next | previous | table of contents

C. A System Fraught with Serious Error

The serious nature of the errors leading courts to reverse thousands of capital verdicts during the 23—year study period may be shown systematically, as well as with the above examples.

Every decision that a death verdict is too flawed to be carried out is serious because it

  • torments victims, forcing them to relive excruciating events at a retrial, and risks either dashing their expectations with a noncapital verdict or causing them to endure another round of appeals;

  • requires a costly retrial of one or both phases of a capital trial—meaning, more judges, prosecutors, police officers, witnesses, defense lawyers, and court personnel to compensate, and more jurors and lay witnesses to inconvenience;150

  • risks a new set of appeals—meaning still more judges, defense lawyers, states attorneys, and lay and expert witnesses to compensate or inconvenience;

  • delays the processing of other court cases; and

  • casts the court systems in a bad light.

Reversible error is also serious because of its devastating impact on the capital system. The rejection of more than two thirds of all finally reviewed capital verdicts over nearly a quarter—century reveals a system collapsing under the weight of its own errors:

  • In the 23—year study period, 2349 death verdicts were reversed and sent back for retrial.

  • In the same period, it took an average of 9 years from trial to execution; by the end of the period, it was taking 11 years, and the figure now is 12 years.151

  • Only 5% of the 5826 death verdicts imposed from 1973 to 1995 were carried out in the period.

  • On average since 1981, capital states have managed to execute only about 1.2% of their death row inmates each year. Those inmates now number about 3700.

The case studies above and the analysis below reveal an even more compelling reason to take seriously the huge amounts of error found by A Broken System: Neither modern capital trials nor multi—layered appeals are reliable ways to keep people from being sent to death row or executed for crimes they did not commit or for which the law bars the death penalty.

This section explores the answers to four questions:

  • Given what we know about reviewing judges, should we expect them to be stingy or generous with their findings of reversible error?

  • Given the rules governing when error is, and is not, reversible, and given the actual reasons for reversal, is it likely that reversible error affects verdicts' accuracy, or is merely technical?

  • What do the outcomes of retrials following reversals reveal about the effect of reversible error on the reliability of death verdicts?

  • Do patterns of reversible error across states, time and court systems suggest that high rates of error are localized or widespread?

Answering these questions confirms what the case studies show: Not every known error that occurs at capital trials—nor even every error that sends an innocent person to death row—qualifies as reversible error and is included in the 68% rate of serious error discovered by A Broken System. To qualify as reversible error and be counted by us, mistakes in capital verdicts must be serious; they must stand out in the record and undermine faith in the result; they must impress reviewing judges—and should impress others—as matters of grave public concern.152

1. Reviewing judges' political and professional incentives dispose them to affirm, not reverse, death verdicts and to make full use of legal rules that allow them to avoid reversing death verdicts despite known error.

Nearly all judges in capital states are elected, and for decades their decisions in capital cases have been closely scrutinized by opponents and voters.153 Rulings overturning death verdicts are frequently cited as reasons to remove judges from office, including in well—known cases in California, Mississippi, Tennessee and Texas where judges were turned out of office because of such rulings.154 Election promises to impose and uphold death sentences for convicted killers are just as routinely given as reasons to vote for judicial candidates—or, indeed, as a threshold qualification for being a judge.155 By contrast, a careful search of the relevant sources reveals not a single report of a judge being turned out of office or denied election for imposing or affirming a death verdict, including the 99 verdicts imposed since 1973 on people who were factually or legally innocent.156 Even candidates for appointment to the state and federal bench have frequently been denied appointments to higher courts—and, when appointed, denied confirmation—because of occasional votes to overturn death sentences (among many others to affirm them) while serving on lower courts.157

These political realities are pronounced and one—sided, and convey a powerful message:

Unless the reason is strong and apparent—and even then—voting to reverse a death verdict creates a big risk that judges will be voted off the bench, or denied promotion to higher office. Voting to affirm a death verdict has no adverse political effects.

Given this message, the fact that reversals occur with great frequency—and, indeed, that reversal is the usual final outcome—indicates the seriousness of error in capital verdicts.

This is particularly so given that just under 90% of the 2349 capital reversals from 1973 to 1995 were by elected state judges.158 Because reversals by elected judges are good indicators of the seriousness of capital error, the fact that those judges were responsible for the vast majority of capital reversals during the study period is another indication that most reversals reflect poorly on the reliability of the capital system.

The same conclusion may be drawn from the party affiliations of a majority of the unelected federal judges responsible for the remaining 10% of reversals:

  • In 54% of those cases, appointees of Republican Presidents were a majority of the judges voting to reverse the verdict—even though, as we show below, Republican appointees are less disposed to reverse death verdicts than Democratic appointees.159

    1. In 73% of those cases, one or more Republican appointees were on the review panel, and a majority of them voted to reverse the capital verdict.

    2. In 80% of the cases, half or more of the Republican appointees voted to reverse.

    3. In 83% of the cases, at least one Republican appointee agreed that the capital verdict had to be overturned.

  • All told, about 95% of the thousands of reversals documented by A Broken System were by elected state judges or by Republican appointees to the federal bench.

A good measure of the seriousness of defects inspectors find is how disposed they are to reject products. We consider that question in the case studies above and in this section's study of the disposition against reversal of the vast majority of reversing judges. Studies of appellate review of death verdicts in Illinois, Ohio and Texas address the same issue.160 They, too, show that reviewing judges use a wide mesh to sift capital verdicts for reversible error: Only clear and acute error—and not even all of that—is corrected. Yet even these loose inspections find large amounts of reversible capital error. By this measure, reversible capital error is serious, and the need to study its causes is substantial.161

2. Legal rules bar judges from reversing based on errors with no effect on the outcome, and most documented reversals are for errors that probably affected the outcome or are inherently prejudicial.

a. Stringent rules limiting reversal.

The available evidence indicates that judges are "tough sells" when it comes to finding capital error. But do the rules they follow require them to reverse death verdicts for technicalities or other unimportant errors? The case studies above of four innocent men sentenced to die and then approved for execution by three sets of state and federal reviewing judges, reveal the opposite: Technical rules often bar courts from reversing, even for serious error. Courts repeatedly found defects in capital verdicts but did not reverse because the defendant's lawyer "waived" the error by not following the rules for pleading claims; because the violation was ruled "harmless"; or because the defendant could not prove what was in fact the case, that admitted errors had destroyed the accuracy of the jury's guilt and sentencing decisions. Because of these limits on relief, courts let pass:

  • the confession of retarded defendant Earl Washington containing statements directly contradicted by the most obvious facts of the crime such as the victim's race—which courts held reliable enough to support the conviction;

  • the failure of Washington's lawyer to tell the jury that the blood type of the semen at the scene excluded his client as the rapist—ruled non—prejudicial;

  • the admittedly incompetent failure of Anthony Porter's and Lloyd Schlup's lawyers to interview or call witnesses known to have identified someone else as the killer—also ruled non—prejudicial;

  • the seating in Porter's case of a juror known to have misled the trial judge about her close relationship with the victim's mother—held non—prejudicial and waived; and

  • the apparent unreliability of crucial eye—witness accounts in Frank Smith's case—ruled not unreliable enough to undermine the conviction.

Given these and other case studies,162 it is probable that waiver, harmless error and non—prejudicial error findings led judges to ignore mistakes of a similar magnitude in most of the 60—plus cases where innocent defendants were approved for execution by one or more sets of reviewing courts.163

b. Serious reasons for reversal.

The above analysis suggests that cautious judges applying stingy legal rules are more likely to miss serious error than to reverse based on non—serious error. We are led to the same conclusion by data we collected on the reasons for court reversals at the final two review stages (state post—conviction and federal habeas). These stages inspected only the cleanest 59% of death verdicts, because those stages occurred after direct appeal already had removed the 41% of the verdicts with the most glaring errors. For example, a lack of evidence of guilt is a fairly common reason for reversal at the first, direct appeal stage, but few such obvious errors are left to be caught at later review stages.164 The two review stages covered by our data also are the ones critics of multi—stage review cite as the most likely to reverse due to technical error not affecting the reliability of verdicts.165 If the reasons for reversal at those stages—with the cleanest cases and (some say) the most suspect judges—are clearly serious, there is reason to think that many or most errors caught at the first review stage by the least suspect judges are also serious.166

At the state post—conviction review phase, four errors account for 80% of the reversals:

  • egregiously incompetent lawyering (39%);

  • prosecutorial suppression of evidence of innocence or that death is not a proper penalty and other police and prosecutor misconduct (19%);

  • improper instructions to jurors on the law governing when defendants may be convicted and sentenced to die (19%); and

  • judge or juror bias (4%).167

Reversal on the first three these grounds requires proof that the error probably affected the outcome, and most errors in the last category are inherently prejudicial.

At the federal habeas stage, four analogous errors also account for most of the reversals:

  • egregious ineffective assistance of counsel (a basis for reversing 27% of the death verdicts reviewed on habeas),

  • misinstruction of jurors (39%);

  • prosecutorial suppression of seriously exculpatory evidence or other police and prosecutor misconduct (18%), and

  • judge or juror bias or the deliberate exclusion of African-Americans from the pool of prospective jurors (7%).

These errors sometimes occurred in the same case. After accounting for the contribution of each error to all reversals, the four errors are responsible for 74% of all federal habeas reversals.

Together with these errors, four others account for 81% of the federal habeas reversals:

  • forced trials of mentally incompetent defendants who could not communicate with their lawyers or understand the proceedings against them—including one who was acquitted after medical treatment restored his competency;168

  • prejudicially denying impoverished defendants funds for experts on DNA and other forensic analysis;

  • coerced confessions not found to be harmless; and

  • trial court rulings excluding exculpatory evidence of innocence.169

As is true of these seven types of error accounting for over 80% of state post—conviction and federal habeas reversals, legal mistakes are rarely a basis for reversing capital verdicts unless they are found to have impaired the reliability of the outcome. This tendency to undercut the reliability of death verdicts is another reason the reversible errors we count are serious errors.

Television and other fictionalized accounts sometimes suggest that the most common reason convictions are reversed is to punish the police for not following technical rules when gathering reliable and damning evidence. As far as our data show, however, courts never reverse capital verdicts based on technical error freeing demonstrated perpetrators of serious crimes:

  • Not a single capital verdict reversed at the state post—conviction or federal habeas stage during the 23—year study period was due to a police officer's or other official's unlawful search, seizure or presentation in court of reliable evidence.170

  • Instead, most errors found at those stages were:

    • defense lawyers' incompetent failure to discover or present evidence that the defendant was innocent or not subject to the death penalty;

    • prosecutors' unlawful suppression of the same kind of evidence; or

    • trial judges' improper exclusion of or instructions to ignore such evidence.171

  • Statistical analysis below of traits of cases in which reversals tend to occur finds a strong relationship between the weakness of the evidence supporting a death verdict and the probability of reversal.172

Our data show that reviewing courts are inclined to, are legally permitted to, and usually do reverse capital verdicts only because of real doubts about the verdicts' accuracy. If anything, courts err on the side of ignoring doubts, and approving flawed verdicts. That, even so, most capital verdicts are reversed suggests that capital error, and the frequency with which it occurs, pose a serious problem calling for careful study of why so much error occurs.

3. Retrial results following state post—conviction reversals provide strong evidence that verdicts with reversible error are unreliable.

Above we find that most capital reversals require evidence that the verdict was the result of error rather than a reliable assessment of the facts, and that 95% of the reversals were ordered by elected and Republican—appointed judges who have strong incentives to approve death verdicts absent convincing evidence of unreliability. When reversals do occur, the case usually goes back for a retrial of one or both phases of the capital trial and a new verdict. The question we consider here is whether curing such error on retrial changes the outcome often enough to indicate that most reversed death verdicts are indeed inaccurate and thus that most reversible errors are serious.

Tracing post—reversal outcomes is difficult. They are not reported, and there is no central source of even unofficial information. When we decided late in the data—collection process to make the painstaking inquiries needed to count state post—conviction reversals,173 we also undertook to collect retrial results for those cases as well. For the 352 known state post—conviction reversals between 1973 and April 2000, 306 retrial outcomes were known as of April 2000. (The other retrials were still pending, except in 5 cases where the defendant died before retrial or the outcome is unknown.) Our findings run in the expected direction: Reversed death verdicts tend to end in non—capital outcomes on retrial. The lopsided proportion of changed outcomes surprised us, however:

  • Discovering reversible error and curing it on retrial changed the outcome of 82% of the underlying capital verdicts.

  • 73% of those retrials ended in a sentence less than death; 9% ended in acquittals.174

These findings provide strong additional evidence that reversible error is serious error. Assessing their full implications requires answers to two questions. First, how likely is it that capital retrials end in lesser verdicts because evidence was lost in the meantime, and not because reversible errors led to inaccurate outcomes that retrials cured?

This possibility is substantially dispelled by (1) how evidence is preserved between trials and retrials and the rules on presenting evidence at retrials, (2) the kinds of error leading to retrials and (3) the nature of decisions to accept pleas to lesser crimes. Apart from a finding that evidence at the first trial was false, unreliable or illegal—in other words, unless the verdict was reversed because it resulted from unreliable evidence—all evidence admitted at the first trial is admissible at retrial. Moreover, all such evidence is preserved while appeals take place. Physical evidence is kept in the court record subject to strict rules for its handling. Testimony at the first trial is preserved in a trial transcript the court reporter prepares. The transcript then may be used by prior witnesses to refresh their recollections before testifying at retrial, as may prior statements to police, photographs and other available information.175 If these aids still do not enable witnesses to remember their prior testimony, or if they cannot attend the retrial, their prior testimony and other recorded recollections are themselves used as evidence and read to the jury.176 Thus, although important evidence can conceivably be lost between trials, it rarely happens.

Moreover, death verdicts typically are reversed only after defendants show (among other things) that a second jury will be better informed than the first. Most reversals—e.g, for incompetent lawyering, police suppression of exculpatory evidence, court rulings excluding such evidence and jury instructions improperly telling jurors to ignore it177—occur only after defendants prove that the jury at the first trial was illegally denied available evidence that would likely lead to a different outcome at a proper trial.178 In addition, retrial investigators often find entirely new evidence, further enhancing the information available to retrial, as opposed to original, juries.

Finally, there is reason to conclude that agreements to impose non—capital verdicts or to acquit the defendant that prosecutors sometimes reach at retrials are reliable indicators of how the first trial would have ended, but for the errors that occurred there. Nationwide, prosecutors settle more than 85% of all criminal cases, relying on their expert analysis of the facts and law, their experience identifying proper outcomes, and their commitment—enforced by frequent elections—to the imposition of stiff sentences for dangerous criminals. Prosecutors who originally put someone on death row, and the victims they consult, are unlikely to agree to a lesser sentence or release, therefore, unless their now better informed and more deliberate judgment justifies that outcome. Below we report evidence that reversible error is related to pressures to impose death verdicts in marginal cases where the evidence of guilt or facts calling for a death sentence are weak.179 This tendency may be especially strong when prosecutors make snap charging decisions absent full investigation in response to outrage at a serious crime or doubts about the ability of existing law enforcement strategies to solve the crime and prevent its recurrence.180 These kinds of pressures on prosecutors at the original trial may lead to precipitous refusals to agree to outcomes less than death that turn out on retrial, following cooler and better informed reflection, to be justified. Thus, pressures that sometimes arise at initial capital trials may trigger over—charging and error, whereas the better informed and more reasoned decisions that are possible at retrials may be more conducive to accurate outcomes.

A second question is posed by our data on retrial outcomes following state post—conviction reversals: What, if anything, do these outcomes following reversals at the second review stage suggest about retrial outcomes following reversals at the first review stage (direct appeal) and the third review stage (federal habeas)? Absent data on outcomes following first— and third—stage reversals, what we can say is that there is no evident reason to expect that retrial outcomes following those reversals are dramatically different from the retrial outcomes on which we have data.

We begin with a comparison of the first, state direct appeal, and the second, state post—conviction, review stages. State practices vary, but there often are claims that may be raised at the first review stage that cannot be raised at the second stage, and vice versa. This could mean that errors leading to reversal at the second stage are more serious than errors triggering reversal at the first stage, which could suggest in turn that first—stage reversals are less likely than second—stage reversals to prompt a lesser outcome on retrial. Four factors tend to disprove this hypothesis:

  • Because direct appeal comes first, that review process is likely to catch the most glaring errors —and thus many of the most serious mistakes.

  • State high court judges who oversee state post—conviction review are the same judges who review verdicts on state direct appeal. Use of the same judges with the same dispositions about how serious error must be to warrant reversal suggests similarity, not difference, in the seriousness of error prompting reversal at the two stages.

  • Certain errors of a non—constitutional nature are bases for direct appeal reversal but not for state post—conviction reversal. Nearly all such error, however, is subject to the same harmless error rule as applies on state post—conviction review. Because the same judges apply the same ultimate decision rule to the same evidence of guilt at the two stages, there is little reason to suppose that trivial errors are substantially more likely to lead to reversal on direct review than on state post—conviction review.

  • Certain serious constitutional errors may be raised on state post—conviction review but not on direct appeal, such as incompetent counsel. But other serious constitutional errors may be cured on direct appeal, but not generally on state post—conviction review, including the most serious error of all—that there was not enough evidence to convict the defendant.181 Because this latter error automatically leads to post—reversal acquittal, and because no category of post—conviction reversals does so, it is likely that more direct appeal reversals lead to retrial acquittals than do post—conviction reversals. Claims of insufficient evidence to allow a death sentence also are usually adjudicated on direct appeal, not state post—conviction, and they too— unlike any kind of claim available on state post—conviction—automatically require a non—capital sentence on retrial.

  • Overall, the allocation of serious constitutional claims between direct appeal and state post—conviction review is sufficiently balanced that their retrial outcomes should not differ greatly.

Retrial outcomes following federal habeas reversals are also unlikely to differ much from those following state post—conviction reversals. The main reasons for reversals at the two stages are similar,182 as are the procedures and rules that apply and the lawyers who litigate there.

Available data support the judgment that retrial outcomes do not greatly differ depending on the review stage where reversals occur. John Shiffman recently examined the aftermath of all death row reversals ordered by Tennessee direct appeal, state post—conviction and federal habeas courts between 1980 and mid—2001.183 Of the 76 retrials ordered during that period, 58 had reached a verdict or other disposition by the time Shiffman completed his study. Of those, 89% (51) resulted in an outcome less than death—which dovetails with our finding, based on retrial outcomes following reversals at the state post—conviction stage, that 82% ended in sentences less than death. Although retrials following state post—conviction reversals were somewhat more likely to result in lesser outcomes than retrials following direct appeal and federal habeas reversals, Shiffman's central finding is that rates of changed verdicts were remarkably high at all three stages:184

  Direct Appeal State Post-Conviction Federal Habeas
Outcomes Less than Death /
Completed Retrials (Percent)
24/29 (83%) 24/25 (96%) 3/4 (75%)


All available data suggest that a large majority of capital retrials end in a non—capital sentence or acquittal. This would not be true if courts often reversed capital verdicts for technical reasons with no effect on the accuracy of the verdict. Instead, it supports the same conclusion as the cautious predilections of most reversing judges, and the seriousness of their reasons for reversing: Reversible error is error that tends to produce unreliable outcomes.

4. The stability of high rates of reversible error across states, courts and time shows that flaws in death verdicts are serious, chronic and continuing.

Below, we study variations in the amount of reversible error in death verdicts across states, counties and time to help identify factors that lead error to occur. But before focusing on what varies across place and time, it helps to see what is constant across those dimensions. Although the Rocky Mountain peaks crossing central and western Colorado from north to south vary greatly from 9,000 to 14,000 feet, and although higher and lower peaks bunch together in places, their chief attribute is not variation, but constancy. They are all high. If one is looking for temperate winters, easy passage or long stretches of sandy beach, the same answer applies: You're in the wrong place.

The same is true for error rates in death verdicts nationwide, from 1973 to 1995, as the Figures discussed in this section reveal. Despite significant differences at the top of the range, rates of error in capital verdicts almost always are high, no matter when the death verdicts were imposed during the 23—year study period, where they were imposed among the 34 states studied, and which court system—state or federal, trial or appellate—reviewed them. High error rates are not the problem of a few atypical jurisdictions or periods with unusually flawed verdicts or unduly meticulous judges. They are the norm—the solid and unrelieved massif underneath the irregular peaks and valleys; a feature of the modern death penalty terrain so chronic and disturbing that it consistently commands the attention and unfavorable decisions of reviewing judges.

a. Consistency across states.

Figure 1A, p. 50 below, compares overall rates of reversible error found in the 26 states with death verdicts finally reviewed at all three review stages from 1973 to 1995.185 The horizontal line indicates the national overall error rate of 68%. As Figure 1A reveals, 24 of the 26 states with verdicts reviewed at all three inspection stages during the 23—year study period had overall error rates over 50%. In the vast majority of these leading capital states, that is, death verdicts were more likely to fail than pass inspection for reversible error.

Figure 1B extends this analysis to all 34 states in which at least one death verdict was fully reviewed at one or more review stages from 1973 to 1995. It compares the state direct appeal reversal rates of six states that had death verdicts reviewed at only the first review stage,186 to the overall direct appeal and federal habeas reversal rates for two states for which we have data on outcomes at only those two stages,187 and compares all those rates to the overall direct appeal, state post—conviction and federal habeas reversal rates for the 26 states that had one or more death verdict reviewed at all three stages.188 On this comparison, 85% of the states had overall reversal rates of 50% or higher.

A wide variety of additional information on each of the 34 states with active death penalties during the study period is reported in the State Capital Punishment Report Cards in Appendix A.

Figure 1A. Combined Reversal Rate on State Direct Appeal, State Post-Conviction and Federal Habeas, 1973-95

Figure 1B. Combined Reversal Rate for Completed Stages of Review, 1973-95

Figure 1B establishes the order in which states are displayed on most of our comparison graphs below. Because reversal rates are the condition being explained, ordering states based on their reversal rates enables graphs to give a rough and preliminary sense whether there is any relationship between states' reversal rates and the factor studied in the graph (e.g., states' homicide rates).189 That relationship is only roughly indicated in this manner, however, and is much more reliably tested by the regression analyses below for two reasons, among others:

  • Some states have only a few cases reviewed at one or more stages (e.g., Connecticut on direct appeal, Kentucky on federal habeas), while other states have many cases reviewed at all stages. Multiple regression analyses more reliably indicate the strength of relationships between reversal rates and other measured conditions than these graphs, because they give more weight to relationships appearing in states with more reviewed death verdicts than to relationships appearing in states with fewer reviewed verdicts. This is appropriate given the higher degree of confidence we can have in the existence of relationships that are based on many observations as opposed to those that seem to appear after only a few observations.

  • Graphs displaying states in the order of their reversal rates and comparing states based on another factor (say, homicides rates) examine only the relationship between two factors: reversal rates and that particular factor. Often, though, two conditions that seem to be related when compared only to each other turn out not to be related when other factors are analyzed. Conversely, factors that do not seem to be related when compared to each other may turn out to be related after controlling for other factors. For this reason, as well, multiple regressions are more informative than the bivariate comparisons in these graphs.

b. Consistency over time.

i. 1973 to 1995. The regression studies below carefully analyze the effect of time on capital reversal rates after controlling for other explanatory factors.190 Figures 2A—2D, pp. 55—58 below, are a cruder, more preliminary analysis of the effect of time which does not account for other factors.

Figure 2A, p. 55 below, shows the combined reversal rates for death verdicts finally reviewed on state direct appeal and federal habeas in each year from 1973 to 1995.191 Figure 2B, p. 56 below, shows the combined reversal rate on state direct appeal and federal habeas for all verdicts that were imposed in each year and were finally reviewed at those two review stages by the end of the study period. Because it takes years for verdicts to move through each of the two state stages of review and then to be decided at the federal habeas stage, Figure 2B stops in 1989. No death verdict imposed after that year was finally reviewed on federal habeas as of the end of the study period, in 1995.

Both these measures of reversal rates over time are sensitive to two things: changes in how carefully reviewing judges scrutinize capital verdicts for error, and changes in the quality of capital verdicts imposed at trial. Analyses of reversal rates for each study year in which verdicts were finally reviewed are more sensitive to changes in how carefully judges review verdicts for error in different years.192 Changes in reversal rates for each study year in which verdicts were imposed are more responsive to changes in the quality of verdicts over time.193 Here, we are interested in the stability of high rates of capital error, no matter how change over time is measured, so we examine reversal rates for both review year (Figure 2A) and sentence year (Figure 2B). The remainder of the Report focuses on factors related to changes in the quality of death verdicts, and for that reason studies changes in reversal rates based on the year death verdicts were imposed.

Figure 2A shows that reversal rates for verdicts reviewed at the state direct appeal and federal habeas stages in each study year fluctuated during the early years but stabilized around 1983. The combined reversal rate for each decision year thus hit peaks in 1973 and 1975 (100%) and in the early 1980s (86—100%) and a valley in 1976 (32%), but remained at around 60% during most of the last 13 years. In all but one of the last 18 study years, the overall reversal rate for the direct appeal and habeas stages was greater than 50%.

Figure 2B shows consistently high (63—88%) but somewhat declining overall reversal rates for verdicts imposed from 1973 through 1981. After that, reversal rates wobble between 40% (1988 verdicts) and 73% (1989) with an average rate of about 52% from 1982 until 1989.

Our regression analyses below generate important findings about the trend of reversible error over time at the state direct appeal stage after controlling for other factors. The direct appeal stage is particularly important because it accounts for 79% of the reversals during the study period.194 To provide a benchmark for our later analyses, Figures 2C and 2D, pp. 57—58 below, display changes in direct appeal reversal rates over time without accounting for other factors. Figure 2C analyzes direct appeal reversals by the year in which they occurred. Figure 2D analyzes direct appeal reversals by the year in which the reviewed death verdicts were imposed, ending in 1993 because only a handful of verdicts imposed after that year were finally reviewed on direct appeal by the end of the study period.

Both figures reveal a pattern similar to the one noted for overall reversal rates—volatility toward the beginning of the period, with more stable levels of reversals starting in the early 1980s.

Figure 2A: Combined Reversal Rate for State Direct Appeal and Federal Habeas by Year of Review, 1973-95

Figure 2B: Combined Reversal Rate for State Direct Appeal and Federal Habeas by YEar of Death Verdict, 1973-89

Figure 2C: Direct Appeal Reversal Rate by YEar, 1973-1995

Figure 2D: Direct Appeal Reversal Rate by Year of Death Verdict, 1973-1993

The reversal rates by decision and sentence year in Figures 2A—2D do not include state post—conviction error rates. As we note above, because state post—conviction decisions often are unpublished and difficult to find, we only collected information on state post—conviction reversals.195 Although this makes the task of estimating state post—conviction error rates manageable,196 it does not allow us to determine the trend of reversal rates over time for this middle stage of review. Figures 3A and 3B, pp. 60—61 below, are a rough effort to gauge the direction of that trend.

Figure 3A shows that the number of state post—conviction reversals by year tripled as of 1987, then rose again in the early 1990s. To provide a rough sense of whether the number of death verdicts also was increasing at the time—which could explain a rise in the number of reversals, without indicating a rise in reversal rates—Figure 3B indicates the number of death verdicts imposed each year in the study period. Because state post—conviction review generally occurs between four and seven years after a death verdict is imposed, it is possible to project forward from increases in death verdicts to increases in post—conviction review. Figure 3B shows that the number of death verdicts imposed nationally increased substantially from 1977 through 1982 (the spike in 1975 and 1976 was caused by mandatory death verdicts that the Supreme Court wiped off the books in the latter year,197 keeping them from reaching state post—conviction review), but held fairly steady after 1982. Together, Figures 3A and 3B suggest that an increase in the number of verdicts under review may explain the first increase in state post—conviction reversals beginning in 1988, but cannot explain the second increase after 1992. State post—conviction reversals rates thus appear to have risen starting in the early 1990s, at a time when reversal rates at the other two stages were holding steady.

Overall the chronological picture is the same as the geographic one: High error rates have plagued the American death penalty system for years.

Figure 3A. Known State Post-Conviction Reversals by Year, 1973-2000

Figure 3B. New Death Verdicts by Year, 1973-2000

ii. 1996-2001. Because many state post-conviction decisions are unpublished, data on them must be collected by direct contact with informed persons in each capital state. Since we already were making those contacts to enable us to count state post—conviction reversals during the study period, we decided to ask as well about state post—conviction outcomes in the 1996—2000 period.198 As Figures 3A and 3B above reveal, the number and, apparently, the rates of state post—conviction reversals remained at about the average for the 1990s as a whole during the last half of that decade.

We do not have the resources required to collect information on the larger number of decisions approving and reversing death verdicts at the state direct appeal and federal habeas stages since 1995. What information is available suggests that reversal rates may have climbed some or significantly in Florida (state direct appeal reversals), Illinois (direct appeal), Kansas, Kentucky, Nevada, Pennsylvania, Tennessee (direct appeal), Virginia and Washington; held steady in California; dropped slightly (from 77%) in Alabama; and dropped significantly in Louisiana (direct appeal), North Carolina (direct appeal) and Ohio (direct appeal).199 The patterns elsewhere are unknown.

Whether high rates of serious error in capital verdicts during the 23—year study period are a guide to reversals in following years depends on answers to two questions: First, is there reason to think the quality of death verdicts has changed much recently? If systematic steps were taken to improve or undermine the quality of capital trials recently, one might expect a corresponding decrease or increase in overall reversal rates. Absent systematic changes, one might expect error rates to remain near the 50% to 60% plateaus prevailing in the second half of the study period. Second, is there reason to think that the care with which state and federal courts scrutinize death verdicts for serious error has changed substantially during that period? Here we ask about 1996—2001 the same question asked above about 1973—1995: Do court reversals validly reflect the amount of serious error? For the 23—year study period, we have found that court reversals are a conservative measure of serious capital error.200 If the review process has become more careful since then, the correspondence of reversals to serious error might be something closer to one-to-one, as opposed to the current situation where there probably are more serious errors than reversals. If, on the other hand, the review process has become less careful recently, there would be reason to worry that, instead of conservatively indicating how often serious error occurs in capital trials, the number of reversals would provide an inaccurately optimistic picture of the system's reliability.

Although the 2001 legislative cycle may signal a different trend,201 there were only two major nationwide changes in imposing and reviewing death verdicts in the latter half of the 1990s:

  • the 1995 de-funding of resource centers that, from the mid-1980s until then, had provided assistance to capital trial lawyers in most capital states, and had themselves represented death row inmates on appeal and in state and federal post—trial review;202 and

  • the adoption of major pieces of state legislation (e.g., Texas 1995) and federal legislation (1996) imposing strict times limits on state post-conviction and federal habeas review and creating numerous technical barriers to reversal because of waiver and default, non-retroactivity of rights, harmless error, non-exhaustion of remedies, "reasonable error," and successive assertions of rights—even where capital verdicts were demonstrably the result of serious legal violations that substantially undermined their reliability.203

Starting in 1995, the first of these changes immediately increased the probability that lawyers appointed for capital defendants lack the expertise and resources needed to try death cases competently.204 Both changes increased the probability that errors in verdicts imposed before and after 1995 but reviewed after then will not be corrected.205 The changes' combined effect is just beginning to be felt, because flawed verdicts imposed after the withdrawal of defense resources in 1995 are only now reaching the state and federal post—conviction stages that were truncated in that and the next year.

Given all these events, it is reasonable to infer that rates of error seriously undermining the reliability of capital verdicts have not decreased since 1995, and may well have increased, though the rates at which serious error is being detected may have declined in the federal courts and in some state courts, while rising in other state courts including those noted just above.206

A more complete answer to the 1996—2001 question awaits the collection of additional data.

c. Consistency among state and federal court systems.

i. The general rule: similar state and federal court reversal rates. Figure 4, p. 66 below, compares the rates of reversible error discovered by state courts on direct and post-conviction review and by federal courts on habeas review in the 26 states where we have data on verdicts fully reviewed at all three review stages during the study period. The states are arrayed to reveal those where the rates of reversible error discovered by state courts (the grey line) and federal courts (the black line) are similar or different. In the middle range of the graph are states where state and federal reversal rates are similar; towards the left and right margins are states with successively larger differences between the two rates. In states on the left side, state court reversal rates exceed federal court rates; in states on the right side, federal court reversal rates exceed state court rates. Figure 4 shows that:

  • Differences between states in regard to their overall—state plus federal—reversal rates are greater than differences within states, comparing their reversal rates in state vs. federal court: If a state's verdicts tend to be reversed relatively more often by state courts, they also tend to be reversed relatively more often by federal courts. Even the actual rates of state and federal court reversals are similar. This is especially so if one removes Kentucky and Tennessee (far right margin), which had only one death verdict reviewed on federal habeas during the study period—not enough to give a good picture of federal reversal rates—and when one notes that both North and South Carolina (far left margin) are in the Fourth federal judicial Circuit which has very low capital reversal rates compared to all other circuits.207

  • Figure 4 distinguishes (1) states where state and federal courts roughly agree on the amounts of error in capital verdicts (in absolute terms, and relative to the amounts of error other courts find in verdicts from other states) from (2) states where there the state and federal courts disagree. Notably, states where local and federal courts roughly agree outnumber states where the two sets of courts do not see eye to eye. This, again, suggests that in general it is not the different dispositions of different types of judges that determine whether they find reversible error. Instead, it is the presence of clear and serious error that drives judges—state and federal, elected and life—tenured—to reverse death verdicts.208

ii. A few exceptions: federal court compensation for lax state court review, and vice versa; California and Georgia vs. North and South Carolina. Even where state and federal courts differ in the amounts of error they find in capital verdicts, they do not differ in ways that are predictable based on judges' status: State courts reverse higher proportions of their own verdicts than federal courts in 11 states; federal courts reverse higher proportions than state courts in 13 states; reversal rates are identical in 2 states. Considering all verdicts nationally, state court reversal rates are 46%, compared to 40% for federal courts. This, and Figure 5, p. 68 below, suggest that disagreement in amounts of error detected is not a function of the different proclivities of state as opposed to federal judges and, instead, that inter—court compensation may be occurring:

  • In reviewing verdicts from some states, federal courts may find it necessary to reverse at relatively higher rates because the courts of that state are unusually tolerant of error. This may be the case in, e.g., California and Georgia. As Figure 5 reveals, both states have especially high federal court reversal rates and atypically low state court reversal rates.

  • Figure 5 suggests the reverse pattern in North and South Carolina. There, state courts may exercise more care because they know that death verdicts receive atypically forgiving review in the relevant federal court (the Fourth Circuit).

This, again, suggests the stability and seriousness of error in capital verdicts: If one set of courts (state or federal) is unusually willing to let error pass—e.g., California state courts, where harsh political discipline has been exacted against judges thought to have reversed too many death verdicts209—the seriousness of error slipping through that system's review process may pressure other courts to redouble their efforts to catch error—as federal judges report having to do when reviewing California death verdicts.210

Figure 4. Percent of Capital Verdicts Reversed on State Direct Appeal or State Post-Conviction and on Federal Habeas, 1973-95

Figure 5. Percent of Death Verdicts Reversed by Selected State Direct Appeal and Federal Habeas Courts, 1973-95

iii.. Virginia: low state and federal reversal rates. As Figure 5 illustrates, Virginia has the unique confluence of state and federal reviewing courts with the lowest rates of error detection in capital cases among state and federal courts nationally. Together, those courts generate an overall error rate for Virginia (18%) more than two standard deviations below the mean for the other states in Figure 1A above: Less error is detected in Virginia capital verdicts than the experience of other states suggests should be the case. This could mean there are disturbingly low rates of error detection in Virginia capital cases—given the confluence of state and federal court systems with high tolerances for serious error and no desire to compensate for the other system's lax review.211 This explanation is suggested by comparing Virginia to the three neighboring states in the same (Fourth) federal judicial circuit. In Maryland, North Carolina and South Carolina, atypically low federal court reversal rates are compensated for by unusually high state court reversal rates.212 Alternatively, Virginia death verdicts may have commendably low rates of error—the likes of which other states have not come near achieving given Virginia's unique traits. We come back to Virginia below.213 Here, we simply note two warning signals from emergency systems that catch some errors missed in the normal course of state and federal court review:

  • The U.S. Supreme Court can but rarely does grant review of death verdicts after state or federal circuit court review. From 1995 to 2001, it did so in fewer than 25 cases. But a substantial majority of those cases were from only 1 of 50 states—Virginia—or 1 of the 12 federal judicial circuits—the Fourth, including Virginia—or (usually) from both. The reasons for this disparity are hard to know (the Court doesn't explain decisions to grant or deny review), but our study suggests one reason: In the unique situation of Virginia, where both the state and the federal courts are unusually disposed to let capital error pass—and where neither has beefed up its review to compensate for the other's review—it has fallen to the court of last resort to provide the inspection function that elsewhere is performed by state courts, lower federal courts, or both.214

  • During the last decade, Virginia Governors have spared two inmates from execution under suspect convictions that state courts on direct and post—conviction review and the courts on habeas review had declined to cure through the regular judicial channels.215 This experience (which is duplicated in no other state) also suggests a break—down in the court system for containing the effects of serious capital error, and a need to rely on fall—back systems that are not designed to provide routine protection against unreliable death verdicts.

iv. The high risk that courts miss some serious error. High reversal rates at the last as well as the first review stages reveals a substantial risk that appellate courts cannot catch all serious errors that compromise capital verdicts. Multiple inspections are designed to increase confidence in the outcome, not only by sending flawed products back to be reworked or scrapped, but also by showing that prior inspections have done their job. To serve the latter goal, however, the amount of error found at successive review stages must drop substantially, with the last stage finding no or only infrequent errors. By this measure, the capital inspection system is a not working properly:

  • The 41% — 10% — 40% pattern of reversal rates at successive stages of review does not reveal the downward trend needed to create confidence that prior review stages catch most serious errors. Indeed, our regression analyses identify types of error that the first two review stages do a poor job of catching, leaving it to the last stage to fix.216

  • The 40% rate of serious error found at the last review stage undermines confidence that no or few errors will pass that final checkpoint and go on to cause serious harm. This is especially so given evidence in our regression study that federal reviewers have blind spots of their own.217

    The 3-tier review process reveals the size of the problem of serious capital error. But that process does not solve the problem. Instead, it leaves a high risk that some seriously flawed verdicts are approved for execution and carried out. The problem may be especially acute in states like California, Georgia, North Carolina and South Carolina where there is evidence that one review stage is having to compensate for the weakness of other review stages, and could be most worrisome in Virginia, given the potential laxness of both state and federal review.