next | previous | table of contents

VII. Summary and Interpretation of Results: The Strongest Predictor of Serious Capital Error Is Aggressive Use of the Death Penalty, Extending to Weakly Aggravated Homicides, in Response to Political, Race-Related and Law-Enforcement-Related Fears and Pressures

A. Summary of Methods

Parts IV-VI of this Report describe the results of 19 separate statistical analyses of state- and county-level factors related to high state and county rates of reversible capital error, and of case-level factors associated with a probability of federal habeas reversal of death verdicts. The analyses use a variety of statistical techniques, including classic logistic, over-dispersed binomial logistic and over-dispersed Poisson logarithmic regression analyses to identify factors that explain why some states and counties have more capital error than others and why some capital verdicts are reversed on federal habeas review and others are not. To assure that they are comprehensive, conservative and reliable, the analyses:

  • use a variety of statistical methods with different assumptions about the arrangement of the condition being studied—capital reversals and reversal rates—to ensure that it is relationships in the data, not statistical methods, that drive the results;

  • analyze reversals and reversal rates at each of three stages of court review of capital verdicts—state direct appeal, state post-conviction and federal habeas—and at the three stages combined;

  • use different methods to analyze the simultaneous effect on reversals and reversal rates of important general factors, such as state, county, year and time trend, and specific conditions that may explain capital reversals and reversal rates;

  • examine explanatory factors operating at the state, county and case level;

  • were all subjected to tests for statistical significance, variance left unexplained, fit between predicted and actual results, and effect size; and

  • were tested for consistency within analyses and across analyses to determine which form of analysis, which individual factors identified as statistically significant, and which interlocking sets of significant factors are the most robust and warrant the most confidence.

The 19 regression analyses were supplemented with two sets of case studies, each of which broadened the focus from serious, reversible capital error, to another kind of serious error: the capital conviction and sentencing of people later shown to be innocent of a capital crime. The first set of case studies examine why full sets of reviewing courts at all three review stages approved the execution of four innocent men who thereafter were saved only by the fortuitous, eve-of-execution discovery of exculpating DNA, the reinterpretation of an exonerative video tape after a decade of apparently minor discoveries cumulated to discredit false testimony that had emarginated the tape at the original trial, and an actual perpetrator's confession to intrepid college students taking part in a class project.697 The second set of analyses examines the capital-error records (including for convicting and condemning people later shown to be not guilty) of paired sets of American counties with similar numbers of homicides but different rates of using the death penalty.698

Our basic approach in using this array of statistical methods and case studies, explanatory factors and controls, and diagnostic tests was to start with one analysis reflecting our best judgment about the most reliable way to study conditions associated with serious, capital error (Analyses 1), then systematically to address possible objections to that analysis with alternative methods that evaluate or avoid the objection. Our choice of results to treat as worthy of attention and analysis, and to carry forward to this section's interpretation of all results as a whole, is conservative: Unless there is a substantial basis for confidence in a result, given the methods used to reach it, its statistical significance, its performance on the other diagnostic tests, its consistency with results of other analyses and its consistency with logic and experience, we omitted it from further consideration. We use the same approach here in analyzing the factors and interpretations that have survived this gauntlet of tests and comparisons.

As we show in Broken System, Part I, and in Part III of this Report, high rates and amounts of serious, reversible capital error have broken the nation's death penalty system. We begin here with our single, principal conclusion about the condition most strongly and consistently associated with high rates and amounts of reversible capital error:

  • The more aggressively officials use the death penalty—the more often they use it and the more frequently they apply it to homicides that are not highly aggravated—the greater is the risk that any death verdict they impose will be seriously flawed.

We also reach five supporting conclusions grounded in the study results that expand our understanding of the principal conclusion:

  • Several conditions that are strongly associated with serious capital error have a common tendency to increase pressure on officials to use the death penalty aggressively:

    • the risk of homicide to the entire community, especially when the risk to politically influential citizens approaches or exceeds that to other citizens—as measured here by how close the homicide risk to whites comes to equaling or surpassing the homicide risk to blacks;

    • crime fears associated with racial and possibly economic conditions—as measured here by the proportion of the population that is African-American, and by the amount of spending and number of residents on welfare;

    • well-founded doubts about the ability of the state's law-enforcement system to deal effectively with crime through arrest, conviction and incarceration; and

    • state trial judges' susceptibility to being harmed politically if their capital rulings do not conform to popular sentiment.

  • Overuse of the death penalty causes harms beyond serious, reversible error, including cost, delay and the system's inability to achieve its most basic goals.

  • Poor quality trial proceedings—which are in part a function of heavy use of the death penalty—also appear to increase the risk of serious, reversible error.

  • After controlling for other factors, conditions leading to capital reversals at the state direct appeal stage of review—which accounts for 79% of all reversals—have gotten substantially worse over time, given the strong association between later verdicts and higher reversal rates. The same may be true at the other review stages. There is no reliable evidence that conditions creating serious capital error have improved over time.

  • State and federal reviewing judges are themselves susceptible to political pressure and mistake, and thus are not a reliable substitute for careful and accurate capital trials.

The principal conclusion and most of the supporting conclusions are obvious implications of strong and consistent study results requiring little interpretation. The first and last supporting conclusions rely additionally on logic and experience. We are confident in the reliability of all of these conclusions and their strength and sufficiency as bases for changes in policy. All of them inform our sense of urgency about the need for serious policy reforms. The principal conclusion drives most of the policy suggestions in Part VIII below.

B. Principal Conclusion: Heavy Use of the Death Penalty Extending Beyond Highly Aggravated Homicides Substantially Increases the Risk of Serious Capital Error

Recently, the Washington Post quoted a statement by Joshua Marquis, District Attorney of Clatsop County, Oregon, and a Board Member of the National District Attorneys Association, that "[t]here is a growing acknowledgment generally that the death penalty should be reserved for the worst of the worst."699 A few weeks earlier, Virginia's Governor, James Gilmore, expressed the same sentiment on CNN: The death penalty should be "reserved only for the worst possible cases."700 The state-, county- and case-level results underlying our major finding reveal the wisdom of these views, and the need to enforce the "worst of the worst" principle strictly in order to bring serious capital error under some sort of control.

1. High state-level capital-sentencing and high capital-error rates.

States vary widely in how often they punish homicides with death. During the 23-year study period from 1973 to 1995, 34 active capital states imposed death verdicts in one or more of those years, totaling 519 sets of states and years. The average rate of death verdicts imposed in all 519 states and years was 18 per 1000 homicides. But rates ranged from about 1 death verdict per 1000 homicides (e.g., in Illinois in 1977) to 208 in Idaho in 1982.

Figure 11, p. 121 above, compares states based on how often they used the death penalty during the entire study period. Death-sentencing rates per 1000 across that period homicides ranged:

  • from less than 5 in Connecticut, Colorado and New Mexico, and between 5 and 10 in Maryland, New Jersey and Washington;

  • to around 10 in California, Kentucky and Louisiana, and around 12 in Illinois, Indiana, and Virginia;

  • to from 32 to 37 in Alabama, Florida and Montana, and around 45 in Arizona, Delaware, Nevada and Oklahoma;

  • to 60 in Idaho.

The most consistent finding of our 19 analyses is that these disparities in capital-sentencing rates are strongly associated with disparities in capital-error rates. The more death verdicts jurisdictions impose per 1000 homicides, the more likely it is that any single death verdict they impose will later be reversed due to serious capital error. This is a significant finding of:

  • our main regression Analysis 1;701

  • most of the 17 confirming state and county regression analyses;702

  • analyses of all three review stages combined,703 and of two of the three review stages individually,704 with supporting results from our case-level study of the remaining (federal habeas) stage;705

  • analyses designed to explain county reversal rates, as well as state reversal rates;706

  • analyses identifying explanatory conditions operating and measurable at the county level— death-sentencing rates being the main, significant county-level explanation for county reversal rates707—as well as analyses identifying explanatory conditions operating and measurable at the state level;708 and

  • our county case studies of capital-sentencing and capital-error rates.709

This explanatory factor has a large predicted effect on rates of serious capital error. Analysis 1—the most complete analysis of our detailed data on capital reversal rates710— predicts that capital-error rates will increase from less than 15% to more than 75% as death-sentencing rates rise from the lowest to the highest levels among states and years in our study, holding other explanatory factors at their averages.711 Predicted increases in error rates are especially steep around the average death-sentencing rate, meaning small changes in death-sentencing practices within the range where most states operate are predicted to have large payoffs in terms of reduced capital error.712

Table 18, p. 344 below, ranks the 34 states based on the degree of risk each faces from six conditions associated with higher rates of serious capital error. The risk posed by all but one factor, holding other factors constant at their averages, is based on the results of main Analysis 1A.713The comparative risk posed to each state is based on each state's weighted average value on the relevant condition during the study period, with weights assigned based on the state's yearly contribution to the pool of capital verdicts being studied.714 In addition to each state's rank and weighted average value on each factor, Table 18 indicates whether the capital error rate Analysis 1 predicts for the state based on the explanatory factor is above or below the predicted 34-state average error rate based on that factor (holding other factors constant at their averages), and how far—how many percentage points—above or below the 34-state average each state's predicted error rate falls.

In using Table 18, a strong caveat is in order. Because the data in each column are based on a single explanatory factor, holding other factors constant at their averages, and because our results indicate that capital error rates are a function of several significant factors, and also because of the statistical methods used to generate the information there, Table 18 is most appropriately used to identify conditions in each state that pose a particularly large risk of serious capital error and might be an important target of reform efforts there. No single column in Table 18, nor the table as a whole, may appropriately be used to assign a particular overall predicted reversal rate to a given state.

Column A in Table 18, p. 344 below, compares states' predicted risk of capital error based on their capital-sentencing rates, holding other factors constant at their averages. Based only on states' death-sentencing rates—and with the above caveat in mind—Analysis 1A indicates that:

  • The states with the highest weighted average number of death sentences per 1000 homicides—Idaho and Delaware—are at risk of capital error rates 23 percentage points higher than the 34-state average, and as much as 45 percentage points higher than the error rates predicted for the lowest death-sentencing states.

  • States in the next cohort in terms of their risk of serious capital error given their high death-sentencing rates are Utah, Wyoming, Nevada, Oregon, Oklahoma and Arizona—which are at risk of capital error rates from 10 to 18 percentage points above the predicted 34-state average rate.

  • Three states prominently associated with the death penalty in the public mind because of their high numbers of executions—Texas, Virginia and Louisiana—face a lower risk of error based on this factor—suggesting that their relative success in carrying out the death verdicts they impose may be due in part to their comparatively low death-sentencing rates and thus their lower expected reversal rates. ("Success" in this regard is only relative, however, given that no state carried out even 30% of its verdicts during the study period, and the national average was 5%.715)

Table 18: states' Rank, and Difference from Predicted 34-State Average Error Rate, Based on Six Explanatory Factors, Holding Other Factors at the 34-State Average*

A

B

C

State

Death-Sentencing Rate Per 1000 Homicides

Proportion of Blacks in State Population

Homicide Risk to Whites Relative to Blacks

Rank

Value

Difference from 34-State Avg. Error Rate

Rank

Value

Difference from 34-

State Avg. Error Rate

Rank

Value

Difference from 34-

State Avg. Error Rate

Connecticut

34

5.5

-21.6%

19

8.3

-6.9%

21

.170

-2.7%

Kentucky

22

16.9

-7.0%

23

7.1

-8.7%

5

.280

+1.2%

Maryland

27

14.2

-9.6%

6

23.7

+5.9%

20

.190

+2.2%

Tennessee

21

18.4

-5.7%

11

15.7

+0.7%

18

.230

-0.4%

Mississippi

15

27.0

+0.2%

1

35.2

+11.0%

6

.270

+0.9%

Oregon

6

53.2

+11.1%

30

1.6

-22.2%

23

.170

-2.9%

California

31

10.1

-14.3%

21

7.5

-8.2%

9

.250

+0.3%

New Jersey

26

14.2

-9.6%

14

12.9

-1.7%

24

.170

-3.0%

Idaho

2

113.8

+22.9%

33

.6

-28.2%

32

.070

-9.1%

Montana

9

43.5

+7.9%

34

.5

-29.5%

33

.001

-29.4%

Georgia

18

24.3

-1.4%

4

26.6

+7.3%

12

.250

+0.2%

Arizona

8

47.7

+9.4%

27

3.0

-17.3%

10

.250

+0.3%

Alabama

11

38.9

+6.0%

5

25.3

+6.7%

13

.240

-0.1%

Colorado

32

8.5

-16.4%

25

3.9

-14.9%

11

.250

+0.3%

Washington

33

5.8

-20.9%

28

3.0

-17.4%

22

.170

-2.7%

Wyoming

4

64.0

+14.1%

31

1.2

-23.9%

34

.001

-29.4%

Florida

12

32.7

+3.3%

13

13.7

-1.0%

17

.230

-0.3%

Oklahoma

7

49.7

+10.0%

22

7.2

-8.6%

8

.260

+0.7%

Indiana

23

16.6

-7.3%

20

7.7

-7.9%

29

.130

-5.1%

Arkansas

17

25.3

-0.8%

10

16.2

+1.1%

19

.210

-1.1%

North Carolina

16

26.9

+0.1%

7

22.0

+5.0%

7

.270

+0.8%

Nebraska

10

41.4

+7.1%

26

3.2

-16.8%

31

.080

-8.8%

Nevada

5

55.4

+11.8%

24

6.6

-9.5%

14

.230

-0.2%

South Carolina

13

28.0

+0.8%

3

29.8

+8.8%

2

.340

+2.8%

Utah

3

81.7

+17.9%

32

.7

-27.3%

28

.130

-5.1%

Louisiana

24

15.0

-7.9%

2

29.8

+8.8%

16

.230

-0.3%

Illinois

28

14.0

-9.8%

12

14.7

-0.2%

25

.150

-3.8%

Pennsylvania

14

27.5

+0.5%

18

9.1

-5.9%

27

.140

-4.6%

Texas

25

15.8

-8.0%

15

11.9

-2.7%

3

.330

+2.8%

Missouri

20

19.0

-5.3%

16

10.7

-4.0%

30

.120

-5.4%

Delaware

1

116.4

+23.2%

9

16.7

+1.4%

4

.280

+1.2%

New Mexico

30

12.0

-11.9%

29

2.1

-20.2%

1

.590

+7.7%

Ohio

19

23.9

-1.7%

17

10.5

-4.2%

26

.150

-4.0%

Virginia

29

13.4

-10.3%

8

18.8

+3.0%

15

.230

-0.2%


* Data on all explanatory factors in this table are based on Analysis 1A, except for the data on per capita spending on state courts, which are based on Analysis 3A.

2. High county-level capital-sentencing and high capital-error rates.

The above discussion focuses on state differences in capital-sentencing and capital-error rates. Similar disparities exist at the county level. Most counties in most active capital states imposed no death verdicts in particular study years.716 Those localities may be contrasted with the six American cities that imposed over 100 death verdicts, and the nine additional cities that imposed between 50 and 100 verdicts, during the study period—listed in Table 19 below in order of death-sentencing rates, to show the wide variation even among high death-sentencing cities. Even here, however, the influence of states is felt. The top city in each cohort is in Arizona. Five of the top 15 death-sentencing localities measured in this way are located in Florida.

Table 19: Cities with More than 100, and with 50-100, Death Verdicts, 1973-1995,
by Death Sentencing Rate per 1000 Homicides
717

City # Death Verdicts Rate/100 Homicides
Phoenix (AZ) 114 41
Philadelphia (PA) 127 27
Houston (TX) 190 19
Miami (FL) 103 15
Chicago (IL) 138 11
Los Angeles (CA) 150 8
Tucson (AZ) 63 64
Las Vegas (NV) 71 55
St. Petersburg (FL) 51 50
Oklahoma City (OK) 68 50
Tampa (FL) 67 36
Jacksonville (FL) 66 30
Birmingham (AL) 55 25
Ft. Lauderdale (FL) 55 21
Dallas (TX) 61 11

Variation in county capital-sentencing rates is the rule, not the exception. Death-sentencing rates for counties with five or more death verdicts during the study period718 ranged from:

  • 0 per 1000 homicides in, e.g., Denver (0 out of 1057 homicides) and Baltimore City (0 out of 2933 homicides);

  • 3 per 1000 in St. Louis City, Shreveport, and Dayton;

  • 4 per 1000 in Newark (NJ) and Atlanta; and

  • 5 per 1000 in San Francisco and Richmond (VA);

to:

  • 40 to 49 per 1000 in Phoenix, Cincinnati, Montgomery (AL), Columbus (MS), DuPage County (IL), and four Florida counties;719

  • 50 to 59 per 1000 in Oklahoma City, Las Vegas, Reno, suburban Baltimore County and eight Florida counties;720

  • 60 to 75 per 1000 in Tucson, two other Arizona counties, and five additional Florida counties;721

  • 90 to 200 per 1000 in Kent County (DE), Lexington County (part of Columbia, SC), Randall County (part of Amarillo, TX), Coos Bay (OR), Carson City (NV), six Georgia counties, five Alabama counties, one additional Arizona county and four additional Florida counties;722 and

  • 267 per 1000 homicides in Missouri's capital, Jefferson City.723

Included on this list of high capital-sentencing counties are Nevada's three most populous counties with nearly 90% of the state's population, five of Arizona's six most populous counties with 85% of its population, and 21 of Florida's 67 counties with over a quarter of its population.724

As recent commentaries have highlighted, these and other death-sentencing disparities from one locality to the next often occur within the same state.725 Examples are in Table 20 below (sources: DRCen, Vital Statistics).

Table 20: Examples of High and Low Death-Sentencing Counties in the Same State
(Death Verdict per 1000 Homicides Indicated in Parentheses)

  Relatively High Death-Sentencing
City/County
vs. Realtively Low Death-Sentencing
City/County
California Redding/Shasta (62)
Modesto/Stanislaus (35)
Bakersfield/Kern (23)
San Francisco (5)
Los Angeles (8)
Richmond.Contra Costa (9)
Florida Pensacola/Escambia (55)
St. Petersburg/Pinellas (50)
Tampa/Hillsborough (36)
Palm Beach (12)
Miami Dade (15)
Gainesville/Alachua (15)
Georgia Atlanta suburbs/Gwinnett (47)
Atlanta suburbs/Cobb (36)
Columbus/Muscogee (33)
Atlanta/Fulton (4)
Augusta/Richmond (10
Macon/Bibb (13)
Maryland Baltimore suburbs/Baltimore County Baltimore City (0)
Washington suburbs/Prince George's County
Missouri Jefferson City/Cole (267)
St. Louis suburbs/Jefferson (46)
St Louis suburbs/St. Louis County (26
St. Louis City (3)
Kansas City (Jackson) (6)
Oklahoma Muskogee (52)
Oklahoma City (50)
Tulsa (16)
Ohio Akron/Summit (54)
Cincinnati/Hamilton (40)
Dayton/Montgomery (3)
Columbus/Franklin (16)
Oregon Coos Bay (94) Portland/Multnomah (13)
Pennsylvania Scranton/Lackawanna (76)
Philadelphia suburbs/Bucks (33)
Philadelphia (27)
Pittsburgh/Allegheny (12)
Philadelphia suburbs/Delaware (12)
So. Carolina Columbia (pt.)/Lexington (93)
Charleston (23)
Columbia (pt.)/Richland (9)
Greenville (11)
Tennessee Johnson City (pt.)/Washington (88)
Chatanooga/Hamilton (28)
Nashville/Davidson (6)
Texas Lubbock (20)
Corpus Christi/Nueces (20)
Houston/Harris (19)
Austin/Travis (10)
Dallas (11)
Galveston (11)
Virgina Danville City (53) Richmond (5)

As is true of state-level death-sentencing disparities, these county-level disparities are associated with county-level capital error rates. Our county case studies above726 and several of our regression analyses (Analyses 7-10 and 18)727 indicate that the more death verdicts per homicides a county imposes, the higher its capital-error rates are likely to rise. This county factor operates independently of, and in addition to, the effect of state death-sentencing rates.

3. Low or modest aggravation and a high case-level probability of reversal.

Analyses 1-5 and 7-18 and the county case studies thus lead to the conclusion that excessive use of the death penalty is associated with high rates of capital error. A final study, Analysis 19 of case-level federal habeas outcomes, helps answer a question this conclusion poses: Excessive by what measure? Given that the probability of error, reversals and retrials is decreased by less frequent, more judicious capital-sentencing, how should policy makers and officials go about narrowing the category of potentially capital cases?

Analysis 19 finds that the cases that present the greatest risk of federal habeas reversal, and thus that policy makers and officials would be best advised to exclude from death-eligibility, are those in which the degree of aggravation, offset by mitigation, is not high.728 As the case for death gets weaker—i.e., as aggravation net of mitigation or the quality of the evidence decreases—the probability of reversal due to serious error rises. Holding other factors at their average, Analysis 19 predicts that the probability of federal habeas reversal due to serious capital error decreases by 15% or more for each additional statutory or supplemental aggravating circumstance in the case, and increases by 15% for each additional mitigating factor in the case.729 As indicated by the decisions of federal habeas judges—and, on this common sense point, there is no reason to expect judges at other stages to evaluate serious capital error differently—uses of the death penalty are excessive, creating a high risk of serious capital error, when they extend the penalty to cases that are not very highly aggravated.

Our principal conclusion thus strongly supports the statements of District Attorney Marquis and Governor Gilmore quoted above: Jurisdictions that reserve the death penalty for only the very worst offenses do the best job of avoiding serious, capital error and the risks and costs that go with it. By contrast, states and counties that use the death penalty aggressively (i.e. relatively more often per every 10, 100 or 1000 homicides) and extend it to homicide offenses that are not extremely aggravated, are likely to have the worst records of serious, capital error.

Our analyses also indicate that the harmful effect of a propensity to overuse the death penalty in cases that are not highly aggravated occurs at the level where capital-sentencing policy is made, not where policy is applied. Federal habeas reversals are most common in close or marginal cases judged by the amount of aggravation net of mitigation—i.e., in non-highly aggravated cases that get swept into the capital net by broad death-sentencing policies—rather than in especially egregious cases where case-level pressures to sentence capitally might be highest.730 This suggests that it is state or local policies setting a low threshold of seriousness or aggravation for the kinds of crimes that trigger capital prosecutions and verdicts, and not pressures to use the death penalty in particular cases, that are most associated with high rates and amounts of error.

C. Supporting Conclusions

1. High capital-error rates are associated with four conditions that create pressure to use the death penalty in weakly aggravated cases where the risk of error is great—high crime rates, low punishment rates, race and politics.

For many policy purposes it is enough to conclude based on reliable and consistent study findings that heavy use of the death penalty is associated with high capital-error rates. But our regression analyses reveal four additional factors associated with high rates of serious capital error whose common attributes suggest something more about the forces leading to heavy capital sentencing and a high risk of error. High capital error rates are significantly related to:

  • well-founded doubts about the ability of state law enforcement policy and officials to deal effectively with crime;

  • state judges' susceptibility to being harmed politically, given how they are selected and promoted, if their rulings do not conform to popular sentiment;

  • the homicide risk to whites, particularly when that risk approaches or exceeds the high risk of homicide that African-Americans typically face; and

  • the size of the state's black community relative to its overall population (and to a lesser extent the proportion of its population receiving welfare).

As we develop below, each of these factors is a potential indicator of the threat of crime felt by politically influential members of the community, or of the pressure on capital policy makers and officials to respond forcefully to that threat. We conclude that each factor is an indicator of the pressure felt by capital jurisdictions and officials to respond to influential citizens' fear of serious crime by extending the death penalty to cases where its use is not warranted by the especially aggravated nature of the offense and instead invites serious error. After discussing each factor, we address attributes they share that invite the extension of the death penalty to weakly aggravated cases where the need to commit error to secure a death verdict is high.

a. Well-founded doubts about the ability of state law enforcement policy and officials to deal effectively with crime.

Main Analysis 1 and nearly all other analyses find that states which arrest, convict and punish fewer serious criminals (as indicated by the number of incarcerated criminals per 100 FBI Index Crimes) have significantly higher capital-error rates.731 This relationship is highly significant, and the size of its predicted effect on capital reversal rates is large. Typically, predicted capital reversal rates (holding other factors constant) increase 5- to 7-fold as rates of apprehending, convicting and imprisoning serious criminals fall from their highest to their lowest levels among states in our study.732 In the same way as poorly funded and overburdened court systems generate more serious capital error (as we discuss below733), ineffective state law enforcement systems—those with the worst records of arresting, convicting and incarcerating serious criminals—are the most likely to conduct seriously flawed investigations, prosecutions and trials of capitally charged defendants.

When considered with our principal finding, this result supports a further conclusion. The less effective law enforcement is at capturing, prosecuting and punishing criminals, the more pressure is likely to be placed on officials to do more to fight crime. This is especially the case when the crime that people and neighborhoods fear is homicide, and when those in fear have the political influence to translate their concerns into public action. One response such political pressure invites is expanded use of the death penalty as a visible demonstration of officials' intolerance for crime and their commitment to punishing it severely. Because expanding the death penalty costs little at first—although eventually it triggers lengthy appeals that often end in costly reversals and retrials—and because that response is available to any jurisdiction, no matter how poor its crime-fighting capacity may be, expanding the death penalty is an especially attractive response by states with the worst crime-fighting records. Where pressures generated by well-founded doubts about the effectiveness of state law enforcement systems trigger expanded death sentencing, our principal finding predicts that higher capital error rates will result as officials cast the capital net more widely, pulling in more cases where the evidence of a highly aggravated crime is weak.734 Lower crime-fighting competence thus is associated both with heightened pressures to expand the death penalty in response to ineffectively controlled crime, and with lower competence in investigating and prosecuting those progressively weaker capital cases. The mutually re-enforcing effect is the one our study documents: Higher rates and amounts of serious capital error.735

Column D in Table 18, p. 345 above, compares states based on their rates of arresting, convicting and incarcerating criminals per 100 FBI Index Crimes, and based on whether and by how much the capital reversal rates this factor predicts for each state diverge from the average reversal rate predicted for all 34 states. States with the lowest law enforcement scores and the highest risk of error considering only this factor are Utah, Montana and several other western states. Nebraska, Illinois and Florida round out the top 10 states with the highest predicted capital error rates based on this factor alone. According to our best analysis, states in this low-law-enforcement category risk capital error rates anywhere from 8 to 34 percentage points higher than the 34-state average—and 22 to 48 percentage points higher than the state with the best record in this one regard.

Comparing state risk rankings based on this factor to rankings based on high death-sentencing emphasizes the caveat given above.736 Although Colorado and Washington are at the high end of the spectrum of risk based on this low-law-enforcement factor, they are at the low end of the risk spectrum when it comes to their death-sentencing rates. The opposite is true of Delaware, Alabama, and Nevada, which have a relatively low risk of capital error judged by their law enforcement record, but a high risk of error based on their death-sentencing rates. Because our analyses reveal that all these factors are important, it is inappropriate to base an assessment of a state's overall proneness to capital error on state comparisons that are attentive to only one factor. What these figures instead identify are different high-risk factors for each state, which could become a focus of local reforms. Given that nearly all states have disturbingly high (50%-plus) overall capital error rates,737 all have room for improvement, whether or not they do comparatively well on one or another measure.

b. State judges' susceptibility to being harmed politically if their rulings do not conform to popular sentiments.

Another study finding identifies a political mechanism through which public fears about crime, and doubts about the effectiveness of a state's response to it, can pressure officials into adopting policies that increase capital error. This result is found in main Analysis 1, and in confirming analyses of all three review stages combined, and the state direct appeal and federal habeas stages by themselves.738 States, and counties in states, with judicial selection methods that make judges more vulnerable to political discipline if their rulings are not consistent with popular sentiment have higher capital-error rates.739 In other words, courts in states that directly elect judges from the outset—or subject judges to more frequent, more often contested and more partisan elections—more often produce seriously flawed capital verdicts than courts whose judges are insulated from direct political influence from voters and contributors.

This finding is important. It reveals a way in which politically influential members of the public who are threatened by serious crime and doubt the effectiveness of their state's response to it can pressure policy makers to demonstrate their resolve to respond the problem aggressively—including by extending the death penalty to more cases where the risk of error is greater. Judges, however, are not the only actors whose decisions affect the breadth of the state's death penalty. Governors, legislators, attorneys general and district attorneys also have an important impact on death-sentencing policy.740 Unfortunately, the effect of political pressures on those officials is harder to demonstrate statistically, because doing so requires measurable variation among states in the kinds of political pressure their officials feel, and there is little variation from state to state in how and how often they select governors, legislators, attorneys general and district attorneys.741 Thus, the sizeable effect of judicial selection techniques on capital error rates—2- to 6-fold increases in predicted error rates as selection methods change from placing the least to the most political pressure on state judges (other factors held constant)742— probably underestimates the effect of all types of political pressures on all capital officials.

Column E in Table 18, p. 345 above, compares states based on the amount of political pressure their judicial selection techniques put on state judges, and based on the difference between the reversal rate for each state that is predicted by this factor alone, and the average rate it predicts for all 34 states. Because there are only nine possible scores on the political pressure index—only eight of which actually apply to any of the 34 study states—a number of states are in a tie for most rankings.743 Only two states are tied with no other: Virginia, with the lowest rank on this risk factor, given that its judges are appointed,744 and New Mexico, with the highest rank. This top ranking in terms of the pressure on judges to conform their rulings to public sentiment puts New Mexico at risk of capital error rates 14 percentage points higher than the 34-state average. The 13 states with judicial selection techniques that place the next highest level of political pressures on their judges—including, for example, Alabama, Georgia, Oklahoma and Ohio—are at risk of capital error rates 8 percentage points above the 34-state average, based on this factor. On the other hand, judicial selection techniques that immunize state judges entirely from regular or potential elections by the public at large are associated with predicted capital reversal rates nearly 20 percentage below the 34-state average, and over 30 percentage points below the predicted reversal rates of states that put judges under the most pressure to conform their rulings to popular sentiment.

c. A high risk of homicide to politically influential citizens.

By taking each state's homicide rate among whites and dividing it by the state's homicide rate among blacks, it is possible to determine whether—and how closely—the homicide risk to whites in each state approaches the typically high homicide rates that afflict African-Americans communities in this nation. Put another way, this factor compares states based on whether homicides there mainly threaten blacks, or whether the homicide risk also falls fairly heavily on whites.745

In main Analysis 1, and in most other analyses, the greater the share of the homicide risk that is borne by whites relative to blacks, the higher the state's rate of serious capital error.746 Effect size is moderate. Holding other factors at their averages, predicted reversal rates double or triple across the spectrum of conditions among states and years in our study.747 Likewise, Column C of Table 18, p. 344 above, shows that in our best analysis this factor predicts capital reversal rates for New Mexico that are 17 percentage points higher than the predicted reversal rate for Nebraska, given that in New Mexico the risk that a white person will be killed by homicide comes the closest to equaling the risk that a black person will be killed by homicide (the white risk is 60% of the black risk), while in Nebraska the homicide risk faced by whites is only 8% as high as the risk faced by blacks.748 At p. 365 below, we explain why the share of the homicide risk borne by whites as opposed to blacks may have an even bigger predicted impact on reversal rates, when the interaction of that factor and the racial makeup of the general population is considered.

In a minority of analyses, high homicide rates by themselves are significantly associated with high error rates, over and above the effect of a high homicide risk to whites relative to blacks.749 In some other analyses, homicide rates by themselves were significantly associated with error rates until the white-compared-to-black homicide rate was introduced, at which point the white/black homicide rate was significant (and fit and other diagnostic measures improved), and homicide rates by themselves became non-significant. Similarly, in nearly all analyses, the homicide rate exclusively among whites was not as powerful a predictor of error rates as the homicide threat to whites compared to blacks.750 This reveals that, although high homicide rates by themselves predict high capital error rates, a better predictor of high error rates is the distribution of the risk of homicide between whites and blacks—more specifically, whether the homicide risk to whites approaches or surpasses that to blacks (in which case capital error rates are higher), or on the other hand, whether blacks bear the brunt of the homicide risk (in which case capital error rates are lower).

We included this factor based on strong evidence in a number of studies, and the recent conclusions of two highly regarded legal scholars representing a wide spectrum of political views, that law enforcement officials are more responsive to the threat of crime to white as opposed to black communities.751 These observers offer two explanations for their findings. The first is that law enforcement officials and policy makers pay more attention to the law enforcement needs of affluent and politically influential people and communities, and less attention to people and communities with fewer resources and political influence, because the latter groups are less organized, have fewer resources and less time to devote to the civic and political mobilization needed to secure the attention of law enforcement officials or to fund contributions to political campaigns, and have lower social status. In this view, African-American communities are one of a number of communities that tend on average to be less organized and wealthy and to have lower status with officials, and thus are less well served by law enforcement policy and officials. Because reliable data are kept on the race of crime victims, but not always on other indicators of low political influence, it is easier to detect and measure under-enforcement of the criminal laws in the black than in other, similar communities.

The other explanation is that race discrimination leads officials to pay less attention to the threat of crime to blacks as opposed to whites, explaining why the race of victims strongly predicts how well they are served by law enforcement policies and officials. There is substance to both explanations. For our purposes it is unnecessary to choose between them.

A central finding of these prior studies is that, after controlling for degree of aggravation and other variables, death verdicts are substantially more likely for homicides against white victims than for those against black victims.752 This finding predicts that jurisdictions with a relatively large homicide risk to whites, or to members of other influential communities that tend to get more law enforcement attention, are likely to have higher per-homicide rates of capital prosecution and sentencing. But why would states with a relatively high homicide risk to whites have significantly higher rates of serious error in those verdicts?

Our study's principal finding suggests and answer to this question: Jurisdictions that use the death penalty more often per homicide have higher capital error rates. The strong association between high error rates and greater use of the death penalty predicts that conditions prompting aggressive use of the death penalty may also be associated with high error rates.753 This, then, helps explain why states in which a relatively heavy share of the homicide risk is borne by whites as well as blacks have higher capital error rates. The greater the share of the homicide threat borne by whites or other politically influential communities, the more pressure officials may feel to broaden the death penalty to demonstrate a resolve to deal forcefully with homicides. Resolve is just as vividly demonstrated when the death penalty is used for weakly aggravated homicides as when it is limited to highly aggravated cases—indeed, it may be more vividly demonstrated when aggravation is weak. And in any event, in any given jurisdiction, there are likely to be many more medium-range than extremely aggravated cases through which to demonstrate a determination to fight crime. Expanded capital sentencing in response to crime fears thus invites capital verdicts in weakly aggravated cases where the probability of serious error is the greatest.

A homicide risk that is not borne almost entirely by blacks, and also falls fairly heavily on whites, thus appears to pressure officials to set a low threshold on when the death penalty can be imposed. Low capital thresholds in turn prompt high capital error rates, by inviting prosecutions where the offense is not "the worst of the worst"—where the evidence of an offense warranting the death penalty is weak enough that corner-cutting and other errors may be needed to assure a death verdict.

There is one sense in which our study qualifies the conventional wisdom about the link between race and the death penalty. The conventional understanding might suggest that, given the link some studies have found between the race of the victim of a particular murder and an increased probability of a death sentence, our finding of a link between higher death-sentencing rates and higher error rates would lead to higher error rates in death verdicts imposed for homicides against white victims. As we show above, however, capital error occurs just as often in black-victim as in white-victim cases.754 This is part of a pattern of results indicating that high capital-error rates are mainly associated with broad capital-sentencing policies, not individual decisions in particular (e.g., white-victim, or especially aggravated) cases.755 Once factors like high concentrations of homicides in politically influential communities lead to aggressive capital laws and policies, those policies—and associated increases in capital error—evidently affect defendants of all races equally. The people most adversely affected by broad capital-sentencing policies and resulting error thus are defendants of all races who happen to be tried in jurisdictions with high death-sentencing rates, and particularly defendants of all races as to whom the evidence of an offense warranting the death penalty is the weakest.756

The results discussed here and in the previous section have a further implication. As a matter of principle, law enforcement officials must do everything the law permits to lessen the threat of homicide to all residents of the jurisdiction. Our regression results reveal that expanded use of the death penalty against an ever-widening set of homicides is not an effective strategy because it increases the likelihood of mistake, including that innocent people are caught in the net and perpetrators go free. Nor is it a strategy the law permits, because it multiplies reversible capital error. Nor, finally, is it a strategy designed to protect all communities because it is more responsive to concentrations of homicide in the white community. The results in the previous section reveal an alternative strategy for lowering the homicide threat that is an effective response to crime, is permitted by law, and protects all communities. Rather than applying the death penalty to an ever-expanding set of arrested suspects for whom the evidence of an offense aggravated enough to warrant the death penalty is fairly weak, the better strategy is to leave the death penalty focused on "the worst of the worst" and to divert the resources saved by a more judicious use of the death penalty to apprehending, convicting and incarcerating a wider array of perpetrators of a broader set of serious crimes.

d. Large numbers of African-Americans and welfare recipients.

In main Analysis 1, and nearly all supporting analyses, the larger the proportion of a state's population that is African-American, the larger the state's rate of serious capital error.757 At the federal habeas stage, the same thing is true of the proportion of the state's population receiving welfare and its per capita cost.758 Effect size is considerable. In our main analysis, predicted capital error rates more than quadruple as the size of the black population rises from its lowest to its highest levels among states in our study, holding other factors constant.759 Likewise, in our federal habeas regression, predicted reversal rates more than quadruple as welfare recipients and costs rise from their lowest to highest levels among study states and year.760

We explain above why there is no clear link between the proportion of blacks in the state population and the number of black state policy makers, judges, prosecutors, jurors and the like, and why those conditions are unlikely to explain high capital error rates.761 Instead, given that the explanatory condition is the racial makeup of the state's overall population, not that of participants at particular trials or even of the county where the crime and trial took place,762 and given extensive research documenting powerful, inaccurate stereotypes linking contact with African-Americans to a perceived threat of violent crime,763 we conclude that the size of a state's African-American population is a strong indicator of the intensity of crime fears among politically influential citizens. Like the race of homicide victims discussed just above, this racial factor is a powerful indicator of the pressure officials face to respond forcefully to crime. This explains why the factor strongly predicts high capital-error rates, which are strongly associated with the broad and indiscriminate use of the death penalty that can occur when officials face pressure to expand the penalty as a forceful demonstration of their resolve to fight crime.

As we note above, the problem is not with officials who are determined to fight crime.764 The problem is with expanded and indiscriminate use of the death penalty, which is not an effective solution to the problem. When that response is adopted, the result is not more successful law enforcement, but instead a greatly increased risk of serious capital mistake, reversal and costly retrials. At the extreme—as has demonstrably occurred on just short of 100 occasions in the modern death-sentencing era—it means convicting the innocent, while actual killers remain at large.765

We reach this conclusion sadly, given what it suggests about race relations. But we reach it with confidence. To begin with, the conclusion follows from those above. Higher death-sentencing rates are associated with higher capital error rates—with the biggest risk factor being the indiscriminate extension of the penalty to cases where aggravation levels are not extremely elevated. And high error rates are linked to two indicators of crime fears among politically influential individuals that can pressure officials to extend the death penalty to weakly aggravated cases as a way of demonstrating a firm resolve to fight crime: (1) low rates of apprehension, conviction and incarceration of serious criminals, and (2) a high risk of homicide borne by whites as well as blacks. It is an unfortunate but demonstrated fact that the race of people in the community is yet another, powerful indicator of crime fears, given the association people report and display between the race of people they encounter and a perceived threat of violent crime.766 This association is partly based on actual crime and homicide rates, which are higher among African-American and poor communities than among others.767 But as the literature demonstrates, the association is also due to stereotypes that lead people to greatly overestimate the threat of cross-racial violent crime.768 (In fact, most crime occurs among members of the same race, community and class.769) Our analyses provide important new evidence of this effect. When examined separately, higher homicide rates indeed have the same relationship to higher reversal rates as our two racial measures of the actual and perceived threat of homicide. But when all these factors are examined together, it is the racial measures and not homicide rates themselves that are significantly and powerfully related to serious capital error. The condition related to pressure to use the death penalty that most strongly predicts high capital error rates thus is not the actual threat of homicide (the homicide rate), but instead the perceived as well as actual threat of homicide to whites and other influential residents from African-Americans and poor people.770 Given the linkage between the size of the black (and the poor) population and the perceived threat of crime, and given our consistent finding that indicators of crime fears predict high rates of capital error, it is reasonable to explain the strong association between capital error rates and the size of the black (and poor) population as another instance of the effect on capital error rates of the real and perceived threat of crime.

Second, our analyses of factors that increase the risk of capital error reveal that the size of the black population is significantly connected to two other recognized indicators of the intensity of fears of crime, particularly among politically influential citizens. We already have noted the relationship between homicide rates and the relative size of the black population as potential explanations for reversal rates. Those two factors are correlated, given the relatively higher rate of homicide committed by blacks than by whites. And tested separately, both factors are significantly associated with capital-error rates. But when tested together, the size of the black population remains a powerful predictor of capital error rates, while the homicide rate is no longer significant.771 From this we conclude that it is not so much the actual rates of homicide as a perceived threat of homicides by blacks that is associated with higher capital-error rates.

High African-American populations also interact with another established indicator of crime fears among politically influential citizens—the distribution of homicide risk between whites and blacks. In main Analysis 1, and in several other analyses of state and county error rates, states with a combination of homicide risks concentrated relatively heavily on whites compared to blacks and large black populations relative to the total population had significantly higher capital error rates than either of the two factors by itself or the two together would predict.772 This indicates that the two factors have a similar effect on reversal rates that is magnified when both are present. Given a strong consensus about the pressure the threat of crime to the white community puts on law enforcement officials to respond forcefully to crime,773 and given the interaction of that factor and the relative size of the black population, it is reasonable to understand all three effects (each factor by itself and the two together) as indicators of crime fears that put pressure on officials to broaden the availability of the death penalty, and in the process increase capital error rates.

Column B in Table 18, p. 344 above, ranks states based on their weighted average proportion during the study period of residents who were African-American. Column B then compares states based on the difference between the reversal rate predicted for each state, and the 34-state average predicted reversal rate, based on this factor alone, holding other factors at their average. As Column B shows, main Analysis 1A predicts that states with large black populations such as Mississippi and South Carolina are at risk of capital error rates over 10 percentage points higher than the average predicted reversal rate, and as much as 40 percentage points higher than predicted reversal rates in states with low African-American populations.

As we just pointed out, in main Analysis 1A and in a majority of others, the explanation for high reversal rates based on the racial makeup of the general population, and the separate explanation based on the racial makeup of homicide victims, interact: States where blacks make up a higher proportion of the population and where the homicide risk to whites comes the closest to equaling the (typically higher) homicide risk to blacks have an especially high risk of serious capital error. Predicted reversal rates cannot reliably be calculated for interaction effects of this sort, but the states may be ranked based on their comparative risk from this factor, as is done in the accompanying note. States that are most at risk from this factor, holding others constant at their average, are South Carolina, Mississippi, Louisiana, Georgia, Alabama, North Carolina, Delaware, Maryland, Virginia and Texas. Appropriately assessing the risk from each of the two racial factors requires that the risk from the interaction of the two also be considered.774

* * * * *

States for which these racial factors create an especially high risk of serious capital error cannot very well change their demographic profile, and thus may wonder how they can reduce the risk of error. As we develop above, however, it is not the demographic realities, but the pressures they create to apply the death penalty broadly, including in cases that are not highly aggravated, that appear to be linked to a high risk of serious capital error. And as we develop in Part VIII below, therefore, there are ways in which capital-sentencing policy may be changed to decrease the incentive and capacity to impose the death penalty in cases that are not highly aggravated where the risk of error is great. The main point for now is to catalogue the risk factors for each state as a prelude to the policy discussion below.

e. Summary: conditions with a common capacity to pressure policy makers to extend the death penalty to cases that are not highly aggravated, where the risk of error is great.

The four factors discussed here have two common attributes which explain why they invite policies that extend the death penalty to cases that are not highly aggravated, where the risk of error is high. First, the fears and pressures the four factors create seem to operate at the level at which state and county capital-sentencing policy is madeC—i.e., where the threshold level of aggravation sufficient to trigger a capital prosecution and sentence is set for all cases—rather than at the level where policy is applied to particular cases.775 The higher the level of government at which policy is set, and the broader and more divorced the decision is from particular cases, the less likely it is that the policy will be sensitive to the nuances of aggravating and mitigating circumstances in individual cases, and the greater the chance that the policy will encompass less aggravated cases.

Second, all four conditions reflect either generalized fears about serious crime, or the capital system's vulnerability to pressures generated by such fears. Some of the fears and pressures are empirically well-founded—those based on high homicide rates and low rates of apprehending and punishing criminals. Others are less justifiable, or even illegitimate—the influence of political considerations on judicial outcomes, and the role of race in gauging the threat of crime. What is crucial, however, is that all four factors prompt fears and pressures that are far removed from the facts and circumstances of each case and invite responses—including broadened use of the death penalty—that demonstrate officials' intolerance for crime in general, and not just for offenses where close inspection of the circumstances and evidence reveal high levels of aggravation. Particularly in states with poor crime-fighting records, a desire to demonstrate a determination to fight crime is no less well-served—and may even be better served—by a threshold level of evidence and aggravation for the death penalty that sweeps in marginal cases where the evidence is weak and where as a result the risk of error is large.

Our findings indicate that it is not every additional use of the death penalty, but only those uses where the crime is not "the worst of the worst," that especially enhance the risk of serious capital error. The four factors discussed here encourage this indiscriminate use of the penalty. So may other conditions that are harder to measure, such as political pressure on district attorneys.776 The relationship between high death-sentencing rates and high capital-error rates thus serves as a residual explanation for capital error rates, which captures the effect of pressures to use the death penalty broadly that, unlike the four pressures discussed here, cannot be measured more directly.777

2. Aggressive use of the death penalty is also linked to heavy court congestion and delay.

Main Analysis 1 and most supporting analyses find a significant relationship between high numbers of capital verdicts awaiting appeal and low rates of progress in moving capital verdicts through the system either to approval and execution, or reversal.778 Effect size is large. Analysis 1 predicts that the process of moving capital verdicts from trial to a decisive result on appeal essentially comes to a halt in states with 20 or more capital verdicts awaiting review at one time.779

This finding is predictable: Capital verdicts caught in the review process cannot serve the purpose for which they were imposed—and those that are flawed cannot be corrected. The findings have added significance in conjunction with our principal finding that higher death-sentencing rates lead to higher rates of serious capital error. Higher rates of death verdicts also mean more death verdicts, each of which makes an inordinate contribution to court congestion, and even a fairly small number of which can effectively clog and close down the system.780 States with fewer death verdicts not only limit the risk that any verdict will be found seriously flawed, but also increase the probability that verdicts that are not flawed will get through the review process quickly.

The table in note 788 below compares states based on their weighted average number of death verdicts awaiting review at one of the three review stages during the study period.781 States vary substantially in this regard, from California with an average of about 27 capital verdicts awaiting review each year, and Texas, Florida, Pennsylvania and Ohio with average capital backlogs of 12 to 18, to Nebraska, Montana, Washington, Connecticut and Wyoming, with fewer than 1 backlogged capital case on average.

Consideration of this factor reveals a hidden cost of the current capital system. Delayed appeals limit the amount of completed review, generating lower numbers of reversals.782 Delayed appeals also lead to lower rates of reversal. First, when reversal rates are calculated as proportions of all imposed verdicts, lower rates of review automatically mean lower reversal rates—even if verdicts remain equally flawed—because there are fewer outcomes of any sort.783 Although that rate is not the true error rate, which is the number of reversals as a proportion of reviewed, not imposed, verdicts,784 members of the public sometimes mistakenly think that fewer reversals per imposed verdicts means fewer errors.785 Second, reversals take a year or two longer than affirmances to occur at the federal habeas stage, artificially increasing the number of affirmances and decreasing the number of reversals that have occurred as of any moment, which in turn artificially decreases the error rate.786 Third, our regression results suggest that large backlogs of delayed appeals sometimes pressure appellate courts into approving verdicts that otherwise would be found seriously flawed, further lowering reversal and error rates.787 This means that states like California, Texas, Florida, Pennsylvania and Ohio, have fewer reversals and lower reversal rates (as a proportion of imposed verdicts)—and appear to have lower error rates (as a proportion of reviewed verdicts) than otherwise would be true—because capital verdicts move so slowly through their appeals process. From the perspective of these states' reversal records, their inefficiency becomes a saving grace because it lowers their numbers and rates of reversals. But from the perspective of victims and communities seeking finality, taxpayers financing costly appeals, and wrongly convicted and sentenced defendants needing redress, that inefficiency is costly.

The reverse holds for states like Nebraska, Montana, Washington and Connecticut. They are penalized for having efficient review systems: Although their reversal records accurately reflect the amount of error in their capital verdicts, those records are comparatively worse than the records of states like California, Texas and Florida, where delay artificially deflates reversals. Based on this factor alone, holding other factors at their averages, our regression analyses predict very high reversal rates for Nebraska, Montana, Washington, and Connecticut. But that prediction is based entirely on these states' admirably low backlogs of pending appeals, which keep them from taking advantage of delayed appeals to obscure their true error rates.788

3. Overburdened and underfunded courts are associated with a high risk of capital error.

In main Analysis 1, and in most other analyses of capital error found at the three review stages combined, a combination of high numbers of capital verdicts awaiting review and high per capita rates of court cases of all types awaiting decision is significantly related to high capital error rates.789 In analyses of the initial, direct appeal review stage—where nearly 80% of capital reversals occur—low per capita funding on the courts is also related to high capital error rates.790 For states with below average funding for their courts, effect size is large: Relatively small decreases in direct funding below the 34-state average of about $1.80 per capita are associated with steep predicted increases in the amount of serious capital error state high courts discover on direct appeal, holding other factors constant.791

These findings indicate that state court systems with below average operating budgets—or what may be the same thing, with too many capital and non-capital cases to process reliably with available resources—tend to produce more flawed capital verdicts. High proportions of flawed verdicts and the high reversal rates associated with them lead, in turn, to high retrial rates— further burdening the courts, and generating more error, more work for appellate courts, and more reversals and retrials.

Results of particular cases reveal the same thing. At the two phases of review where data are available, the largest single reason why courts reverse capital verdicts is egregiously incompetent representation of capital defendants by mainly state-funded lawyers—prompting close to 40% of all state post-conviction reversals, and close to 30% of all federal habeas reversals.792 The main reason inexperienced, unskilled and untrained lawyers are often the only ones who seek capital trial assignments—the most demanding assignments lawyers can receive—and the main reason the performance of even conscientious appointed capital lawyers is often below par, is the low level of compensation and reimbursement for expenses (investigators, mental health exams, DNA testing and the like) that is available in most states.793 Because funds for capital trial lawyers and for necessary support services often come out of state court operating budgets, it is not surprising that our aggregate-level analyses reveal a link between financially strapped state courts and high rates of capital error.

Case-level Analysis 19 of federal habeas outcomes also reveals a link between poor quality state court proceedings and high capital reversal rates. State court denials of evidentiary hearings on review of claimed capital errors are associated with a higher probability that federal habeas courts will reverse capital verdicts.794 One reason state courts decline to hold hearings is that they cannot afford the accompanying costs: reimbursement of counsel for indigent prisoners, witness and court reporter fees, and salaries for judges, court clerks and security personnel.

Resources available for capital trials are a function of two conditions: the funds and personnel available to process capital cases, and the number of cases to be processed. This explains why high rates of serious capital error are linked to low funding for capital courts and high numbers of capital and other cases to process. This in turn reveals how closely this supporting conclusion is tied to our principal conclusion: More capital prosecutions and sentences lead to more strain on the system, more delay and more serious error.

Column F of Table 18, p. 346 above, compares states based on their weighted average direct expenditures on their court systems.795 States vary substantially in this latter regard, from less than $1 of direct court funding per capita on average in Nebraska, Utah and Georgia, to over $3 of court funding per capita on average in Connecticut. Although as we note above, this explanatory factor has only a modest effect for differences in spending levels at or above the 34-state average, below-average funding of courts adds as many as 10 percentage points to predicted capital reversal rates, holding other factors constant.796

Our measure of the effect of high backlogs of capital and non-capital cases awaiting disposition by the courts is an "interaction" effect for which predicted reversal rates cannot be accurately calculated. The states, however, can be compared based on the extent to which the combination of high capital and non-capital caseloads increases their risk of serious capital error. That comparison, in the attached note, reveals that this factor poses an especially high risk to five states: Texas, Illinois and Pennsylvania and especially California and Florida.797

4. Controlling for other factors, more recent death verdicts are much more likely to be reversed on state direct appeal than earlier verdicts; there is no reliable evidence that the quality of death verdicts has improved much since the early 1980s.

Figures 2A and 2B, pp. 55-56 above, reveal that after fluctuating in the 1970s, capital reversal rates for the three review stages combined were high (50%- or 60%-plus) and fairly stable from the early 1980s through the end of the study period. Those charts plus Figures 2C-3B, pp. 57-58 and 60-61 above, and a figure in our earlier Report, reveal the same stability from the early 1980s forward for direct appeal and federal habeas reversal rates but suggest that state post-conviction reversal rates may have risen somewhat in that period.798 Our regression analyses ask a different question about changes over time: Beyond the effect of other significant factors, have error rates increased or decreased in a statistically significant way during the study period? What this inquiry measures is the influence of forces that are not captured by specific explanatory factors in the analysis but whose effect is time-sensitive and thus is registered by a general measure of patterns of change over time. The question this factor poses is whether forces other than those captured by the specific explanatory factors in the analysis drove reversal rates higher or lower than they would have been had the specific factors been the only ones at work.

Our first conclusion is that in all of our analyses that calculate reversal rates as a proportion of imposed, rather than reviewed, death verdicts, a force with a downward effect on reversal rates over time is at work. That force, however, is not related to changing amounts of error over time, but to changing amounts of unfinished appeals. Appeals that were not completed as of the end of the study period artificially depress reversal rates, because fewer finished appeals means fewer outcomes of any sort, including reversals, as a proportion of imposed verdicts.799 Because the later a death verdict was imposed, the more likely it is that the verdict did not finish being reviewed by the end of the study period, later verdicts are automatically associated with lower reversal rates as a proportion of imposed death verdicts. Because this relationship between later verdicts and lower reversal rates holds true for flawed, as well as unflawed, capital verdicts—the relationship is sensitive to whether review occurred, not whether flaws were discovered when it occurred—the use of time trend as an explanatory factor nicely controls for the effect of delay (unfinished appeals),800 but does not gauge changing rates of error over time.801

The downward influence of delay on reversal rates over time is exacerbated in federal habeas cases where reversals due to serious error take longer to occur than affirmances.802 As a result, flawed verdicts are under-represented among verdicts finally reviewed by the study end date, and over-represented among verdicts remaining to be finally reviewed on that date, with the bias affecting later cohorts of verdicts more than earlier ones, because higher proportions of later verdicts were still awaiting final review as of the study's end date.

These delay-driven biases against counting reversals and (in the latter case) in favor of counting affirmances guide the interpretation of significant changes in reversal rates over time:

  • In analyses calculating reversal rates as proportions of imposed verdicts in which reversal rates decline over time, the result cannot be interpreted with any precision. We know that at least some of the decline is due to the delay-related, error-neutral effects just described. But we cannot say how much of the decline is attributable to delay. It could be that, apart from the effect of other factors, improvements in the quality of death verdicts are also causing reversal rates to decline over time—adding an error-related decline in reversal rates on top of the delay-related decline just discussed. But it could just as easily be that, after accounting for other factors, later verdicts were actually more flawed than earlier ones—thus counteracting some of the delay-driven decline in reversal rates that otherwise would have appeared. Thus:

    • When reversal rates calculated as a proportion of imposed verdicts drop significantly over time, it is impossible to determine whether that drop is entirely delay-related or is also affected by changes in error over time.

    • On the other hand, in analyses of reversal rates calculated as proportions of imposed verdicts in which reversal rates do not drop significantly over time, it is likely that an increase in error over time (after controlling for other factors) has occurred. In that event, it is only increasing error rates over time (after accounting for other factors) that can have counteracted the delay-related biases that otherwise would have caused reversal rates to decline significantly over time.

  • Declining reversal rates at the federal habeas stage are also difficult to interpret. At least some part of that decline is due to systematically longer delays in federal habeas review of flawed verdicts than in habeas review of unflawed verdicts. This again makes it impossible to tell whether error-related decreases or increases in flawed verdicts reaching that stage are adding to or counteracting the delay-related decline.

  • Analyses of relationships between later verdicts and reversal rates calculated as proportions of reviewed (as opposed to imposed) death verdicts at review stages other than the habeas stage are subject to no delay-related biases. Reversal rates in these analyses are not sensitive to delay because delay affects the base number of death verdicts (the number reviewed) as much as the number reversed.803 Nor do flawed verdicts take systematically more or less time to be reviewed at stages other than the federal habeas stage. As a result, any changes in reversal rates over time that these analyses find are reliable indications of the size and direction of changes in error rates that are not captured by other factors in the analysis.

Analyzed under these guidelines, our analyses reveal the following:

  • After the effect of all other factors on error rates is accounted for, state high court judges on direct appeal found substantially higher rates of serious, reversible error in recent death verdicts than in earlier ones. Analyses 3, 4 and 10 reliably evaluate the relationship between the year death verdicts were imposed and the amount of serious reversible error found at the state direct appeal stage, without any delay-related bias. All three analyses find that, after accounting for other important factors, the later a death verdict was imposed, the higher the probability that it would be reversed on state direct appeal based on a finding of serious error. The result is highly significant, and the upward effect on reversal rates of a verdict's having been imposed later rather than earlier in the study period is large. Holding other explanatory factors at their averages, Analysis 3 predicts a 9-fold increase in direct appeal reversal rates over 23 years (from about 9% to about 80%).804 This finding is important because state direct appeal is the only stage that reviews nearly all death verdicts, and it accounted for about 8 of every 10 reversals during the study period.805

  • In order to make the best use of our data on capital reversal rates, it was necessary in many of our analyses to measure reversal rates as proportions of imposed verdicts.806 Most analyses also included the federal habeas stage as at least one of the review phases being studied. As a result, most of our analyses are affected by both delay-related, error-neutral biases noted above.807 And two studies of the federal habeas stage were affected by the second bias, but not the first.808 As those biases would predict, later verdicts were associated with lower reversal rates in a number of these analyses.809 Contrary to expectations, however, the size of the effect was fairly small.810 And in three analyses, there was no statistically significant relationship between later verdicts and lower reversal rates.811 These latter results suggest what our direct appeal studies found: that there is an upward trend over time in the amount of serious error that is not accounted for by the other factors in the analysis, which partially—and in some analyses entirely—neutralizes the downward force of the two delay-related biases discussed above.812

Given these circumstances, our regression analyses modestly enhance what the raw trend of reversal rates over time—depicted in Figures 2A-3B—tells us about the effect on reversal rates of the passage of time. Those analyses are most informative as to the state direct appeal stage, because it is only at that stage that they provide a relatively accurate picture, undistorted by the effect of delay, of the relationship between the passage of time and the amount of serious error discovered by the courts after accounting for other factors. Those analyses show that after controlling for other factors, death verdicts imposed later in the study period were substantially more likely to be reversed at the state direct appeal stage—where nearly four-fifths of all capital reversals occurred during the period—than verdicts imposed earlier in time. Our best analysis predicts that, if other factors had remained constant at their averages, direct appeal reversal rates would have risen 9% per year during the 23-year study period.

Other significant factors did not, of course, remain constant at their averages, and reversal rates in fact were fairly steady during the latter half the study period.813 What increased over time, therefore, is the amount of error found on direct appeal that is not accounted for by the specific explanatory factors we have identified, and instead is registered by our general measure of time trend. This suggests that reforms aimed at alleviating the specific conditions that our analyses have shown to be significantly related to reversals may have less effect than is desired because of the influence of other factors—picked up in our analyses by our measure of time trend—that are associated with increasing amounts of capital error over time.

What we can say with confidence based on these results is that:

  • Overall capital reversal rates remained high and fairly steady from the early 1980s through the end of the study period, averaging about 60% of the verdicts reviewed each year.

  • There is no evidence that conditions causing high capital error rates are curing themselves over time.

  • Most disturbingly, at the direct appeal stage, factors beyond those specifically identified by our regression analyses are linked to increasing amounts of serious error over time.

5. Reviewing courts do not effectively keep serious mistakes from being made or death verdicts from being carried out.

State direct appeal and post-conviction courts and federal habeas courts are the capital system's quality control inspectors, whose job it is to detect seriously flawed death verdicts imposed at trial and to send them back to be retooled or scrapped. Our analyses examine the outcomes of thousands of these inspections mainly to identify causes of serious flaws at trial. But the analyses also shed light on the effectiveness of the inspection system. Such systems have two goals—to catch individual mistakes before they cause unintended harms, and to feed back information and sanctions to those who made the mistakes—particularly information and sanctions focused on patterns of problems—so that error does not occur in the future. This sections concludes that the review process is not a failsafe method of achieving either of these goals. We begin with the second.

a. The review process fails to keep high rates and amounts of serious error from recurring.

The capital review system fails utterly to keep serious mistakes from being repeated. Rates of serious capital error were disturbingly high during the entire 23-year study period— with an overall rate for the period of 68% that remained around 60% even in the last years of the study.814 Although there is some evidence suggesting (among other possible conclusions) that the burden of catching error has shifted somewhat from federal courts at the third inspection stage to state courts at the first and second stages,815 there is no reliable evidence that rates and amounts of error have declined substantially since the early 1980s.816 Moreover, for nearly two decades, the rate at which people sentenced to die have thereafter been exonerated has been fairly steady at 1 innocent death row inmate for every 7 or 8 people executed.817 Nor—at least apart from last year's incipient and scattered reforms818—is there any evidence of ameliorative changes since the study period that were designed to, or can be expected to, lead to lower rates of serious error in capital cases. Instead, as we note above, the most important changes in the years between 1995 and 2000 were designed to substantially decrease the level of scrutiny and feedback that appellate courts give to the capital trial process, and that federal reviewing courts give to state reviewing courts.819

For this reason alone, the capital system is broken. This is best illustrated by asking whether decades of 50%-plus rates of serious error would be tolerated in any other public or private enterprise in this country. If goods coming off the production lines at Ford Motor Co., General Dynamics or Dell were so seriously flawed that they had to be sent back for repair or scrap 68% of the time, it is doubtful the enterprise would last a year—and it is certain that investors, regulators and consumers would shut down the operation long before its failures went on for decades.820 The same is true of 50%-plus rates of serious error in public operations, such as issuing social security checks, constructing schools or air traffic control. Nor would it be any consolation that the enterprise's chronic failures have not yet killed any innocent people—at least so far as can be proved.821 Meticulous inspections or not, it is simply unreasonable—especially over the course of decades—to continue tolerating:

  • the costs of operating consistently failing enterprises and having to fund multiple overlapping inspections systems and repairs;

  • the delays that complex, redundant and painstaking inspections require;

  • the inconveniences and injuries that people suffer from persistently faulty products and outcomes; and

  • the risk that a day of reckoning will arrive when inspections fail, and when a seriously flawed product or system causes an innocent person's death.

 

It thus is clear that the capital review process fails as a means of feeding back information and, where necessary, sanctions on defense and government lawyers, law enforcement officers, and judges who conduct flawed capital trials. The first reason for this failure, as a number of investigative journalists have recently documented, is that appellate courts understand their role as examining each case separately. They accordingly keep no aggregate data about how frequently they reverse death verdicts due to errors committed by particular lawyers and law offices, police officers and police forces and lower court judges. And they entirely pass over many errors they find as non-prejudicial, harmless or waived—even where those errors contribute to patterns of abuse that previously or subsequently have resulted in reversals. As a result, although court decisions in fact often reveal egregious patterns of error by particular defense lawyers, prosecuting offices, police forces, and trial judges, those patterns rarely are noticed, much less sanctioned in any way, by reviewing courts. Consequently, problems can fester for years.822

In addition, a review process taking 12 years on average before executions occur is unlikely to be an effective way of informing, instructing or disciplining the actors responsible for flaws the review process finds. As investigative news reports also have recently documented, by the time the capital review process is finished and a reversal occurs, the offending trial-level actors have usually moved on to other jobs.823 In most cases, moreover, trial-level actors do not have to defend flawed capital trial verdicts on appeal, because that task is handed over by defense lawyers to new appellate lawyers in a different office, and is handed over by local prosecutors to lawyers in the state attorney general's office. In neither case do the new lawyers have authority to discipline trial-level actors whose mistakes the later lawyers must defend. Instead, appellate lawyers for the state are often blamed for having "lost" the case on appeal when the verdict is reversed.824

Nor, as those same reports have shown, do court reversals ever lead to bar discipline for lawyers, loss of jobs for law enforcement officers or other state employees, or sanctions for judges who repeatedly commit serious error.825 Rather the only "sanction" imposed is an order to retry the case—typically handed down many years after the fact. For all these reasons, nearly the entire cost of the review process and its outcome is borne, not by the trial-level actors who committed the errors in the first place, but by taxpayers spread throughout the entire state (who fund the state court system and state attorney general's office) and throughout the entire United States (who fund the federal court system and the lawyers who represent indigent capital defendants in those courts). Because local taxpayers do not have to bear most of the costs of the mistakes local officials make, they have little reason to discipline local officials for their mistakes by voting them out of office. And because the state and federal taxpayers who do foot the bill are removed from the local situation, they typically have no idea what is happening and, if they do, have no recourse against the responsible officials.

Our study provides evidence of disturbing ways in which the chronically failing capital system may actually reward actors who are responsible for many of its flaws. Our principal finding is that excessive death-sentencing is the most crucial source of serious capital error. An important supporting finding is that serious error is especially common in states where judicial selection techniques give judges strong incentives to conform their rulings to popular sentiment. Together, these findings suggest that judges and probably other officials826 benefit politically from each additional death verdict they are at least partly responsible for securing, including in weak or marginal cases where the probability of reversal is great. Particularly given that most of the costs of curing the resulting errors fall on others, the clear incentive the system gives officials is to cast the net of capital punishment law and policy still wider, pulling in progressively weaker cases in which the likelihood of error is progressively higher. Added to this is the fact that higher numbers of death verdicts mean more delays on appeal, which in turn tends to dampen and obscure reversal and reversal rates and to delay the point when the case will be sent back for retrial827—further weakening any disciplining force of reversals when they finally come.

An analogous process affects the work of skilled capital defense lawyers—mainly from out-of-state civil rights organizations and law firms—whom our study shows have the greatest success in overturning seriously flawed capital verdicts at the final, federal habeas stage of review.8288 Because there are so few of these lawyers and so few resources to fund their work—a problem Congress and the states made worse when they shut down the "capital case resource centers" in 1995829—these lawyers cannot handle the thousands of capital trials taking place all over the country each year, and instead can only intervene at the last stage of review after state court reversals and review delays have narrowed the number of pending cases to a manageable number. Given how often their clients' death verdicts are overturned due to persistent flaws in capital verdicts, it is not surprising that these lawyers work hard to preserve a robust three-stage review process in which they are largely responsible for the last stage. Nor is it surprising that they are mistrustful of promises to trade meaningful trial-level improvements, which thus far have not materialized, for limits on post-trial review that by themselves will make things worse.830 As understandable as these views are, however, they have the same counterproductive effect as the actions of the opposing camp. They divert good lawyers from the trial phase, leaving poor lawyers to contribute to high death-sentencing and high error rates, and they preserve the lengthy review process that the weak trial system requires. In other words, they keep a broken system going, for decades, chronically generating too many death verdicts—most of which, as a result, are seriously flawed and unreliable—which in turn require an expensive review process that is so delayed that it stymies execution of valid verdicts and so overburdened it misses egregious mistakes.831

b. The review process does not catch all serious mistakes.

Our results also indicate that the capital review process has not achieved the other goal of an inspection process: catching flawed products before they harm innocent people. Our case studies of some of the death row inmates shown to be innocent after judges at all three review stages had approved their verdicts for execution reveal that the judicial inspection process has failed on several occasions to catch the most serious capital error of all—the conviction and capital sentencing of an innocent man or woman.832 Of the 99 death row inmates who have been exonerated during the modern death-sentencing era, over 60% had their capital verdicts approved by at least one set of appellate courts.833

Our results also help explain why appellate courts fail to catch even the most egregious capital errors. In each case study of an innocent man approved for execution by a full complement of state and federal courts, the courts took note of the questionable procedures later shown to have put an innocent man on death row and even acknowledged doubts about the reliability of the resulting verdict. Nevertheless, the courts refused to overturn the verdicts because the innocent defendant was unable to satisfy the strict standards for proving that he pleaded the claim properly at trial and on appeal, and that the acknowledged errors in his case had "prejudiced" him.834

Our regression analyses in turn reveal evidence that reviewing courts sometimes set the bar to reversal high in response to political pressures and a desire to avoid the controversy that frequently accompanies reversals but almost never accompanies affirmances.835 In addition to the political pressures discussed above to impose death verdicts at trial in cases that are not highly aggravated, where error rates are the highest,836 our results provide evidence of pressures to approve death verdicts on appeal despite the presence of error that renders the verdicts unreliable:

  • The more political pressure imposed on judges by a state's method of selecting—which usually means electing—judges, the higher is the risk that capital trial verdicts imposed in the state will be seriously flawed.837 State judicial selection techniques have the strongest association with the discovery of reversible error at the federal habeas stage, where the judges are appointed and life-tenured and thus are immune to the pressures generated by state judicial selection techniques (Analysis 6). The association between the discovery of error and state selection methods is somewhat weaker but still close to significant at the state direct appeal stage, where pressure on elected judges triggered by particularly notorious capital cases is moderated by the passage of time between the commission of the crime and the appellate ruling and by the fact that most constituents of appellate judges come from communities besides the one where the verdict under review was imposed and thus are not as interested in how the court decides the case (Analyses 3, 4, 10). There is no evident relationship between judicial selection techniques and the discovery of error by state post-conviction judges, who usually are the same trial judges who imposed the death verdict in the first place, and who face the most direct political pressure from cases under review because all their constituents come from the community where the crime occurred (Analysis 5).838 This suggests that political pressures that are associated with high rates of error at trials supervised by elected judges may also keep the same judges from correcting errors during subsequent state post-conviction proceedings, and may discourage elected high court judges from reversing verdicts on direct appeal.

  • As is revealed by main Analysis 1 and a wide array of confirming analyses (Analyses 2-5, 7-18), state direct appeal judges and state post-conviction judges are significantly more likely to find serious error and reverse death verdicts imposed in more urbanized and populous states and counties, and less likely to reverse verdicts imposed in less urbanized and populous places. Analysis 6 reveals the opposite pattern for federal habeas judges, who are more likely to find serious error and reverse death verdicts from less urbanized and populous states and less likely to reverse those from relatively urbanized and populous states.839 These opposing patterns are additional evidence of political pressures on state reviewing judges to affirm verdicts that, apart from such pressures, would be reversed due to serious flaws. Urban areas have more homicides and impose more death verdicts, any one of which is not very likely to make a strong and lasting impression on most local citizens. By contrast, less densely populated areas have a smaller number of homicides, each of which—and any death verdict imposed for it—is likely to be well known and important to many local citizens.840 Over the long run, therefore, reversing rural or small-town death verdicts is likely to be more controversial than reversing urban death verdicts, especially for state judges who face direct electoral discipline for locally unpopular decisions.841 At the two state stages of review, the predictable result of a desire to avoid locally controversial reversals is fewer reversals of verdicts from less populous areas than of verdicts from urban areas. This result helps explain why the flawed verdicts found at the final federal habeas review stageC—by appointed, life-tenured judges who are relatively isolated from local political pressures—are disproportionately from more rural states. More generally, it helps explain why the proportion of flawed verdicts found at each successive review stage does not shrink—as otherwise should occur in a properly functioning series of inspections—and instead why almost as high a proportion of flawed verdicts is found at the final capital inspection stage as at the first stage: 40%.

    • The table in the appended note ranks states based on their population size and density and indicates the difference this factor makes, holding other factors constant, in whether states have above-average or below-average reversal rates. As the table reveals, when other factors are held constant at their average, states with low population density are prone to reversal rates as much as 30 percentage points below the 34-state norm when the reversal rates being explained are mainly those of state judges who are especially likely to suffer adverse political consequences from reversing death verdicts imposed in rural communities.842

  • All analyses of reversals taking place at only the state direct appeal stage and at only the state post-conviction stage show that state courts with large backlogs of cases are more likely to affirm death verdicts than courts without such backlogs (Analyses 3, 4, 5 and 10).843 This suggests that pile-ups of cases awaiting review, and associated delays and controversy, pressure state judges to move cases along as quickly as they can, including by affirming verdicts that in calmer times would be found to be seriously flawed. Again, analyses of reversals taking place at only the federal habeas stage, where life-tenured judges are less susceptible to local political pressures show no similar effect.844

  • These results validate the explanation for federal court review of state court decisions famously given by Alexander Hamilton in The Federalist Papers. Federal court review of state decisions, Hamilton wrote, helps assure "an inflexible execution of the national laws" by national courts immune from "a local spirit" that sometimes compromises decisions of local courts. This is especially so, he wrote, when the national laws are designed to bar "arbitrary methods of prosecuting pretended offenses, and arbitrary punishment upon arbitrary convictions."845 But the fact that federal judges are relatively immune from local political pressures does not make the final, federal review stage a firewall against all political influence on the review process. On the contrary, case-level Analysis 19 of federal habeas decisions provides evidence that federal reviewing judges are influenced by national political pressures associated with the process by which they are appointed and promoted. Holding other factors constant at their average, Analysis 19 predicts that the probability that a capital verdict will be reversed rises or falls as much as one-third depending upon whether the review is by judges mainly appointed by Republican Presidents or by judges mainly appointed by Democratic Presidents.846

  • Analysis 19 also provides strong evidence that reviewing federal habeas judges are forced to serve as replacement sentencers, to screen out the many death verdicts induced at trial as a result of excessively broad death-sentencing policies.847 Even so, federal review is not a failsafe check on excessive, error-prone death-sentencing, given federal judges' susceptibility to political pressure, and given the proneness of the strict rules those judges apply to let some, even very serious, errors slip through.848

Reviewing judges thus are demonstrably incapable of curing all of the flawed verdicts imposed at capital trials. This is so in part because reviewing judges are susceptible to political pressures to affirm flawed death verdicts analogous to the pressures trial judges and other trial-level officials face to impose flawed verdicts in the first place—pressures that call for a forceful response to serious crime in general, but are divorced from the strength of the evidence and circumstances supporting a death verdict in particular cases.

c. The probability that innocent people have been executed is high.

As we discuss above, it is impossible to know how many innocent people have been capitally convicted, sentenced and executed—in part because officials are permitted to withhold DNA samples and other crucial information needed to determine the scope of the problem. The best researchers and policy makers can do, therefore, is to use available evidence to estimate the risk that innocent people have been executed.849 Our conclusion on that question is the same as the one Justice Sandra Day O'Connor reached in addressing bar groups last summer and this fall: "If statistics are any indication, the system may well be allowing some innocent defendants to be executed."850

The best evidence we have been able to assemble based on counts, regression studies and case studies of the results of all three stages and each separate stage of court inspection of 4500 capital verdicts imposed in 34 states and 1000 counties across 23 years is as follows:

  • 50%-plus rates of reversible error across nearly all states and years;851

  • strong indications, using multiple measures, that the errors causing these reversals are serious;852

  • deep-seated and disturbing racial and political factors that are strongly associated with that error;853

  • reviewing judges' inability to catch serious error even when it has caused an innocent person to be convicted and condemned;854

  • reviewing judges' susceptibility to pressures to approve flawed capital verdicts;855 and

  • high reversal rates persisting from the first to the last review stage, as opposed to the steadily shrinking rates of discovered error needed to instill confidence in the efficacy of inspection processes.

Other analyses show that for every 7 or 8 death row inmates who are executed, another inmate in line to be executed is proven to be factually or legally innocent.856 Moreover, among the events helping to save innocent inmates before being executed were a documentary film maker's accidental discovery of flaws in one case while examining another; an investigation by college students as a class project in a second case; a police clerk's accidental release of a suppressed file in a third case; and a burglary at a prosecutor's office in a fourth—fortuities that cannot be relied upon to keep miscarriages from occurring.857

Together, these findings convince us that the probability that an innocent person has been executed during the modern death-sentencing era is high. The findings also convince us that lesser but still serious harms are rampant in the capital system, including the execution of individuals who were guilty of some offense but not one for which the law allows the death penalty.

D. Higher-Risk and Lower-Risk States, Given this Analysis

1. Connecticut and Colorado Compared to Florida, Georgia, Texas and Alabama.

As we warn above, Table 18 cannot not give a full picture of the risk of serious capital error that states face based on the factors our regression analyses identify.858 The table analyzes the effect of each factor while holding other factors at their 34-state average. It thus does not measure the combined effect of all factors operating simultaneously. In addition, Table 18 does not account for three general factors our regression analyses consider—year, state and time trend—which gauge the influence of still other forces that are not studied directly but are associated with the location and timing of the relevant death verdicts and reversals. Subject to these limitations, however, it is possible very generally to associate a particularly high risk of error with a few states that fall fairly consistently on the high-end of the risk spectrum—and to compare those states to ones that more consistently fall towards the low end of the risk spectrum. In doing so, we consider the six important factors in Table 18 and the four additional factors addressed in the tables in notes 774, 788, 797 and 842.859

As a review of Table 18 and the accompanying tables makes clear, most states' 10 risk rankings are widely distributed across the spectrum from first (most risk of high capital reversal rates) to 34th (least risk of high reversal rates). In most cases, therefore, the information in Table 18 and the accompanying tables suggests particular areas where each state might focus policy attention without providing a strong basis for distinguishing the state from any other. In a small number of cases, however, states' risk rankings fall fairly uniformly towards one end of the risk spectrum or the other. On the low side, for example, are Connecticut and Colorado. Based on average conditions across the 23-year period,860 and on analyses of each of the 10 risk factors, holding other factors constant at their averages:

  • Seven of Connecticut's 10 risk rankings place it in the bottom half of the 34 states in terms of the probability of serious capital error, including four rankings in the bottom five of 34. Most importantly, given our principal finding above, Connecticut is ranked last in terms of the risk of error posed by its (low) capital-sentencing rate. Thus, although Connecticut was one of four states with 100% reversal rates during the study period, that rate is based on a total of only two decisions and does not provide a fair estimate of the state's risk of serious capital error over the long haul. Our analyses suggest that Connecticut capital verdicts pose less of a risk of serious error than verdicts in most other states.

  • Six of Colorado's risk rankings are in the bottom half of all states, with an additional ranking on the border between the top and bottom half (17 out of 34).861 Colorado is ranked third-to-last in terms of the risk of error posed by its capital-sentencing rates. Colorado's reversal rate during the study period was 75%—based on only four decisions, three ending in reversals.

Connecticut and Colorado may be contrasted to Florida, Georgia, Texas and Alabama. Based on average conditions across the 23-year period,862 and on analyses of each of the 10 risk factors, holding other factors constant at their averages:

  • Eight of Florida's 10 risk rankings place it in the top half of states based on its predicted risk of serious capital error, including two placing it in the top five among the 34 states. A ninth ranking is on the border between the top and bottom halves of the 34 states (18 out of 34). The only ranking out of 10 on which Florida has a substantially below average risk of capital reversals is the result of its large backlog of capital appeals awaiting review—the third highest backlog in the country. As we note above, delay in the review process has the perverse effect of lowering reversal rates.863 Adding to concerns about the risk of serious capital error in Florida:

    • The state's death-sentencing rate is 12th highest out of 34.

    • Three of the top ten counties in the nation with the highest death-sentencing numbers and rates are Florida counties.864

    • Florida has had more people removed from its death row following findings that they were not guilty than any other state.865

    Florida's overall capital reversal rate during the study period was 75%.

  • Seven of Georgia's 10 risk rankings put it in the top half of all states in terms of the predicted risk of serious error. Four rankings put it the top five of all states. Georgia is the only state among the 34 that is not in the bottom 10 states on at least one risk factor. And it lowest ranking (21 out of 34866) is due to its above-average number of death verdicts that are stuck in the appeals process awaiting final review. Working modestly in Georgia's favor, its death-sentencing rate ranks only 18th out of 34. Georgia's overall capital reversal rate during the study period was 80%.

  • Seven of Texas's 10 risk rankings are in the top half of the 34 states. Two are in the top five. As in the case of Florida and Georgia, the factor on which Texas ranks the lowest in terms of predicted reversals is a result of its high backlog of capital cases awaiting review— the second highest in the nation. Also moderating predicted reversal rates is Texas's relatively low death-sentencing rate—25th out of 34. During the study period, Texas had an overall capital reversal rate of 51%. Although high in absolute terms, this rate is towards the low end compared to other states. See Figures 1A and 1B, pp. 50-51 above. One important line of inquiry for Texas, given its high rankings on most risk factors, is whether—as some have recently claimed—its relatively low capital reversal rates are due to excessively lax state court review of capital verdicts.867Other explanations are Texas' high backlog of verdicts awaiting review, which tends to depress reversal rates, and the state's relatively low death-sentencing rate.

  • Six of Alabama's 10 risk rankings place it in the top half of the 34 states. Three risk rankings place it in the top five among the 34 states. During the study period, Alabama's death-sentencing rate was 11th in the nation. Alabama's overall reversal rate during the study period was 77%.

2. Virginia.

As is discussed above, Virginia has extremely low capital reversal rates.868 Compared to other states with cases decided at all three review stages during the study period, Virginia's 17% overall reversal rate—the product of the lowest state direct appeal reversal rate in the county and the lowest federal habeas reversal rate in the country—is more than two standard deviations below the mean. Two theories have been offered to explain Virginia's low reversal rates—uniquely high-quality death verdicts or, on the other hand, uniquely low-quality court review.869 Our findings suggest that the truth lies in between those two poles. In fact, Virginia's rankings on the 10 risk factors tend to cluster around the two poles of fairly low, and fairly high, risk of serious capital error:

  • On the one hand, Virginia falls among the bottom five states in terms of its risk of serious capital error in four of the ten risk categories in Table 18 and the allied tables. Chief among these low-risk categories is Virginia's death-sentencing rate, the sixth lowest in the nation. Virginia also ranks low in terms of the political pressure put on state judges through the electoral process, and given the state's relatively strong record of apprehending and punishing serious criminal—both of which tend to relieve pressure to use the death penalty as a stop-gap response to ineffective law enforcement strategies.870

  • On the other hand, on three of the remaining six risk factors, Virginia ranks in the top ten among the 34 study states—including with respect to the two racial factors that pose a high risk of capital error. The state ranks eleventh on still another factor.

Based on the factors our study identifies as important, we conclude that the risk of serious capital error in Virginia is, on the whole, fairly moderate, but that the risk is not low enough to explain the state's extremely low reversal rates. Our findings tend to confirm those of the state's Joint Legislative Audit and Review Commission, which recently concluded a year long study of the state's death penalty ordered by the state legislature. The Review Commission concluded that federal and state judges' adherence to strict rules limiting review for serious error in capital cases, and the state high court's narrow review of the appropriateness of death sentences in particular cases, may have let stand the convictions and sentences of some death row inmates who did not receive proper trials.871 We, too, conclude that lax state and federal court review of Virginia death verdicts has probably depressed the state's reversal rate below its actual rate of serious capital error.

next | previous | table of contents