Figure 9

Figure 10

Especially when we put to one side the (asterisked) states that had too few capital habeas reviews to permit analysis, both charts reveal a strong degree of similarity between the rates of capital-sentencing error detected by state and federal courts in each state (i.e., in how close together the two lines are). More important is the even stronger degree of similarity between state and federal courts' judgments about the various state's comparative rates of capital-sentencing error (i.e., in how closely each line's upward and downward ticks as it moves from one state to the next are paralleled by the other line's upward and downward ticks). What Figures 9 and 10¹⁹⁵ thus suggest is that state and federal courts examining the same pools of capital judgments generally find-and react similarly to-the same relative levels of serious capital-sentencing error. In plain English: Where state courts find comparatively high, low or average rates of error in a particular jurisdiction's capital judgments relative to error rates found elsewhere, so do the federal courts reviewing the same jurisdiction's capital cases. Figures 9 and 10 thus refute the notion that elected state judges as a group react differently to the possibility of error in capital cases from the way that federal judges react as a group. In fact state and federal judges' reactions to capital error on both these measures of comparative amounts of error¹⁹⁶ are very much in sync.

That said, it is interesting to consider the relatively small numbers of states that fall on the left and the right edges of the chart where the state and federal error-detection lines diverge. In doing so, we focus on Figure 10 (the more informative of the two charts¹⁹⁷) and on the (non-asterisked) states with sufficient numbers of federal habeas cases.

One interpretation of Figure 10 is that the courts of North Carolina, South Carolina, Louisiana, Florida and Alabama-the states on the left side of the chart-are doing the lion's share of error detection for capital judgments in those states, leaving significantly less error to be detected by the relevant federal courts. Alternatively, the courts of those five states may have increased their level of vigilance to compensate for what they perceive (based, e.g., on past experience and (more probably) on information transmitted by lawyers) to be unusually lax error-detection by the federal courts. This latter interpretation might explain the North and South Carolina courts' robust error-detection in capital cases. Both states fall within the United States Court of Appeals for the Fourth Circuit, which has by far the lowest capital-error detection rate of any federal judicial circuit in the country.¹⁹⁸

The corresponding interpretation for Georgia, Montana and California-on the right side of the chart-is that federal courts have taken the lead error-detection role as to capital judgments from those states to compensate for low state court error-detection.

The hypotheses offered in the preceding two paragraphs present important questions for future research.

We conclude our discussion of Figure 10 by again noting a discrepancy between Virginia and the other states. Unlike almost every other state (Missouri, again, and Texas are in an intermediate category) Virginia's state-review "square" and its federal-review "circle" are both located at the very bottom of the chart. In this respect, the Virginia courts may be contrasted to those of the other states in the Fourth Circuit, which are discussed on pp. 51 and 65 above: unlike the courts of the neighboring states, there is no evidence that Virginia's courts have tried to compensate for very low error detection by the Fourth Circuit. Quite the contrary, Virginia courts have the lowest error-detection rates of the 28 study states. As a consequence of simultaneously low state and federal error detection, the rate of error detected in Virginia capital judgments is both extremely, and unusually, low.

F.   Overall Rates of Serious Error Found on State Direct Appeal, State Post-Conviction, and Federal Habeas Corpus

Figure 12

Table 10 and Figure 12 reveal that:

• For the study states as a whole, the overall rate of serious error was 68%.²⁰¹
  
  
• Overall error rates vary enormously, from 18% in Virginia to 91% in Mississippi.²⁰²
  
  
• All but two states (Virginia and Missouri) had overall error rates of 52% or higher. All but four states (those two, plus Texas and Pennsylvania) had overall error rates of 64% or higher.
  
  
• Put the other way around, only two states out of 26 produced capital judgments that passed inspection for serious error more than half the time.
  
  
• Numerous states, in all sections of the country-including Alabama, Arizona, California, Georgia, Indiana, Mississippi, Montana and Oklahoma-had error rates of three-quarters or more, with Mississippi's being more than 9 out of 10 (a success rate of less than 1 in 10).
  
  
• As noted above, the Governor of Illinois cited evidence of high rates of serious error in Illinois capital judgments, and particularly a spate of exonerations of innocent men released from death row, as the reason for declaring a moratorium on executions there.²⁰³203 This prompted other policymakers, including the Governors of Florida and Texas, to suggest that actions in Illinois are not relevant elsewhere, because high error rates are unique to Illinois.²⁰⁴ In fact, the rate of error detected by state and federal courts in Illinois capital sentences, 66%-while high in absolute terms-is not at all unique. On the contrary, the Illinois error rate is very close to, and a bit lower than, the national average of 68%.
  
  
• As one would expect from our previous discussion, and as Figure 12 demonstrates, Virginia is a distinct outlier here, falling almost literally "off the charts" on the low side of error detection. Virginia's overall rate of detected error is barely half that of the next closest state (Missouri, which itself is much lower than all the other states), and barely a quarter the national rate. In technical terms, Virginia's overall-error detection rate is nearly 3 standard deviations below the mean (2.88).
  
  

Figure 13 below plots (1) the combined state direct appeal and state post-conviction reversal rates, (2) the federal habeas reversal rate, and (3) the overall error rate that is a composite of the other two.²⁰⁵ Figure 13 illustrates three points that (for the most part) we have discussed above:

• High overall error rates across most states.
  
  
• Similar state and federal patterns of error detection in most states, with some exceptions where high state court error detection compensates for low federal court error detection (e.g., North and South Carolina state courts compensating for the federal Fourth Circuit), and vice versa (e.g, the federal Ninth and Eleventh Circuits compensating for low state court error detection in California and Georgia, respectively).
  
  
• Virginia's outlier status.
  
  

Figure 13

G.   Length of Time of Review

Figure 14²⁰⁷ below compares states on the basis of how many years elapsed between each state's first death sentence and its first non-consensual²⁰⁸ execution (not necessarily in the same case):

• In 16 (57%) of the 28 study states, it took (or will take²⁰⁹) 15 or more years to get from the state's first death sentence to its first execution following full review.
  
  
• In 71% of the states it took 10 or more years.
  
  

Figure 15²¹⁰ compares the 23 study states in which at least one execution (consensual or non-consensual) took place between 1973 and 1995 based on the amount of time that elapsed, on average, between the same prisoner's death sentence and execution. Subject to missing data, and the fact that the table counts consensual executions, which causes it to understate the time needed for full review,²¹¹ Figure 15 reveals that:

• In the vast majority of states, executions took place on average 7 or more years after death sentences during the study period.
  
  
• In over two-thirds of the states, executions took place an average of 9 or more years after the death sentence was imposed.
  
  

Figure 14

Figure 15

Figure 16²¹² below compares states based on the proportion of their 1973-1995 death sentences that were awaiting direct review in 1995. As is discussed above, this comparison provides a rough measure of the extent to which state direct appeal is a bottleneck in the inspection process.²¹³ Nationally, 21% of capital sentences imposed between 1973 and 1995-about 5-years-worth of death sentences-were awaiting direct appeal in 1995:

• In over a third of the 28 states, 20% or more of all post-Furman death sentences were backed up at the state direct appeal stage 23 years after Furman.
  
  
• In three of the nations most prolific capital-sentencing states, Texas, Pennsylvania and California, the 1995 log-jam of cases awaiting state direct appeal contained (respectively) 27%, 27% and 47% of the state's post-1972 cases. In Washington and Wyoming, the 1995 logjam contained 45 and 70% of the post-Furman cases.
  
  
• Of note, although the federal Ninth Circuit Court of Appeals is sometimes blamed for holding up executions in the states within its jurisdiction, the three states in that circuit with the largest death rows-California, Arizona and Washington-were all in the top cohort of states as of 1995 in terms of the proportion of cases bottled up in the state courts awaiting direct appeal.
  
  

Figure 16

H.   Capital-Sentencing and Execution Rates, and the Two Compared

Figure 17 below compares states based on their death sentencing rates per 1,000 homicides, per 100,000 population and per 1,000 prisoners. Figure 18 below compares states based on their non-consensual execution rates per the same three populations.

Figure 17

Figure 18

Figures 17 and 18 reveal huge variations among states in both their death-sentencing rates and their execution rates measured per homicides and per population:²¹⁵

• Measured against both populations, some death-sentencing states have death-sentencing rates that are 10 times those in other death sentencing states.
  
  
• In Wyoming, for example, nearly 6% of all homicides result in a death sentence-over four times the national average for death-sentencing states. In Maryland, less that six-tenths of 1% of homicides lead to a death sentence.
  
  
• Nevada condemns nearly 11 people out of every 100,000-about three times the national average for death-sentencing states. Washington State does so to less than 1 person out of every 100,000.
  
  
• Similar disparities characterize the execution rates in the various death-sentencing jurisdictions.
  
  
• The disparities among states in death sentences and executions per 1,000 homicides are particularly interesting, revealing the absence of what one would expect to be a consistent relationship between homicides and capital punishment.
  
  

Figures 19-21 below consider whether high (or low) death-sentencing rates (per homicide, per population or per prisoners) translate, as one would expect, into high (or low) execution rates.

Figure 19

Figure 20

Figure 21

Judging from figures 19-21, there is no relationship between death-sentencing and execution rates. When states are arranged in order of their death sentences per capita, the line representing their executions per capita fluctuates wildly and randomly:

• Idaho, Nevada, Arizona and Oklahoma rank 2nd, 3rd, 4th, and 6th (and range from 3 to 4 times the national average) when it comes to how often homicides result in death sentences. Those same states, however, are tied for 23rd, tied for 24th, 17th, and 14th among 28 states (near or well below the national average) when it comes to how often homicides result in execution.
  
  
• On the other hand, Texas, Virginia and Louisiana rank 18th, 22nd, and 25th in death sentences per homicide (ranging from slightly above, down to two-thirds, the national average) but 4th, 2d, and 7th in executions per homicide (ranging from over twice to nearly four times the national average).²¹⁶ Thus, the three states most associated in the public's mind with executions- Louisiana, which was the nation's execution capital in the late 1980s,²¹⁷ and Texas and Virginia which claimed that distinction in the 1990s²¹⁸-did not attain that status by sentencing disproportionately large numbers of people to death row. Instead, they have done so by translating below-average death-sentencing rates into above-average execution rates.
  
  

Figure 22 below asks a related question: Are states that are most likely to punish homicides with death also most likely to translate death sentences into executions?

Figure 22

Figure 22 reveals no relationship between death sentencing and execution rates. Indeed, for nearly half the states-Louisiana, Virginia, Missouri, and Texas (with comparatively low death-sentencing but high death-sentences-carried-out rates) and Wyoming, Idaho, Nevada, Arizona, Oklahoma, Florida, Alabama, and Mississippi (with comparatively high death sentencing rates but low death-sentences-carried-out rates), the relationship is the inverse: the more frequently states sentence killers to die, the less frequently they execute them, and vice versa.

Overall, therefore, it seems clear that a powerful disposition to sentence offenders to die does not go hand in hand with a strong capacity to carry out the death sentences that are imposed. Figuring out why this is so is a question we will address in a subsequent report. Our analysis so far, however, suggests one place to look for the source of the discrepancy: the distribuingly high rates of capital-sentencing error that we document above.

I.   Demographic Factors

Figures 23 and 24 below consider the relationship between homicide rates per 100,000 population and, respectively, capital-sentencing and execution rates.

Figure 23

Figure 24

If there is any relationship at all between homicide and capital-sentencing rates (a matter requiring more sophisticated analysis), Figure 23 suggests that it is weak and inverse. Figure 24 asks whether variations in rates of serious crime, as measured by homicides per 100,000 population, can explain variations in execution rates, or vice versa. Figure 24's decisive answer is that there is no such relationship between a state's serious crime rate and its willingness or capacity to execute its citizens.

Turning to the issue of race, Figure 25 below compares capital-sentencing states' relative death-sentencing rates (per 1,000 homicide) to their percent nonwhite population.

Surprisingly, perhaps, this chart suggests that proportionately larger minority populations are associated with somewhat lower death-sentencing rates, and vice versa. Figure 25 also reveals the sharp variation among capital-sentencing states in terms of the proportion of their populations that are nonwhite, ranging from 5% in Idaho (which, incidently, has a very high death sentencing rate per homicide) to 37% in Mississippi (where the death-sentencing rate per homicide is relatively low).

Figure 26 below considers whether race influences execution, as opposed to death-sentencing, rates. Here, the relationship is weaker than in Figure 25, and runs in the opposite direction: Although states with larger proportions of racial minorities tend to capitally sentence less often than states with proportionately smaller minority populations, those same states tend to carry out relatively more of the death sentences they impose.

Figure 25

Figure 26

J.   Court Factors

Figure 27 and Figure 28 below consider the impact of political pressure on, respectively, death-sentencing and execution (more specifically, death-sentences-carried-out) rates. Because error rates and the rates at which death sentences are carried out are so highly correlated (see Figure 1, supra p. 11), the latter chart is also a rough measure of the relationship between political pressure and capital error rates.

Figure 27

Figure 28

Figures 27 and 28 reveal a curious and potentially significant pattern: In general, the more electoral pressure a state's judges are under, the higher the state's death-sentencing rate, but the lower the rate at which it carries out its death sentences. Assuming a causal relationship, this suggests that political pressure tends to impel judges-or to create an environment in which prosecutors and jurors are impelled-to impose death sentences, but then tends to interfere with the state's capacity to carry out the death sentences that are imposed.

Whether it is fair to infer a causal relationship here and, if so, what might account for that relationship is a question for further research. One hypothesis is suggested by possible relationships between high death-sentencing rates and high error rates, and between the latter and low execution rates: Public opinion may place a premium on obtaining death sentences.²²⁴ If so, a desire to curry favor with voters may lead elected prosecutors and judges to cut corners in an effort to secure that premium- simultaneously causing death-sentencing rates, and error rates, to increase. In that event, high rates of reversible error would explain why high political-pressure states, after imposing so disproportionately many death sentences-making so many errors in the process-end up carrying out so disproportionately few of their death sentences. These are questions for further research.

Figures 29 and 30 below relate, respectively, states' death-sentencing rates, and the rates at which they carry out death sentences, to their per capita court expenditures.

Figure 29

Figure 30

With some exceptions, Figure 29 appears to indicate that comparatively high expenditures on courts are associated with relatively high death-sentencing rates. It is difficult to know what to make of this relationship, especially because capital cases are themselves costly and thus may partly account for high expenditures. It may be, however, that states whose courts have substantial amounts of resources are more capable of handling capital cases-and thus do so more often-than states with less well-funded courts.

As was the case when we looked at capital punishment and political pressure, the relationship between capital punishment and spending reverses when we move from analyzing death sentencing rates to rates of death sentences carried out: Figure 29 shows a direct relationship between court expenditures and death sentencing (the higher the one is, the higher the other tends to be); by contrast, Figure 30 shows a weak inverse relationship between court expenditures and death sentences carried out-as states' spending on their courts increases, the proportion of the death sentences imposed that are carried out tends to decrease. The cause of that relationship (if any exists) is unclear. If, however, it were the case that the processing of death cases is itself responsible for significantly driving up court expenditures, then Figures 29 and 30 might suggest that spending relatively large sums to secure relatively large numbers of death sentences has little pay off-and, indeed, is counterproductive-when it comes to securing executions. If so, the policy alternative of spending less by securing fewer death sentences²²⁵-each of which, however, is more likely to be carried out-would be indicated.

Figures 31 and 32 below consider the relationship between state court caseloads and, respectively, death sentencing rates and the rate of death sentences carried out.

Figure 31

Figure 32

Judging from Figure 31, there is no relationship between how many cases per capita state courts handle and the rate at which those courts impose death sentences. Figure 32 does, however, suggest a weak relationship between court caseloads and death sentences carried out: As per capita caseloads drop, the rate of death sentences carried out also tends to drop. One might hypothesize that states with smaller courts (ones with lower caseloads) are more likely to generate seriously flawed death sentences at the trial level, thus depressing the rate at which their death sentences are carried out. Alternatively, state appellate courts with lower caseloads may be superior error detectors, thus (given high error rates across all states) accounting for lower rates of executions-or, in this scenario, lower rates of flawed executions. Further research is called for.