenced to death before March 5, 1990, the date McKoy was decided.[87]

Table 5 presents descriptive statistics of the BJS variables used, together with the significance of their relation to whether relief was obtained. It includes death sentences imposed after 1984 in the twenty-one states covered by the appellate data. Strong correlations exist between whether a defendant was under sentence at the time of a capital offense and whether the defendant obtains relief from a capital sentence. Education correlates negatively with relief: The more educated a defendant, the less likely relief is obtained. Education may be functioning as a proxy for the absence of mental illness or retardation. Furthermore, women are significantly more likely than men to obtain relief from death row.

The primary variables of interest are the states’ death-obtaining rate and the states’ partisan election status. Unlike Table 3, Table 5 shows a significant correlation between death-obtaining rates and obtaining relief from death sentences. Like Table 3, Table 5 shows no significant relation between partisan election of judges and obtaining relief from death sentences. This result is independent of how one characterizes Ohio.

Table 5. Descriptive Statistics, Death Sentences in 21 States, 1985-1995

 

mean 

range 

significance*  

number 

Relief granted to defendant 

0.20 

0-1 

 

3,046 

Black defendant 

0.42 

0-1 

0.406 

3,046 

Hispanic defendant 

0.08 

0-1 

0.168 

3,046 

Female defendant 

0.02 

0-1 

0.016 

3,046 

Married defendant 

0.22 

0-1 

0.080 

3,046 

Divorced or separated defendant 

0.20 

0-1 

0.819 

3,046 

Age of defendant 

30.97 

16-77 

0.283 

3,046 

Education of defendant 

4.80 

1-11 

0.000 

2,637 

Charges pending at time of crime 

0.06 

0-1 

0.643 

3,046 

Under sentence at time of crime 

0.30 

0-1 

0.000 

3,046 

Prior murder conviction 

0.06 

0-1 

0.272 

3,046 

State’s death-obtaining rate,
1985-94
 

2.58 

0.85-5.24 

0.000 

3,046 

Year of sentencing 

1990.03 

1985-95 

0.000 

3,046 

Partisan election state 

0.43 

0-1 

0.748 

3,046 

McKoy dummy variable
(for N.C.)
 

0.02 

0-1 

0.000 

3,046 

*In relation to rate of relief from death sentence

Importantly, Table 5 shows that a defendant’s year of sentencing is significantly correlated with the grant of relief. In particular, the earlier the sentencing year, the more likely the defendant is to obtain relief from a death sentence. This raises the problem of what is termed data censoring, which we explore in Table 6.

Table 6 reports, by year, the relief rates based on the BJS data. Each row shows the proportion of cases in which relief was granted for a year of sentencing. For example, 42% of the defendants sentenced to death in 1985 obtained, at some point, some relief from their death sentence or conviction. Table 6 shows a decline in relief over time, with the relief rate reaching zero in 1995, the most recent sentencing year in the data. The data are censored in the sense that it apparently takes a substantial period of time to obtain relief from death sentences. Those who have been on death row only briefly are much less likely to have obtained relief from their death sentences. It may also be that obtaining relief in recent years has become more difficult. But the delay between sentence and relief would lead to a declining relief rate being observed even if there were no decrease in the rate that relief was in fact being granted. Enough time has not elapsed for recently sentenced defendants to have obtained relief.

Table 6. Relief Rates from Death by Year of Sentencing,
BJS Data, 1985-1995

Year of sentence 

Relief rate 

Number of death sentences 

1985 

.423 

279 

1986 

.311 

293 

1987 

.319 

273 

1988 

.274 

292 

1989 

.287 

251 

1990 

.248 

254 

1991 

.167 

263 

1992 

.124 

275 

1993 

.034 

268 

1994 

.010 

303 

1995 

.000 

298 

The presence of censoring suggests the propriety of survival time models for the BJS data.[88] Such models account for the fact that the event of interest, in this case relief from a death sentence, may not have occurred simply because sufficient time had not elapsed.[89] These models also allow for different times of entry into the sample, which we have in the form of different years of sentencing,[90] and exit from and reentry into the sample, which occurs when defendants obtain relief and are later resentenced to death. These features make a survival time proportional hazard model preferable to a simple logistic regression model, which ignores the time to an event (relief from a death sentence in our case) and censoring.[91] Table 7 reports Cox proportional hazard models.[92] Logistic regression models do not yield results that differ materially from those reported here.

Table 7 reports two models. The second model excludes the defendant’s education variable, which is missing for almost 400 defendants. As is common for proportional hazard models, the coefficients are reported as hazard ratios.[93] A hazard ratio of one means that a variable has no effect. A hazard ratio of ten, for example, means that a unit increase in the dependent variable corresponds to a tenfold increase in the likelihood of relief. A hazard ratio of .1 implies that a unit increase in the dependent variable corresponds to a tenfold reduction in the likelihood of relief.

Table 7 confirms the simple statistics presented in Table 5. Of particular interest is the significance of the states’ death-obtaining rate. A strong, significant correlation exists between death-obtaining behavior and relief from death sentences. A 1% increase in a state’s death-obtaining rate corresponds to about a 20% increase in the likelihood of relief being granted.[94]

Table 7. Cox Proportional Hazard Models of Relief from
Death Sentences, 1985-1995

Time variable is months to relief.
Censoring occurs for cases in which no relief has been granted.
 

Black defendant 

1.025

(.795) 

1.082

(.377) 

 

 

 

Hispanic defendant 

.775

(.177) 

.910

(.588) 

 

 

 

Female defendant 

1.802

(.014) 

1.843

(.009) 

 

 

 

Married defendant 

1.034

(.758) 

1.022

(.827) 

 

 

 

Divorced or separated defendant 

.818

(.112) 

.834

(.130) 

 

 

 

Defendant’s age 

1.009

(.082) 

1.009

(.093) 

 

 

 

Defendant’s education 

.944

(.009) 

-

- 

 

 

 

Charges pending at time of capital crime 

.926

(.687) 

1.024

(.891) 

 

 

 

Defendant under sentence at time of
capital crime
 

.735

(.003) 

.758

(.006) 

 

 

 

Defendant committed prior murder 

.634

(.029) 

.770

(.167) 

 

 

 

State’s death-obtaining rate, 1985-1994 

1.180

(.000) 

1.195

(.000) 

 

 

 

Year of sentencing 

.933

(.000) 

.924

(.000) 

 

 

 

State has partisan judicial elections 

.901

(.306) 

.913

(.336) 

 

 

 

McKoy dummy variable for N.C. cases 

4.463

(.000) 

4.549

(.000) 

 

 

 

Prior death sentence 

.841

(.485) 

.793

(.348) 

 

 

 

Number of defendants 

2,551 

2,956 

Number obtaining relief 

521 

584 

Log-likelihood 

-3632.140 

-4156.275 

Chi-squared probability 

.0000 

.0000 

Significance levels are in parentheses.

The other variable of primary interest, whether a state has partisan judicial elections, follows the pattern in the appellate opinion results reported in Table 4. The variable’s effect is small and insignificant. This is true whether or not we treat South Carolina, Virginia, or Ohio as partisan election states, although the models reported in Table 7 do not treat them as partisan election states.

Other results are also worth noting. Given equal crimes, available data suggest that women tend to receive shorter sentences than men.[95] Consistent with these findings, Table 7 contains evidence that women sentenced to death are more likely to obtain relief from their death sentences than are men. Furthermore, defendants who commit capital crimes while under sentence for other crimes and defendants who had prior murder convictions are less likely to obtain relief from their capital sentences.

The explanation of why death-obtaining rates help explain case outcomes in the BJS data but not in the state court appellate data may rest in judicial selection method, but not in interstate differences in judicial selection method. The BJS data include grants of relief by federal courts, whereas the state court appellate data do not. It may be that the independence of federal judges led them to be more likely to grant relief in marginal death sentence cases than state judges. This could lead to the observed correlation between grants of relief in the BJS data and the rates of death-obtaining behavior. That this effect does not emerge in the state court appellate data may show that state judges know they face traditional elections or retention elections at some point in their career. More than a decade ago California’s experience proved that the selection method did not provide insulation from politicization of the death penalty. Only the life tenure and independence of federal judges may provide the luxury of assessing death penalty cases based on their death-worthiness.

But other explanations are also possible. As discussed above, possibly important differences exist between the BJS data and the state court appellate data. They cover different time periods and only the BJS data account fully for post-conviction relief. At this point we can only suggest an explanation based on the federal judiciary’s greater independence. Expanding the appellate opinion database to encompass more years may yield more conclusive evidence. For example, if we had ten years of data from state courts, as we do for federal and state courts combined in the BJS data, the absence of a correlation between cases’ death-worthiness and outcome would be stronger evidence of differences between state and federal court processing of capital cases.

D. Another Measure of States’ Tendencies to Impose
the Death Penalty

As a last effort to detect judicial selection effects, we combine Table 2’s death-obtaining rates and relief rates to construct a new measure of states’ tendencies to impose the death penalty and retain it on appeal. This measure is based on the percent of murders that led to an affirmed death penalty. We then test whether judicial selection methods correlated with this more sophisticated measure of death-obtaining behavior.

To illustrate, the first entry in Table 2, column five, shows that, for 1994 to 1996, 3.97% of Alabama’s murders led to imposition of the death penalty. Table 2, column eight, shows that 22.4% of death penalties in Alabama were reversed. Conversely, 77.6% of death penalties were affirmed. So in a state in which 3.97% of murders led to death penalties, 77.6% of the death penalties were affirmed. We multiply the 77.6 affirmance rate by the 3.97 death-obtaining rate to yield an index of the percent of murders that led to an affirmed death penalty. In Alabama’s case, the multiplication yielded 3.08%. In other words, 3.08 of each 100 murders led to an affirmed death penalty. A similar calculation for each state allows interstate comparison of a death-obtaining rate that accounts for appellate review of death sentences. Table 8 presents the results.

Table 8 confirms that states with large death rows do not necessarily obtain and affirm death penalties at high rates. Although Nevada and Oklahoma obtain and affirm death penalties at higher rates than other states, their death rows were are not among the largest. California’s death row is one of the largest but its rate of obtaining and affirming death penalties is one of the lowest, and is the lowest of any major death penalty state. Texas is not extreme in its rate compared to other states. Mississippi and Florida combine high trial-level death-obtaining rates with high reversal rates to yield net death-obtaining rates that are moderate compared to other states.

Table 8. Number of Affirmed Death Penalty Cases
Per 100 Murders, by State, 1995-1997

State 

Affirmed 

 

State 

Affirmed 

Nevada 

4.75 

 

Arizona 

1.54 

Oklahoma 

4.48 

 

Arkansas 

1.38 

North Carolina 

3.85 

 

Virginia 

1.11 

Alabama 

3.08 

 

Illinois 

0.99 

Ohio 

2.67 

 

Georgia 

0.91 

Pennsylvania 

2.27 

 

Louisiana 

0.86 

South Carolina 

2.09 

 

California 

0.84 

Texas 

2.08 

 

Kentucky 

0.76 

Mississippi 

2.01 

 

Indiana 

0.75 

Florida 

1.89 

 

Tennessee 

0.57 

Missouri 

1.76 

     

Does this new "bottom-line" measure of death-obtaining behavior correlate with judicial selection method? Again, the answer is no. We find no correlation between partisan elections and this measure of death-obtaining behavior. The absence of correlation does not depend on how South Carolina and Virginia, the two states with legislative participation in judicial selection, are classified or on how Ohio is classified.

A possible extension of this analysis would be to account for execution rates. Affirmance of death penalties and the absence of relief do not necessarily lead to executions. Some states have large death rows but no or few executions.[96] The defendants are in a kind of limbo due to executive or judicial obstacles to execution. But this limbo was likely to be unstable overtime. We forego an execution rate analysis because of this volatility.

E. State-Specific Evidence of Political Pressure

Neither the appellate opinion data, the BJS data, nor our new measure of death-obtaining behavior show a robust correlation between judicial selection method and treatment of death sentences. Nevertheless, California and a few other states provide state-specific stories worth exploring. These stories support the existence of a relation between politics and death penalty case outcomes, even if that relation cannot be explained by focusing on judicial selection method.

California combines an extremely low pre-1995 relief rate with a merely low appellate reversal rate from 1995 to 1997. As shown in Table 2 (columns seven and eight), California’s relief rate from capital sentences was less than 1% through 1995 and increased to about 10% from 1995 to 1997 in the appellate opinion data. This is consistent with a substantial period of extreme reluctance to overturn death sentences after Chief Justice Bird’s removal in the mid-1980s, with a softening of that reluctance in recent years. In fact, only California has a pre-1995 relief rate that statistically significantly differs from its 1995 to 1997 appellate reversal rate.[97] But even that softening left California with a low reversal rate. This contrasts with a more liberal reversal rate prior to the period we studied.[98]

Justice White’s removal from office in Tennessee during the period covered by our appellate opinion data allows us to test the hypothesis that her removal affected that state’s reversal rate. For the first two years of the appellate data in this study, 1995 and 1996 combined, reversals occurred in eight of thirteen Tennessee cases (61.5%). In 1997, the first year after Justice White’s removal, reversals occurred in one of fourteen Tennessee cases (7.1%). This difference is highly statistically significant (p=.004).[99]

To assess the effect of the politicization of judicial selection in South Carolina, we added South Carolina appellate capital cases from 1981 to 1994 to the sample. Thus, for South Carolina, we have data from 1981 to 1997. As noted above, the state’s attorney general, elected in 1994, ran on strong pro-death penalty views and a politicization of the death penalty. The attorney general’s election coincided with a noticeable change in the success rates of South Carolina capital cases on direct appeal to the state supreme court. From 1981 through 1993, twenty-eight of sixty-two death penalties (45.2%) were reversed. From 1994 through 1997, five of twenty-five death penalties (20%) were reversed. This difference is statistically significant at the .031 level.[100] Even the 20% recent reversal rate may have been artificially high because of reversals required by United States Supreme Court decisions in South Carolina cases.[101]

In contrast to developments in California, Tennessee, and South Carolina, the politicization of the death penalty in the 1994 Texas Supreme Court elections seems not to have significantly changed that state’s pattern of review in capital cases. Texas capital prisoners had little success in obtaining relief before 1995 and little success in obtaining appellate reversals from 1995 to 1997. But Texas’ much publicized taste for death sentences is partly mythical. Texas’ prosecutors obtain death penalties at lower rates than at least half the states with substantial numbers of death penalties. Similarly, Mississippi’s rates of relief were not discernibly affected by the 1992 campaign against Justice Robertson. Although Mississippi continued to grant relief to capital defendants at high rates, this may have been a function of Mississippi’s high death-obtaining rate.[102]

Taken together, the state-specific stories suggest that judicial selection methods can neither insulate judges from political pressure nor always lead to quantifiable manifestations of such pressure. California lacked partisan judicial elections, yet justices were voted out of office on the death penalty issue. California continued to have a fairly low reversal rate in capital cases. Tennessee and South Carolina also supported the belief that political pressure affects judges in capital cases. But judicial behavior in capital cases had been politicized in Mississippi and Texas without noticeable changes in their treatment of capital cases.

Whether a state is classified as having partisan judicial selection methods is not a useful predictor of capital case outcomes. Specific state political campaigns raising the death penalty issue are more helpful in explaining case outcomes, but even they do not always assure measurable change.

V. Conclusion

Differences in states’ judicial selection methods explain little about interstate differences in capital case outcomes. But politicization of the death penalty issue has affected state court behavior. This seems to have occurred in California, Tennessee, and South Carolina.

We also theorized that judicial behavior in capital cases can only be understood if one accounts for the case selection process.[103] The BJS data, but not our appellate opinion data, show a strong correlation between states’ death-obtaining behavior and reversal rates. In the BJS data, which include federal habeas corpus capital cases, states that obtain death penalties at a high rate tend to have them overturned at a high rate. This finding is consistent with the view that high death-obtaining rates correspond to death penalties being imposed in less death-worthy cases. Courts understandably overturn more capital sentences in such cases.