next | previous | table of contents

C. Results of Analyses 3-6 of Single Stages of Review: More Evidence of the Effects of Race, Politics, Frequent Use of the Death Penalty and Weak Law Enforcement

We next consider analyses that alter those just discussed in two important ways:

  • They examine only one review stage at a time, not all three stages at once.

  • They examine reversal rates among only actually reviewed verdicts imposed in the relevant state and year (or, in the case of state post-conviction Analysis 5, among verdicts available for review), rather than reversal rates among all verdicts imposed in the relevant state and year.

Both changes reduce the number of death verdicts available for study, and thus the number of states and years with reversal rates to study. This accounts for the drop in the number of observed reversal rates from 519 in Analyses 1 and 2 above, to 453 in Analyses 3 and 4, and to 354 in Analysis 5 and 161 in Analysis 6. Because statistical significance is affected by the number of observations, these declines in the number of observations make it more difficult to obtain useable results.

Two other, more favorable effects of these changes explain why we use them to check the reliability of main Analysis 1 and Analysis 2. First, by altering choices made in designing the main analysis, Analyses 3-6 test whether those choices, not actual relationships in the data, account for the main analysis' results. For example, by studying the results of all three stages at once, Analyses 1 and 2 maximize the amount of data being analyzed but might obscure conditions operating only at a single stage. Studying each stage by itself thus checks the completeness and power of the explanations Analyses 1 and 2 identify. Analyses 1 and 2 also maximize the data being analyzed by using a deflated measure of reversal rates—reversals as a proportion of imposed, rather than only reviewed, death verdicts.400 Analyses 3, 4 and 6 alter this feature by analyzing factors related to true error rates, i.e., reversals as a proportion of verdicts actually reviewed.

Second and relatedly, some of our stage-specific analyses more reliably assess the effect of time at the state direct appeal stage. As noted, Analyses 1 and 2 explain differences in reversal rates among all imposed verdicts, whether or not they were finally reviewed. This leads reversal rates in later periods to be lower than would otherwise be the case, not because later verdicts were less seriously flawed, but because there was less time to review them, so that fewer of those with flaws were reversed by the study's end date.401

Analyses 3 and 4 avoid this bias by studying only verdicts that were actually reviewed on direct appeal. Federal habeas Analysis 6 avoids the same time bias in the same way, but is susceptible to a separate bias against a full accounting of flawed verdicts imposed later in the study period. As we demonstrate above, federal habeas reversals take longer to occur than affirmances. See Figure 10, p. 93 above. As a result, a disproportionately high number of flawed (as opposed to unflawed) verdicts that were imposed in the study period were not fully reviewed and reversed on federal habeas by the end of that period.402 State post-conviction Analysis 5 is subject to the same time bias as our Analyses 1 and 2 of the three review stages combined, because we could only measure reversal rates at that stage against verdicts available for review, even if they were not actually reviewed.403

1. Direct appeal Analyses 3 and 4.

a. Tests of the analyses' ability to generate reliable results.

The results of the diagnostic tests in state direct appeal Analysis 3 (a binomial regression analysis) and state direct appeal Analysis 4 (a Poisson regression) show that:

  • Even after accounting for the effect of state, year and the passage of time, there was a significant amount of variation in capital reversal rates from state to state and year to year, warranting consideration of more specific factors that might explain the differences.404

In addition, all five best analyses of more specific factors in Analyses 3 and 4C3A, 3B and 4A-4C:

  • did a significantly better job of fitting the predicted reversal rates to the actual ones than was true of the baseline analyses of only state, year and time trend;405

  • left substantially less unexplained variance than the baseline analyses;406

  • identified a series of specific factors that are significantly related to reversal rates; and

  • fit the data better and left less unexplained variance than other sets of significant factors.

The significant explanatory factors identified by Analyses 3 and 4 also:

  • largely overlap each other, across two different kinds of analyses (binomial logistic vs. Poisson logarithmic regression analyses), and

  • generate nearly the same set of significant explanatory factors as Analyses 1 and 2, notwithstanding the different stages of review being studied (all three stages in Analyses 1 and 2; only state direct appeal in Analyses 3 and 4) and the different definition of the reversal rates being compared (rates of imposed verdicts that were reversed, in Analyses 1 and 2; rates of reviewed verdicts that were reversed, in Analyses 3 and 4).

These tests suggest the reliability of Analyses 3 and 4, and tend to confirm the results of main Analysis 1 and Analysis 2.

b. Factors significantly related to higher capital reversal rates.

We discuss explanatory factors in the same order here as in discussing Analyses 1 and 2 above.

i. Lower court backlogs. We included capital and general court backlogs in Analyses 1 and 2 to control for the effect of unreviewed death verdicts on reversal rates measured as a ratio of imposed (even if not reviewed) death verdicts. Because Analyses 3 and 4 examine only reviewed verdicts, we expected court backlogs to be less important in these analyses, and they were. General (i.e., non-capital) backlogs had no significant effect at all. Capital backlogs are still significantly and negatively related to reversal rates,407 but effect size is somewhat smaller when only actual direct appeal decisions are considered.408 The continuing significance of capital backlogs suggest that large amounts of death verdicts awaiting review limit either the capacity of overwhelmed appellate courts to find and reverse flawed verdicts, or their willingness to do so in the face of public unhappiness at how slowly capital cases move through the courts.

ii. Earlier death verdicts. For reasons given above, Analyses 3 and 4 are our most reliable analyses of the effect of the passage of time on capital-error rates, after accounting for other factors.409 Their results in regard to this factor are important, therefore—and interesting. In the baseline analysis of only state, year and time trend, the passage of time is significantly, and negatively, related to direct appeal reversal rates, suggesting that later death verdicts are less likely to be reversed than early verdicts. But, when other, more specific factors are considered—appreciably enhancing the analyses' explanatory power, given (1) significantly improved fit with the actual data, (2) a sizeable drop in the amount of unexplained variance and (3) an expanded range of significant explanations for reversible error—the passage of time turns out to have a positive relationship to error rates that is far more powerful than the opposite relationship that initially appeared in the baseline inquiry.410 Accounting for other significant factors associated with reversal rates, death verdicts imposed later in time were significantly more likely to be reversed on direct appeal due to serious error than verdicts imposed earlier in time. The effect size for this factor is as large as that of any other factor in any of our analyses: If all other explanatory factors had remained constant at their averages, Analysis 3A predicts a reversal rate for death verdicts imposed in the early years of the study of about 9%, rising to 80% for verdicts imposed at the end of the study period. See Figure 32A (binomial Analysis 3A), p. 202 below. See also Figure 32B (Poisson Analysis 4A), p. 202 below, predicting a similar 8- to 9-fold increase. Using the numerical effect-size estimates, this translates into a prediction by Analysis 4A that, apart from the effect of the other factors in the analysis, error rates would have increased about 9% per year over the 23-year period.

Of course, other factors did not stay constant at their averages, and as Figures 2C and 2D, pp. 57-58 above, reveal, overall reversal rates in fact fluctuated considerably during the first part of the study period, then remained fairly constant during the latter half of the study period. Under these circumstances, the result discussed here indicates that:

  • changes over time in the other significant factors identified by Analyses 3 and 4 predict a drop in reversal rates over time that did not occur;

  • other time-sensitive factors—registered in Analyses 3 and 4 by the general time-trend factor— are associated with large increases over time in the amount of reversible error found on direct appeal; and

  • the downward influence of the specific factors identified by Analyses 3 and 4 and the upward influence of the time-sensitive factors registered by the general time-trend combined to produce the flat pattern of rates over the latter half of the study period.

From a policy perspective this suggests that, even if reforms based on the other factors identified by Analyses 3 and 4 had a downward effect on reversal rates, the factors registered in our analyses by time trend might nonetheless continue to have the opposite effect, with the overall result that reversal rates might abate less than one would hope, or not at all.

iii. Larger African-American population. In all but one of the groups of factors in Analyses 3 and 4, the proportion of states' population made up of African-Americans was significantly related to capital error rates.411 In one analysis, the factor was just above the .05 level.412 As in Analyses 1 and 2, the larger the relative size of the black community, the higher the rate of serious capital error found on state direct appeal. Analysis 3A predicts close to a doubling of reversal rates across the spectrum of African-American populations as a proportion of the general population in states and years in our study, other factors held constant. See Figures 33A and 33B, p. 203 below.

iv. Relatively high homicide threat to whites relative to blacks. In Analyses 3 and 4, as in Analyses 1 and 2, the higher the risk of homicide to a state's white community relative to its black community, the higher the probability that the state's death verdicts will be reversed due to serious error. The relationship is significant in Analyses 3A, 4A and 4B, and barely above significance in 3B and 4C.413 Holding other factors constant, Analyses 3A and 4A predict that reversal rates will more than double as the homicide threat to whites relative to the threat to blacks rises from its lowest to highest levels among states in the study. See Figures 34A and 34B, p. 204 below.

v. The interaction of a large African-American population and a relatively high threat of homicide to whites as compared to blacks. We tested this interaction effect in Analyses 3A, 4A and 4B. In all three analyses, the direction of the relationship was the same as in Analyses 1 and 2. States with large black communities and high homicide risks to their white communities relative to their black communities tend to have especially high reversal rates, even after accounting for the independent effect of the two component factors. Significance levels are unimpressive, however, though high enough to be of some interest.414

vi. Heavy use of the death penalty. In all five sets of factors in Analyses 3 and 4, states with higher death-sentencing rates had higher rates of flawed death verdicts. The relationship was highly significant in all five analyses. Effect size is large. In Analysis 3A, holding other factors constant, expected direct appeal reversal rates are about 15% where the death penalty is used most sparingly, but over 65% where it is used most often. See Figure 35A, p. 205 below. Analysis 4A predicts close to a 4-fold increase in reversal rates across that same spectrum. See Figure 35B, p. 205 below. The largest expected changes in reversal rates are associated with fairly modest changes in death-sentencing rates on either side of the average rate of 25 death verdicts per 1000 homicides. Holding other factors at their average, Analysis 3A (Figure 35A, p.205 below) predicts that:

  • States imposing 5 death verdicts per 1000 homicides will have a direct appeal reversal rate of about 27%.

  • States imposing the average number of death verdicts per 1000 homicides (25) will have a direct appeal reversal rate of over 40%.

  • And states sentencing 50 people to die for every 1000 verdicts—twice the national average, but only a quarter of the top death-sentencing rate—will have a reversal rate of over 50%.

vii. Fewer serious criminals apprehended and imprisoned. Throughout Analyses 3 and 4, as in Analyses 1 and 2, death verdicts imposed by states that apprehend, convict and imprison fewer criminals per serious crimes are more likely to be overturned due to serious, reversible error than death verdicts imposedin states with higher rates of apprehension, conviction and incarceration. The relationship is highly significant in all five analyses, and effect size is quite large. Where the rate of apprehension, conviction and incarceration of serious criminals is at the highest level among states and years in our study, Analysis 3A predicts direct appeal reversal rates of only about 12%, other factors constant. But where the apprehension-conviction-incarceration rate is at its lowest level for states and years in the study, the predicted rate of serious capital error found on direct appeal is over 80%. Analysis 4A predicts about a 5-fold increase in reversal rates across that spectrum. See Figures 36A and 36B, p. 206 below.

viii. More political pressure on state judges. Analyses 3 and 4 are limited to rates of reversal by mainly elected state high court judges reviewing verdicts imposed at trials supervised by other, mainly elected state judges. If, as was found by Analyses 1 and 2Cwhich examine decisions of unelected federal judges as well as elected state judges—increases in the amount of political pressure on state judges are related to increases in the frequency of serious capital error, one would expect this relationship to be somewhat weaker when only the decisions of elected state appellate judges are studied. Because the political pressure we measure—caused by judicial selection techniques—is similar for all state judges, appellate as well as trial judges, the same pressures that drive trial-level judges to make errors at particular trials might be expected to discourage appellate judges from reversing the resulting verdicts. Still, the political pressures a local community concentrates on local judges in the immediate wake of homicides probably weigh less heavily on appellate judges who review cases years after homicides occur, live elsewhere than the community harmed by the offense, and have constituents from many places besides that community. If political pressures are related to high rates of trial error, therefore, the patterns of direct appeal reversals in states where those pressures are stronger and weaker should probably provide some evidence of that relationship.

Subject to these considerations, the results of Analyses 3 and 4 are consistent with those of Analyses 1 and 2: Higher levels of political pressure on judges are associated with higher levels of serious capital error identified at the direct appeal stage. But significance levels are less impressive—.06 for Analyses 3A, 4A and 4B, and over .1 for Analyses 3B and 4C.415 Effect size is moderate, with reversal rates expected to double when the political pressure on judges rises from the lowest to the highest levels among states in our study and other factors are held constant. See Figures 37A (Analysis 3A) and 37B (Analysis 4A), p. 207 below.416

ix. Greater population size and density. At the direct appeal stage studied in Analyses 3 and 4, as at the three review stages combined in main Analysis 1 and Analysis 2, the size and density of a state's population was significantly associated with the likelihood of reversible error.417 Verdicts from highly and more densely populated states were more likely to be reversed than those from more sparsely populated states. Effect size is about the same as in Analyses 1 and 2. Compare Figures 38A and 38B, p. 208 below, to Figures 30A-30D, p. 190 above. Analysis 3A (Figure 38A) predicts a rise in direct appeal reversal rates from about 16% to nearly 60% as population structure goes from the least to the most populous among states in our study, other things equal.

x. A new explanatory factor: low spending on state courts. One explanatory factor with a highly significant relationship to error rates in our direct appeal Analyses 3 and 4 was not significant in Analyses 1 and 2, covering all three review stages: States' per capita direct expenditures on their court systems.418 As expected, death verdicts imposed by states that spend less on their courts per capita are more likely to be reversed on direct appeal due to serious error than verdicts imposed by states that spend more on their courts.419 Effect size is large, especially among states spending less than the average of about $1.75 per resident.420 Holding other factors constant, Analysis 3A predicts a rise in direct appeal capital reversal rates from 25% for jurisdictions with per capita court expenditures at the average for states in our study, to 74% for jurisdictions with per capita expenditures at the lowest extreme among states in our study. See Figure 39A (Analysis 3A), p. 209 below. Although all states support a high court, they probably vary substantially in the amounts they spend on local courts. Evidently, states that spend the least on their courts place an especially heavy quality-control burden on their high courts, which is reflected in higher capital reversal rates on direct appeal.

Figure 31 A-B: Effect Size: Backlog of Capital Verdicts (Analyses 3A, 4A)

Figure 32 A-B: Effect Size: Year of Death Verdict (Analyses 3A, 4A)

Figure 33 A-B: Effect Size: Proportion of African-Americans in State Population (Analyses 3A, 4A)

Figure 34 A-B: Effect Size: Risk of Homicide to Whites Versus Blacks (Analyses 3A, 4A)

Figure 35 A-B: Effect Size: State Death Sentencing Rate (Analyses 3A, 4A)

Figure 36 A-B: Effect Size: Rate of Arrest, Conviction and Imprisonment Per Serious Crime (Analyses 3A, 4A)

Figure 37 A-B: Effect Size: method of Selecting State Judges (Analyses 3A, 4A)

Figure 38 A-B: Effect Size: Population Structure (Analyses 3A, 4A)

Figure 39 A-B: Effect Size: Per Capital Spending on State Courts (Analyses 3A, 4A)

c. Summary: consistent results across Analyses 1-4.

Changing the condition under study from the rate of imposed death verdicts that are reversed to the rate of only reviewed verdicts that are reversed, and focusing only on a single, state-court stage of review rather than on the combined results at all three review stages, did not change the results. Neither did a 13% decrease in the number of states and years under study from 519 to 453. Instead, the results of Analyses 3 and 4 are strikingly similar, not only to each other, but also to the results of Analyses 1 and 2. All four analyses indicate that:

  • Aggressive use of the death penalty is strongly associated with high capital error rates.

  • Factors that can heighten the fear of serious crime among people with influence over public officials—encouraging officials to use the death penalty even when the evidence is weak—also are associated with high capital error rates. Those factors include:

    • the size of the community's African-American population;

    • the extent to which the homicide risk to members of the white community approaches or surpasses the risk to members of the black community; and

    • the weakness of the state's non-capital response to crime, as measured by the rate at which serious criminals are arrested, convicted and incarcerated.

  • Error rates are higher in states that select capital trial judges in ways that increase the pressure on them to conform their rulings to popular sentiment.

  • Large pile-ups of cases awaiting review are associated with lower review and reversal rates.

  • The more reliable the measure of the effect of the passage of time on the quality of capital verdicts, the stronger the evidence that, after controlling for other factors, later verdicts are more likely to be reversed due to serious capital error than earlier verdicts.

  • Error rates for death verdicts from populous and urbanized states are higher than those from sparsely populated states.

  • High error rates are associated with evidence that state courts are overburdened— although at the three review stages combined (Analyses 1 and 2), the indicator that the courts are overburdened is the combination of heavy capital and general caseloads, and at the direct appeal stage ( Analyses 3 and 4), the indicator is under-funding.

Analyses 3 and 4 thus provide strong confirmation that the relationships identified by main Analysis 1 and Analysis 2 are present in the data collected on reversal rates and potentially explanatory conditions, and are not a function of particular study methods.

2. State post-conviction Analysis 5.

a. Tests of the analysis's ability to generate reliable results.

Analysis 5 is a Poisson regression analysis of state post-conviction reversals as a proportion of all death verdicts available for review at that stage.421 Relatively few verdicts became available for state post-conviction review because so many were reversed at the prior, direct appeal stage, and so many others were still awaiting review at that stage.422 The result is fewer states and years in which there was at least one verdict available for state post-conviction review, and thus fewer reversal rates to study. The 354 observed rates in Analysis 5 are a third fewer than the 519 observations in Analyses 1 and 2, and 22% fewer than the 453 observations in Analyses 3 and 4. Fewer observations make it harder to achieve the required degree of confidence about the significance of apparent relationships between reversal rates and explanatory factors.

These limitations led us to doubt that we could obtain useful information by separately studying the state post-conviction phase. The reliability tests we use support these doubts to a degree, but less than we expected. Cutting against confidence in the results, our two best Analysis 5 sets of factors—Analyses 5A and 5BCfit the reversal rate data no better (nor any worse) than the baseline analysis of state, year and time trend.423 On the other hand, the two best analyses identify significant explanatory factors that modestly reduce the amount of state-to-state variability left unexplained by the baseline inquiry and that are estimated to have substantial effect size.424 Because the significant Analysis 5 explanatory factors are the same as or similar to significant factors identified by Analyses 1-4, Analysis 5 provides some additional evidence of the importance of those factors, while permitting less confident conclusions about the state post-conviction stage itself.

b. Significant explanatory factors.

Three factors that were significant in Analyses 1-4 were not significant in Analysis 5: the proportion of the population that is African-American; rates of punishing serious criminals; and methods of selecting judges. The non-significance of the last factor is probably explained in part by the fact that state post-conviction judges typically review verdicts imposed at trials they or other trial judges supervised. If political pressures in a given state dispose trial judges to permit flawed capital trials, the same pressures also likely deter the same judges from identifying and curing those flaws.4255 The significant Analysis 5 factors are discussed below.

i. Backlogs of capital verdicts awaiting state post-conviction review. In Analysis 5, as in Analyses 1-4, a measure of backlogged death verdicts is negatively associated with reversal rates. The measure here Cthe number of capital verdicts available for review at the state post-conviction stage minus the number reversed at that stage—is a stage-specific analogue of the three-stage measure used in the other analyses (the number of verdicts awaiting review at all three stages combined). As in Analyses 1 and 2, the reversal rates being studied in Analysis 5 are reversals as a percent of verdicts available for review, including many that in fact were not reviewed. As we have noted, the capital-backlog measure serves in analyses such as these to control for the non-error-related, downward effect on reversal rates of delay and resulting pile-ups of unreviewed verdicts. Accounting for the effect of delay helps assure that other significant factors are related to capital error.426

ii. Year of death verdict. As in Analyses 1 and 2, time trend serves mainly in Analysis 5 as a control for the automatically downward effect on reversal rates of unfinished appeals in analyses in which reversal rates are calculated as a proportion of cases available for review, not as a proportion of cases actually reviewed. In such studies, reversal rates are automatically diminished by the number of verdicts that were not reviewed during the study period, and thus could not have been reversed during that period no matter how flawed they were. And this non-error-related effect of unfinished appeals increases as the year in which verdicts were imposed gets later, because the later a verdict was imposed, the less time it had to be reviewed at the second, state post-conviction phase by the study's 1995 end date.427 In Analysis 5, as in Analyses 1 and 2, therefore, the significant negative relationship between the passage of time and reversal rates that was found is at least partly—and may be entirely—a consequence of the non-error-related fact that many later-imposed verdicts simply could not be fully reviewed on state post-conviction during the study period and thus could not be reversed during that period, no matter how flawed they were. Impressionistic evidence of rising state post-conviction reversal rates over time provide another basis for interpreting this result as a measure of the inevitably rising number of unfinished appeals as death verdicts become more recent, and not as a reflection of changes in the quality of death verdicts over time after controlling for other factors.428

iii. Higher threat of homicides to the white community. Analysis 5 uses two different measures of the impact of homicides on the white community. The first compares states and years based on the number of whites killed in homicides per 100,000 whites in the population.429 The second measure— the same factor found significant in Analyses 1-4Ccompares that same rate to (i.e., divides it by) the number of blacks killed in homicides per 100,000 blacks in the population. The former measure is the homicide victimization rate among whites only; the latter is a measure of the relative risk of homicide felt by members of the white and black communities. In Analysis 5A, increases in the white homicide victimization rate by itself are associated with increases in the state post-conviction reversal rate, and the relationship is highly significant. In Analysis 5B, a higher homicide risk to the white community relative to the risk to the black community also is associated with higher state post-conviction reversal rates, although the relationship is not quite statistically significant.430 Effect size is large for both factors—indeed, considerably larger than for the analogous factors in Analyses 1-4:

  • In the relevant states and years, white homicide victims per 100,000 whites in the population range from about 2 to 17, with an average of about 6. Analysis 5A predicts that states and years with 8 white homicide victims per 100,000 whites in the population will have over twice the state post-conviction reversal rate of states and years with 4 white homicide victims per 100,000 whites. The analysis predicts reversal rates 17 times greater for white homicide rates at the high extreme among states in the study than for rates at the low extreme, when all other factors are held constant at their averages.431

  • Analysis 5B predicts about a 3-fold increase in capital reversal rates when the homicide risk to whites relative to blacks rises from its lowest to highest levels among states in the study.432

These Analysis 5 findings provide additional support for the Analysis 1-4 findings that capital error rates increase as the threat of homicide to the white community increases, which may be evidence that politically potent pressures to use the death penalty increase the likelihood that any given death verdict will be flawed.

iv. More aggressive use of the death penalty. In both Analyses 5A and 5B, the higher the number of death verdicts a state imposes per 1000 homicides, the higher its rate of serious capital error. The relationship is highly significant in both analyses. Effect size again is large. The lowest number of death verdicts per 1000 homicides for any state and year in Analysis 5 is just under 2; the average is 27; the highest is 208. In Analysis 5A, each time the number of death verdicts per 1000 homicides doubles—e.g., from 5 to 10, from 10 to 20, and from 20 to 40Cthe expected reversal rate increases by a factor of 2.43 (i.e., it more than doubles). This means, for example, that as the death-sentencing rate rises from one-half the average among states and years in this analysis (about 13.3 death verdicts per 1000 homicides) to twice the average (about 53 verdicts per 1000 homicides), expected reversal rates increase six-fold.433 This provides additional evidence that less judicious use of the death penalty is associated with higher rates of capital error.

v. Greater population size and density. In state post-conviction Analysis 5, as in Analyses 1 and 2 of all three review stages and direct appeal Analyses 4 and 5, death verdicts imposed in more urbanized (highly and densely populated) states are more likely to be overturned than death verdicts imposed in more sparsely populated states. Effect size again is considerable—about twice as large as for the same factor in Analyses 2 and 4 (both of which are Poisson analyses, like Analysis 5). Compare Figures 30C, 30D and 38B, pp. 190, 208, above.

c. Summary: more evidence of the effect of heavy use of the death penalty.

Analysis 5 provides more evidence that aggressive use of the death penalty, including in reaction to a high risk of homicide to whites, is related to higher rates of serious capital error.

3. Federal habeas Analysis 6.

a. Tests of the analysis's ability to generate reliable results.

Analysis 6 uses binomial regression analysis to identify factors related to reversal rates at the federal habeas stage of review. Those rates are the number of verdicts from each relevant state and year that were reversed at the federal habeas stage as a proportion of the number reviewed at the stage.434 Because federal habeas comes at the end of a long process of attrition of capital verdicts through reversals and delay, only a relatively small number of states and years had at least one verdict that survived review without being reversed or bogged down at the prior two review stages and was finally decided at the federal habeas stage. Analysis 6 thus examines differences in only 161 reversal rates—less than one-third the number of observed reversal rates in Analyses 1 and 2, and less than half as many as in Analysis 5.

Despite the lower number of observations, Analysis 6 achieves results on a par with those of Analysis 5. The best analyses do not fit the reversal rate data any better or worse than the baseline analysis of only state, year and time trend.435 But those analyses: (1) eliminate the significant amount of unexplained and non-random state-to-state variance left by the baseline inquiry; (2) identify significant explanatory factors that are consistent across both best analyses and in important respects overlap the results of other analyses; and (3) identify explanations that are predicted to have large effects on capital reversal rates.436 The analysis thus increases somewhat our confidence in our findings about factors related to reversal rates at all three stages combined and provides some evidence of special factors operating at the federal habeas stage.

b. Significant explanatory factors.

i. Year of death verdict. Analysis 6 avoids a problem Analyses 1, 2 and 5 encounter in trying to measure the effect of time on error rates, because it examines only reversal rates among death verdicts that were actually reviewed at the federal habeas stage, not all verdicts available for review. Analysis 6 thus avoids the non-error-related downward effect on later reversal rates of unfinished appeals.437 But Analysis 6 still is not a good measure of the effect of time on rates of error (as opposed to rates of review), because a different non-error-related effect of the passage of time comes into play at the federal habeas stage. As is illustrated by Figure 10, p. 93 above, the time from imposition to federal habeas reversal of death verdicts was usually from 1.5 to 2 years longer during the study period than the time from imposition to federal habeas affirmance of verdicts. As a result, the pool of verdicts imposed during the study period that were not finally reviewed on federal habeas as of the study's end date likely includes a disproportionately high number of flawed verdicts, and that disparity grows as verdicts get more recent.438 The disproportionately high number of flawed verdicts imposed later in the study period that were still awaiting final federal habeas review as of the study's cut-off date artificially depresses reversal rates for death verdicts imposed in later years—not because verdicts imposed later are less flawed, but because flawed verdicts take longer to review and because, in a time-limited study such as this, the main effect of that bias is to keep flawed verdicts imposed later in time from being counted. For these reasons, time trend serves in Analysis 6 as at least in part a control for the non-error-related effect of disproportionately delayed habeas reversals. And the negative relationship between federal habeas reversal rates and the passage of time in Analyses 6A and 6B439 is at least in part the product of the time lag for reversing flawed verdicts, and not an indication of declining rates of flawed verdicts, after controlling for other factors.440

ii. Higher proportions of people receiving, and expenditures on, welfare. None of the race factors is significant in Analysis 6, but a factor related to economic status is highly significant. States with high ratings on an index measuring the proportion of a state's population receiving welfare and its per capita expenditures on welfare have higher federal habeas reversal rates than other states. Effect size is large: Predicted federal habeas reversal rates rise from 20% to 90% across the range of welfare burdens among states and years in our study.441 See Figure 40B (Analysis 6A), p. 220 below.

iii. Heavier political pressure on state judges. Federal judges are appointed for life—unlike the state judges in our study, nearly all of whom serve for shorter terms and are subject to some kind of direct electoral influence.442 Reviewing federal habeas judges thus do not face the same kinds of political pressures as the state judges who supervise the trials at which death verdicts are imposed and also decide state post-conviction petitions, or the state judges who review capital verdicts on direct appeal.443 If political pressures are related to higher capital error rates, as Analyses 1-4 find, one would expect that relationship to appear most clearly when federal judges review state verdicts. And it does.

In both Analysis 6A (using our first index of political pressure on state judges444) and Analysis 6B (second index), the greater the pressure state judges are under to conform rulings to popular sentiment, the more likely it is that their death verdicts will be found to be seriously flawed on federal habeas review. The relationship is highly significant in both analyses. Effect size is substantial. When other factors are held at their averages, Analysis 6A predicts 4- to 6-fold increases in capital reversal rates at the federal habeas stage when political pressure on state judges is varied from the lowest to the highest levels among states in the study—with predicted error rates of about 55% at the high end of political pressure vs. 12% (Analysis 6A) and 9% (Analysis 6B) at the low end. See Figures 40C-1 and 40C-2, p. 221 below. These federal habeas effect sizes are larger than those at the direct appeal stage, and at all stages combined.445 Compare Figures 40C-1 and 40C-2, p. 221 below, to Figures 29A, 29B and 37A, pp. 188, 207 above.

iv. Lower population size and density. As in all previous analyses, population size and density are important in Analysis 6. But the relationship is reversed: Federal habeas judges are significantly more likely to find serious error in capital verdicts imposed in thinly populated states than in verdicts imposed in more heavily populated states.446 Effect size is large. Holding other factors constant, predicted reversal rates rise from less than 30% to about 65% across the range of population and density ratings among states in our study. See Figure 40D (Analysis 6A), p. 221 below. This result is consistent with lore among federal habeas lawyers that habeas reversals often occur in cases from relatively rural areas, especially in the South.

The difference in the effect of population structure on state versus federal court review is predictable based on the political factors just discussed. On average, less densely populated areas have fewer murders (in number, not per capita) than more highly and densely populated areas. As a result, any murder in a less populous community, and any death verdict imposed for it, is likely to be more publicly visible than murders and death verdicts in more populous areas. This in turn makes the reversal of any such verdict more controversial on average in less than in more populous areas. And that probably makes elected state judges subject to political pressures more reluctant to reverse death verdicts from less populous areas where reversals are more controversial on average than verdicts from cities with more murders and death verdicts.447 As a result, the pool of verdicts surviving state court review and becoming eligible for federal habeas review probably includes a disproportionately high number of flawed verdicts from less populous areas. And because life-tenured federal judges are less politically vulnerable than state judges, they may be less reluctant than state judges to reverse flawed verdicts from those areas. The fact that politically controversial backlogs of unreviewed capital cases have no significant influence on federal habeas reversal rates, but exert a downward influence on state direct appeal reversal rates, may be further evidence of the lower susceptibility of federal judges than state judges to locally generated political pressures to affirm death verdicts.448

c. Summary: more evidence of the influence of politics.

Analysis 6 provides more evidence that local political influences increase the probability of flawed death verdicts, while decreasing the probability that flaws will be corrected by state courts on state direct appeal and post-conviction review.

Figure 40 A-B: Effect Size: Year of Death Verdict (Influence of Delayed Reversals); Welfare Recipients and Expenditures (Analysis 6A)

Figure 40 C-D: Effect Size: Method of Selecting State Judges; Population Structure (Analyses 6A, 6B)

D. Results of Analyses 14 and 15 of the States' 23-Year Capital Experiences: the Role of Race, Politics, Zeal for the Death Penalty and Weak Law Enforcement Confirmed

The statistical analyses above were all programmed to explain variations in capital reversal rates based on two standard statistical assumptions:

  • reversal rates in the same state but from different years are likely to behave similarly; and

  • reversal rates in the same year but from different states are likely to behave similarly.

This approach provides the most rigorous analysis of factors affecting reversal rates besides state and year. But because of the cross-cutting effect of state and year, this approach does not as directly answer a question policy makers in particular states may have: What factors account for differences between a given state's 23 years of experience with capital reversals and the 23-year experience of all other states? Analyses 14 and 15 provide a more direct comparison of the various states' 23-year reversal rate experiences by programming the analysis to assume that common forces affect all reversal rates for the same state regardless of year, without making the same assumption for all reversal rates for the same year regardless of state. Because the focus is on the various states' overall 23-year experiences with capital reversal rates, the effect of the passage of time is omitted.

Analysis 14 is a binomial regression analysis of all three review stages combined. Analysis 15 is a Poisson regression analysis of the same stages.449 In both analyses, the reversal rates being explained are reversals as a proportion of imposed death verdicts.

1. Tests of the analyses' ability to generate reliable results.

As is described above, all our diagnostic tests for Analyses 14 and 15, as for Analyses 1-4, indicate that the results are reliable:

  • The baseline inquiries for both analyses leave substantial unexplained variance.

  • The four best analyses—14A, 14B, 15A and 15BCall significantly reduce unexplained variance.

  • All four analyses do a much better job than the baseline analysis of fitting the predicted results to the actual reversal rates being studied.

  • All four analyses identify a set of statistically significant explanatory factors that are fairly consistent across the two analyses and four sets of factors within those analyses.

  • With the addition of one possible factor, the significant explanatory factors overlap those identified by our main analysis.

  • In all these diagnostic respects, Analysis 15A slightly outperforms the others, and is the main source of our effect-size graphs, Figures 41A-41I, pp. 229-32 below.450

2. Significant explanatory factors.

a. Non-error-related factors gauging low rates of review.

i. Capital backlogs. Analyses 14A, 14B, 15A and 15B all found that the states' number of death verdicts backlogged at the three review stages during the study period is significantly and negatively associated with average reversal rates.451 As in Analyses 1, 2 and 5, this factor serves to control for the non-error-related effect on reversal rates of delayed appeals and resulting low rates of review, as opposed to low rates of error.452 States with more verdicts awaiting review on average tend to have fewer death verdicts reversed, at least in part because they have fewer verdicts reviewed. See Figure 41A, p. 229 below, for this factor's effect size.

ii. Court backlogs generally. In all four analyses, a combined four-factor measure of the per capita number of court cases of all types awaiting decision in each state is also significantly and negatively related to reversal rates. As in Analyses 1 and 2, however, effect size is too small to warrant additional consideration of this factor.453 See Figure 41B, p. 229 below.

b. Error-related factors associated with higher capital reversal rates.

i. The death-sentencing rate per 1000 homicides. In all four Analysis 14 and 15 analyses, as in Analyses 1-5, the more death verdicts a state imposes per 1000 homicides, the more likely it is that any one of those verdicts will be found to be seriously flawed. The relationship is highly significant in all four analyses, and effect size is very high. Holding other factors constant, Analysis 15A predicts that reversal rates increase 9-fold as death-sentencing rates rise from the lowest to the highest rates among states in our study. See Figure 41C, p. 229 below.

ii. Larger black population. As do Analyses 1-4, Analyses 14 and 15 find that death verdicts imposed in states in which African-Americans make up a comparatively higher proportion of the population were significantly more likely to be seriously flawed than verdicts imposed elsewhere. This factor is significant to highly significant.454 Effect size is considerable. Analysis 15A predicts close to a quadrupling of reversal rates when black population size rises from its lowest to highest levels among states in our study, holding other factors constant. Significance455 and effect size drop some, however, in analyses that include homicide rates as a potentially explanatory factor. On effect size, compare Figures 41D-1 (Analysis 15A, which does not include homicide rates as a factor) and 41D-2 (Analysis 15B, which includes homicide rates as a factor), p. 230 below. We discuss the connection between the Apercent black@ and Ahomicide rate@ factors in point iii below.

iii. Higher homicide rates. Although we considered homicide rates as a potential explanation for capital error in all our analyses, it usually was not significant. In Analysis 15B, however, but not in any other Analysis 14 and 15 inquires, homicide rates were significantly related to homicide rates.456 The higher a state's average homicide rate, the higher its average rate of serious capital error. See Figure 41E-2,457 p. 230 below, for the factor's moderate effect size in Analysis 15B. This result may provide some additional evidence that pressures to use the death penalty as a response to serious crime are related to higher error rates.458

Homicide rates and the percent of the state's population that is black are correlated. States with higher black populations tend to have higher homicide rates.459 There is some evidence that these two factors compete in Analyses 14 and 15. Percent black is clearly the more important of the two. Unlike homicide rates, percent black is always significant at the .05 level and is sometimes highly significant, both in Analyses 14 and 15 and in most other analyses. But when both factors are included in the same Analysis 14 and 15 analyses, the significance level and effect size for the size of the black population diminishes.460 As is noted above, Analysis 15A predicts nearly a quadrupling of error rates, other factors being equal, when the size of the black population rises from its lowest to highest levels among study states. The corresponding rise in Analysis 15B, which adds homicide rates as a factor, is less steep. Compare Figure 41D-1 and 41D-2, p. 230 below. The trade-off in effect between homicide rates and the size of the black populationsCwith the latter factor being the stronger of the two—supports a point made above: Racial stereotypes may lead influential members of the public to treat the size of a state's black population as a proxy for the threat of homicide, pressuring officials in states with larger black populations to pursue more death verdicts in weaker cases where the risk of error is high.461

iv. Relatively higher risk of homicide to whites as compared to blacks. As do Analyses 1-5, Analyses 14 and 15 indicate that capital error rates are higher for verdicts imposed in states where the risk of homicide to members of the white community is more nearly the same as the risk to members of the black community than in states where the homicide risk to whites is lower than to blacks. This factor is significant at the .01 level in Analysis 15A and at the .05 level in Analysis 15B and falls just above the .05 significance level in Analysis 14.462 Following the pattern noted above in regard to the size of the black population,463 the effect size of our measure of whether homicides are more or less heavily concentrated on whites relative to blacks is greater in Analysis 15A, which does not include general homicide rates as a factor, than in Analysis 15B, which does include that factor. Compare Figures 41F-1 and 41F-2, p. 231 below.464 There is evidence, then, that (1) a higher risk of homicide—which probably increases pressure to extend the death penalty to weaker cases—contributes to higher rates of capital error, and (2) that the pressure generated by the risk of homicide is greatest when that risk to members of the white community approaches or surpasses the risk to members of the black community.

v. Interaction of larger black population and higher risk of homicide to whites as compared to blacks. As do Analyses 1-4, Analyses 14 and 15 indicate that, independently of the effect of (1) the proportion of blacks in a state's population and (2) the rate of white, compared to black, homicide victimization, the interaction of the two factors is related to error rates: Death verdicts are especially likely to be seriously flawed in states in which it is true both that African-Americans on average make up a relatively high proportion of the population and that the homicide risk to members of the white community approaches or surpasses the risk to members of the black community. In those states, that is, reversal rates are even higher than the individual components of the interaction would predict. This interaction is significant in Analysis 15A, but not quite in the other analyses.465 If the relative size of the black community and the relative threat of homicide to the white community each generates potent pressures to use the death penalty and a resulting higher risk of capital error, then the presence of both conditions may multiply the effect.

vi. Lower rates of arrest, conviction and incarceration per serious crime. In Analyses 14 and 15, as in Analyses 1-4, the lower a state's ratio of prison inmates to 100 FBI Index Crimes, the higher its rate of serious capital error. The relationship is highly significant in all four Analysis 14 and 15 analyses, and effect size is large. Analysis 15A predicts that reversal rates increase by more than 8-fold as rates of apprehending and punishing serious criminals drop from their highest to lowest levels among states in these analyses. See Figure 41G, p. 232 below. Our tentative interpretation is that pressures to use the death penalty that are generated by the threat of crime, and the higher capital error rates that result, are greater in states with relatively less effective non-capital responses to crime.466

vii. More political pressure on state judges. Analyses 14 and 15 find what Analyses 1-4 and 6 also found: Death verdicts are more likely to be seriously flawed if they are imposed in states where judicial selection methods increase judges' incentives to conform their rulings to popular sentiment. The relationship is highly significant in all four analyses. Effect size in this analysis of all three review stages is moderate.467 See Figure 41H, p. 232 below. If political pressures lead to high error rates, the effect should be greater in states where judges have a greater incentive to act consistently with popular opinion. That , indeed, is what Analyses 14 and 15 (and Analyses 1-4 and 6) find.

c. Other factors.

i. Higher population size and density. As in Analyses 1-6, population structure is significantly related to error rates in all four Analysis 14 and 15 analyses. And as in Analyses 1-5 (but not Analysis 6), the relationship is positive: The more heavily populated a state on average, the higher its average reversal rates. See Figure 41I, p. 232 below, for the factor's considerable effect size.

ii. The interaction of high capital and general court backlogs. Analyses 14 and 15 reach the same result as Analyses 1 and 2 in regard to the interaction of high capital and general court backlogs: Although each type of backlog by itself is negatively associated with reversal rates, when the effect of the two factors is multiplied, that interaction is positively related to reversal rates. Even after accounting for their delay-related slowness to review (and thus their small numbers of reversed) death verdicts, jurisdictions whose courts are snowed under by both capital and non-capital backlogs tend to have particularly high rates of flawed death verdicts.

Figure 41 A-C: Effect Size: Backlog of Capital Verdicts; State Court Caseloads; state Death Sentencing Rate (Analysis 15)

Figure 41 D1-E2: Effect Size: Proportion of African-Americans in State Population; Homicide Rate (Analyses 15A, 15B)

Figure 41 F1-F2: Effect Size: Risk of Homicide to Whites Versus Blacks (Analyses 15A, 15B)

Figure 41 G-I: Rate of Arrest, Conviction and Impirsonment Per Serious Crime; Method of Selecting State Judges; Population Structure (Analyses 15A, 15B)

3. Summary: the relative unimportance of time.

Analysis 14 and 15's more focused comparison of the 34 capital states' 23-year experiences with capital reversals closely replicates the results of Analysis 1 and 2's combined comparison of each year's 34-state experience and each state's 23-year experience. The only difference is the significance in Analysis 14 and 15 of an explanatory factor not identified by Analyses 1 and 2Chomicide rates.468 Even here, however, the difference is modest because the new factor is correlated with two factors Analyses 1 and 2 did identify—the proportion of the state's population that is black, and the extent to which the homicide risk to members of the white community approaches or surpasses the risk to blacks. Moreover, those two factors appear to be more targeted and more powerful measures of the same kinds of crime fears that may be generated by homicide rates and may account for their association with error rates.

Analysis 14 and 15's replication of the results of Analyses 1 and 2, despite omitting time as a factor, has important implications:

  • Assessing the effect of time—the influence of particular years and of any trend over time—is not crucial to an understanding of the factors that contribute to differences in capital reversal rates.469

  • The important differences to be explained in capital reversal rates are between states, not between years, or between earlier and later years.

  • The individual factors our other six state analyses identify, and the set of factors main Analysis 1 and supporting Analysis 2 identify,470 have a stable association with capital reversal rates that appears whether or not the generalized effects of year and time trend are considered.

E. The Reliability and Robustness of the Results of the Eight State Analyses

1. The reliability of the eight analyses.

Below we consider the reliability of the eight state analyses based on the results of eight diagnostic tests:471

  • Test 1: Does the baseline analysis of only state, year and time trend leave significant unexplained differences among states and years?472

      Yes: Analyses 1, 2, 3, 4, 5,473 6,474 14, 15

  • Test 2: Does the amount of unexplained variance decrease when other factors are considered?

      Yes: Analyses 1, 2, 3, 4, 5, 6, 14, 15

  • Test 3: Does the consideration of other factors significantly improve the fit of the predicted and actual results?

      Yes: Analyses 1, 2, 3, 4, 14, 15
      No: Analyses 5, 6

  • Test 4: Are explanations for differences in reversal rates identified that are unlikely to appear by chance (i.e., are statistically significant)?
      Yes: Analyses 1, 2, 3, 4, 5, 6, 14, 15

  • Test 5: Does the set of significant factors the best analyses identify fit the data better than other sets?

      Yes: Analyses 1, 2, 3, 4, 5, 6, 14, 15

  • Test 6: Are the significant factors the analysis identifies congruent with those identified by a parallel analysis using a different regression technique (binomial vs. Poisson )?

      Yes: Analyses 1, 2, 3, 4, 14, 15
      Not tested: Analyses 5, 6

  • Test 7: Are the significant factors the analysis identifies congruent with those identified by parallel analyses examining different review stages (all three stages vs. state direct appeal vs. state post-conviction vs. federal habeas)?

      Yes: Two-thirds or more of the significant factors in each of five analyses—1, 2, 5, 14 and 15Care also significant in analyses of at least two other review stages.

      Yes: Half or more of the factors in each of Analyses 1, 2, 3, 4, 5, 6, 14 and 15 are also significant in studies of at least one other review stage. 475

  • Test 8: Are the significant factors each analysis identifies congruent with those identified by parallel analyses that compare each state's 23-year experiences with reversals?476

      Yes: Analysis 1 (92%), Analysis 2 (92%), Analysis 3 (89%), Analysis 4 (89%), Analysis 5 (80% [or 100%]). In each of these analyses, the indicated percentage of the significant factors are reached significance in Analyses 14 and 15, which directly compare states' 23-year experiences with capital reversals.477

    The results of these tests justify confidence in the reliability of the results of main Analysis 1 as well as supporting Analysis 2. These analyses are, respectively, a binomial logistic, and a Poisson regression analysis of factors related to state-to-state and year-to-year differences in reversals of death verdicts at all three review stages as a proportion of all death verdicts imposed during the study period. The diagnostic tests confirm our judgment that main Analysis 1 and also supporting Analysis 2 are the most reliable methods of extracting the largest amount of information from our detailed data. The tests reveal that the results of Analyses 1 and 2 reflect actual relationships in the data, which are not sensitive to different regression techniques, different measures of reversal rates, the review stage being analyzed, or whether the focus is on state-to-state and year-to-year or only state-to-state differences in reversal rates.

    The tests also justify confidence in the results of Analyses 3, 4, 14 and 15. Analyses 3 and 4 are a binomial and a Poisson regression analysis of factors related to state-to-state and year-to-year differences in reversals of death verdicts at the state direct appeal stage as a proportion of all verdicts imposed during the study period that were actually reviewed at the direct appeal stage. The state direct appeal review stage analyzed by Analyses 3 and 4 is particularly important because it is the only stage that reviews essentially all capital verdicts and it accounts for nearly 80% of all reversals during the study period.478 Analyses 3 and 4 also are important because they provide the most reliable measure of the effect of the passage of time after controlling for other factors.479 Analyses 14 and 15 fill out our knowledge still further by focusing more directly on state-to-state and less on year-to-year differences in reversal rates.

    State post-conviction Analysis 5 and federal habeas Analysis 6 confirm results of the other six analyses by identifying explanations for reversal rates at the two later review stages that overlap the other analyses' explanations for reversal rates at the first review stage and the three review stages combined. Analyses 5 and 6 justify somewhat less confidence in their stage-specific results. Although both analyses perform well on most diagnostic tests, neither improved on the fit achieved by their baseline analyses, and the explanations Analysis 6 identified overlap less than usual with factors identified by other analyses. We do, however, have some confidence in two stage-specific results of Analysis 6, given their confirmation by results from other analyses:

    • A pattern of results from other analyses (1) reveals a positive relationship between capital error rates and judicial selection methods that pressure state judges to conform rulings to popular opinion, and (2) predicts that the relationship will be strongest at the federal habeas stage where judges are life-tenured and more isolated from local political pressures.480 The direction, significance and effect size of the relationship Analysis 6 finds between error rates and political pressure is consistent with this prediction, giving us confidence in this Analysis 6 result.

    • For similar reasons, the different direction of the relationship between capital error rates and population size and density in Analyses 1-5 (where error rates are positively associated with populousness) and Analysis 6 (where error rates are negatively associated with populousness) tends to support the latter result. Because federal habeas is the last review stage and the only one staffed by life-tenured judges, it would be expected to bear more of the burden of reversing flawed verdicts from rural areas where such reversals tend to be especially controversial.481 Here again, the conformity of an Analysis 6 result to conditions predicted for the federal habeas stage by a pattern of results at the other stages gives us some confidence in this Analysis 6 result.

      2. The robustness of the significant explanations for state differences in capital reversal rates.

    Table 7 below lists each of the explanations for reversal rates that was significant in one or more of the eight state analyses discussed above. To measure the robustness of each explanatory factor, Table 7 reports the number of analyses out of eight total in which the factor provided a significant explanation for state-to-state differences in capital reversal rates, and the review stage(s) at which the factor offers that explanation.

    Table 7: State-Level Factors That Were Sometimes Significant, and How Often They Were Significant482
     
    Which explanatory factor?± Significant in how many of 8 analyses? # Significant at what stages?
    Death Verdicts/1000 Homicides 7 all 3,** da, pc
    Population Size and Density 7 all 3,** da, pc
    ­Death Verdicts Awaiting Review, 3 Stages 6 all 3,** da
    ­Same, State Post-Conviction Stage Only 1 pc
    White/Black Homicide Victimization Rate or White Victimization Rate 6& all 3,** da, pc
    % of Population that is African-American 6 all 3,** da
    ­Prison Population/100 FBI Index Crimes 6 all 3,** da
    Index 1, 2 of Political Pressure on Judges 5& all 3,** hc
    ­4-Factor or 1-Factor Measure of All State Court Cases Per Capita Awaiting Decision 4 all 3**
    Death Verdicts Awaiting Decision x 4-
    Factor or 1-Factor Measure of All State
    Court Cases Per Capita Awaiting Decision
    4 all 3**
    % of Population that is African-American x White/Black Homicide Victimization Rate 3 all 3**
    ­Per Capita Expenditures on State Courts 2 da
    +Passage of Time (reliable)^ 2# da
    ­Passage of Time (unreliable)^ 4# all 3, pc, hc
    % Pop., Per Capita Spending, on Welfare 1 hc
    Homicide Rate 1 all 3**
    ­Population Size and Density 1 hc
    3 Measures of Race of Defendant and Victim for Offenses Punished by Death# 1 all 3

        a. Seven factors robustly related to high capital reversal rates: aggressive death sentencing; large black population, relatively high risk of homicide to whites, heavy political pressure on state judges, weak record of apprehending and imprisoning serious criminals, population size and density, and low capital backlogs.

    Given the range of methods the eight analyses use to explain state-to-state differences in capital reversal rates—different regression techniques, measures of reversal rates, review stages, groupings of reversal rates by state and/or year, and combinations of potential explanations—it is reasonable and conservative to express confidence in explanations for capital reversal rates that are significant in three-fourths or more of the analyses.

    Seven explanations—set out below in order of robustness—satisfy this conservative approach. After controlling for other factors in the analyses:

    • States that impose more death verdicts per 1000 homicides have higher rates of serious error than states that use the penalty less often. The more frequently a state uses the penalty per homicide, the more likely it is that any one of its death verdicts is seriously flawed. [7/8 analyses; 3/4 stages;483 2/2 ways of grouping reversal rates by state and time.]

    • More densely populated states have higher reversal rates at the state court stages of review than do less populous states. [7/8 analyses; 3/4 stages; 2/2 ways of grouping rates.] At the third, federal habeas stage of review, the effect is reversed. Federal habeas Analysis 6 finds that sparsely populated states have higher reversal rates than more populous states. [1/8 analyses; 1/4 stages; 1/2 ways of grouping states.] Overall, capital error rates are sensitive to population size and density. [8/8 analyses; 4/4 stages; 2/2 ways of grouping rates.]

    • States with more death verdicts awaiting review have lower reversal rates. This is especially true in analyses where low review rates dictate low reversal rates calculated as proportions of imposed death verdicts. It is also true in analyses where reversal rates are proportions of only reviewed death verdicts, suggesting that backlogs create pressures to approve flawed verdicts. [7/8 analyses,484 3/4 stages, 2/2 ways of grouping rates.]

    • States where the proportion of whites killed by homicide more nearly approaches the proportion of blacks killed by homicide have higher rates of serious capital error than states where the homicide burden is more heavily concentrated on blacks. In one study it was the white victimization rate by itself that had this effect. [6/8 analyses;485 3/4 stages; 2/2 ways of grouping reversal rates.]

    • States with larger African-Americans populations relative to their white population have higher capital error rates than states where blacks are a smaller part of the community. [6/8 analyses, 2/4 stages, 2/2 ways of grouping rates.]

    • States that arrest, convict and imprison fewer criminals for every 100 serious crimes have higher rates of serious capital error than states that bring larger proportions of serious criminals to justice. [6/8 analyses, 2/4 stages, 2/2 ways of grouping rates.]

    • States whose judicial selection methods give judges more of an incentive to conform their rulings to popular sentiment have higher capital error rates. [5/8 analyses plus 2 other analyses just barely above significance;486 (2+1)/4 stages; 2/2 ways of grouping reversal rates.]

    There are additional reasons for confidence in these seven key explanations for capital reversal rates. First, four additional factors were significant in some analyses and tend to confirm the seven explanations by identifying similar or related ones:

    • In three analyses, the interaction of the two racial factors discussed above is independently related to error rates: States where a high homicide risk to whites relative to blacks interacts with high numbers of blacks relative to whites have especially high capital error rates. [3/8 analyses,487 1/4 stages, 2/2 ways of grouping reversal rates.]

    • In four analyses, the negative effect of capital verdicts awaiting appeal on review and reversal rates (the fourth key factor) is paralleled by the negative effect of per capita rates of court cases generally that are awaiting decision. This delay-focused effect is felt only in analyses where low review rates dictate low reversal rates calculated as proportions of all imposed death verdicts. This factor has a low effect size, however. [4/8 analyses, 1/4 stages, 2/2 ways of grouping rates.]

    • In one analysis, states with higher general homicide rates have higher capital error rates. This provides some support for the finding that relatively high homicide rates among whites compared to blacks (the fifth of the seven key factors) are related to high capital error rates. A link between the effect of general homicide rates and the comparative threat of homicides to whites and blacks is indicated by the fact that in the two analyses in which general homicide rates were significant, the effect of the risk of white relative to black homicide victimization diminished somewhat. When general homicide rates were significant in the two analyses, the significant relationship between relatively large African-American populations and high reversal rates (the sixth key factor) also diminished somewhat—suggesting a link between crime fears and resulting capital error rates and the size of the black population.488 [1/8 analyses, 1/4 stages, 1/2 ways of grouping reversal rates.]

    • In the federal habeas analysis, the size of the black population was not related to high error rates, but a combination of the size of the population receiving welfare and per capita expenditures on welfare were related to high error rates. States with higher percentages of residents receiving welfare and higher per capita expenditures on welfare had higher capital reversal rates at the habeas stage. This may be due to the correlation between high African-American populations and high rates of welfare assistance, or because large and visible populations of poor people increase crime fears among more well-to-do residents. Either effect is similar to, and tends to support, the influence of high proportions of African-American residents (the sixth of the seven factors). [1/8 analyses, 1/4 stages, 1/2 ways of grouping rates.]

    Another reason for confidence in the seven key factors is their effect size. Moderate (at least a doubling) to very large increases (4-, 6- and even 9-fold increases) in capital reversal rates are predicted when moving from one end of the spectrum among states in our study to the other in regard to each of the seven conditions, holding other conditions constant.489

    A third additional reason for confidence in the seven key factors is the existence of a convincing explanation for its failure to appear significant in a minority of the analyses. Consider, for example, the one analysis in which judicial selection methods, and resulting pressures on judges to conform to popular sentiment, are not significant. That is the only analysis that studies decisions by reviewing judges who are subject to essentially the same political pressures as judges making the errors in the first place and who thus are unlikely to spot and cure errors generated by that condition.490

    As noted, there is also a good reason why high population size and density are associated with high reversal rates in the seven analyses of, or dominated by, state court reversals, while the same condition is associated with low reversal rates in the one analysis of federal court reversals. Reversals of death verdicts for homicides in rural areas are likely to be more controversial to entire communities on average than reversals of verdicts for homicides in urban areas. And the elected state judges who are mainly responsible for the two state court review phases have greater incentives to avoid such controversy than the life-tenured judges responsible for final, federal review. The likely result is exactly the division of labor all eight analyses find: Elected state judges do a better job of screening urban death verdicts for error, leaving to less politically vulnerable federal judges the more controversial task of reversing flawed verdicts from rural areas and small towns.491

    In the rest of the cases, the minority of analyses in which one of the seven key explanations did not appear significant were ones in which there were fewer observations to be explained, making it more difficult for the analysis to identify important relationships that may nonetheless exist.

    Additionally, all seven key factors were significant explanations for capital reversal rates in all six of our best state analyses (Analyses 1-4, 14, 15), judged by the number of reversal rates that were available to be compared and explained. Indeed, the seven factors together make up the broad core of the set of significant explanations for reversal rates in those analyses. And when considered together with the four allied factors listed above, the seven key factors formed virtually the entire set of significant factors in the best analyses. There thus is reason for confidence not only in the reliability of each of the seven explanations by itself, but also that the seven factors taken together provide a reasonably comprehensive understanding of the important explanations for state-to-state differences in capital reversal rates during the study period.

        b. A single robust explanation with seven interlocking parts: excessive use of the death penalty as opposed to other responses to crime.

    As the last point suggests, a final reason for confidence in the seven key explanations for reversal rates is that together they point towards a single, persuasive explanation for capital reversal rates. They do so, not only because they consistently appear in the best state analyses but, more crucially, because they fit together logically, by virtue of what is known about the capital system, into an interlocking and potentially convincing rationale for differences in capital error rates across states and years. We sketch out parts of that overarching explanation above, and say more about it below after reporting the results of our county and case-level studies. It is useful here to articulate the explanation and note how it connects the seven factors in which our state analyses give us most confidence, as well as the four allied factors listed above:

    When it comes to capital punishment, less is better; more is worse. The fewer death verdicts a state imposes per 1000 homicides, the less likely it is that any given verdict will be reversed due to serious error. And the fewer death verdicts a state imposes, the less overburdened its capital review system is, and the more likely it is to carry out the verdicts it does impose. Conversely, states that more often give in to pressures to use the death penalty and extend it to marginal cases have significantly higher rates of serious capital error, more delay in processing appeals, less success carrying out the verdicts they impose, and a greater temptation to approve flawed verdicts on appeal. Among the sources of pressure to overuse the death penalty in these ways are politics, the ineffectiveness of the state's non-capital response to serious crime, the race and, possibly, the economic status of the state's residents and homicide victims, and on appeal pile-ups of capital cases awaiting review and flawed verdicts' imposition in non-urban communities.

        c. Two other potentially important factors: over-burdened courts and time.

    Our analyses suggest two other explanations for high capital error rates. Our demanding tests warrant somewhat less confidence in the two explanations, but each has some support in the analyses, is consistent with the overarching explanation and is plausible:

    • Verdicts imposed by states with over-burdened and under-funded courts are more likely to be flawed than those imposed by states with average or better caseloads and funding:

    • In analyses of all three review stages, the significant factor is the interaction of heavy capital and non-capital caseloads: Courts with high capital and non-capital case-loads have high error rates. [4/8 analyses, 1/4 stages, 2/2 ways of grouping rates.]

    • At the direct appeal stage, the important factor is per capita spending on courts: Poorly funded state courts generate more capital error. [2/8 analyses, 1/4 stages, 1/2 ways of grouping rates.]

      These two results are different sides of the same coin: Having too many cases to decide likely means having too few resources to decide them reliably. It thus is reasonable to treat the two points as aspects of a single explanation: insufficient resources for capital trials. If the four analyses in which high capital and non-capital caseloads are significant are added to the two analyses in which poor funding is significant, the combined explanation crosses our demanding 75% line for explanations deserving a high degree of confidence. [6/8 analyses, 2/4 stages, 2/2 ways of grouping reversal rates.]

    • Whether death verdicts were imposed earlier or later in the study period seems to affect the probability that they will be found to contain reversible error at different stages of review:

    • Controlling for other factors, death verdicts imposed later in the study period are more likely to be found seriously flawed by judges at the direct appeal stage than verdicts imposed earlier in the study period. [2/6 analyses,492 1/4 stages.] Direct appeal Analyses 3 and 4 are the most reliable measure of changing amounts of reversible error found over time, giving us confidence in this result.493 The result is important because direct appeal reversals are about 80% of all reversals between 1973 and 1995.494 The result has three possible explanations: Controlling for other factors, later verdicts may be more seriously flawed than earlier ones; state direct appeal courts may have detected more serious error over time; or both. As we point out above, the fact that this finding occurs only after controlling for other factors—reversal rates in fact were quite stable during the latter half of the study period495‐has an important implication. Evidently the specific factors we have been able to isolate that account for differences in error rates predict that those rates should have declined some over time. The fact that they did not decline indicates the influence of still other time-sensitive factors, captured in our analysis by time trend, that are associated with an upward trend in error rates. This means that reforms generated by known specific influences on error rates may not be as effective as one would hope because of the competing influence of conditions that are known to be increasing over time but otherwise are not well understood.496

    • Among death verdicts reviewed on federal habeas (i.e., among verdicts surviving inspection at the prior two, state stages of review), and after controlling for other factors, verdicts imposed earlier in the study period were more likely to be reversed than verdicts imposed later in the period. [1/6 analyses, 1/4 stages.] Federal habeas reversals account for 10% of all reversals.497 We are not confident in this result as a reflection of changing amounts of flawed verdicts over time because of a time-sensitive bias: Federal judges take more time on average to reverse flawed death verdicts than to approve unflawed ones. As a result, disproportionately more of the flawed verdicts imposed later in the study period were still under review at the end of the period and were not counted in our study than is true of other verdicts.498 The relationship between later verdicts and fewer federal habeas reversals thus could be entirely a function of longer delays in reviewing flawed verdicts, which led us to count a disproportionately small share of the flawed verdicts imposed later in the study period. Or, the result could be a function of that effect (which we know accounts for some of the result) plus the fact that state judges caught more error over time at the direct appeal stage, leaving less error to be caught at the later federal habeas stage. Support for this latter interpretation is found in evidence presented above that state courts increase their scrutiny for serious capital error to compensate for federal courts known to exercise low levels of scrutiny,499 and in the fact that a series of judicial and other cut-backs on the ability of all federal habeas courts to scrutinize capital verdicts for error—which began in 1986 and accelerated in 1989-93.500 Or, finally, this result could be a function in part of the fact that, controlling for other factors, there is less error in later death verdicts. This last possibility is hard to reconcile with the contrary finding at the direct appeal stage that, controlling for other factors, there is substantially more serious error in later verdicts.

    • The remaining analyses demonstrate the obvious in regard to later as opposed to earlier verdicts: Court review takes time, so later verdicts are less likely to have been finally reviewed by the end of the study period than earlier verdicts, and thus were less likely to have been reversed or affirmed. Because the rest of our analyses calculate reversal rates as a proportion of imposed verdicts, not as a proportion of actually reviewed verdicts, the declines in reversal rates those analyses associate with later death verdicts were partly or entirely a result of the decreasing probability of review over time, not the decreasing probability of error. [3/6 analyses, 2/4 stages.]


    Before reaching a final interpretation of the conditions that are significantly associated with capital reversals and reversal rates (see Part VII below), we consider how the above results of our eight state-level analyses are affected by results from eight analyses of county (as well as state) conditions and from an additional study of case level conditions (discussed in Parts V, VI below).