next | previous | table of contents

D. Stages of Analysis and Factors Considered

Our analysis proceeds in stages based on the capital reversals being explained. First, we study factors explaining differences in state rates of reversed capital verdicts, starting with our main binomial Analysis 1 and an otherwise similar Poisson analysis (Analysis 2), then conducting several follow-up analyses. Still more follow-up analyses identify explanations for differences in county capital reversal rates, starting with explanations operating at the county level, then testing factors operating at both the state and county level. Finally, using a large data set from all capital federal habeas cases decided between 1973 and 1995, we identify traits of capital cases that increase the probability of federal habeas reversals.

In this section, we explain why state, county and case-level factors, and time, are all worth considering as explanations for serious capital error. Then we describe the range of factors considered. Finally, we discuss our sources of information about those factors.

    1. Why study state, county and case-level explanations for serious capital error?

In identifying explanations for reversible capital error, we consider conditions operating (1) throughout each of the 34 states that imposed and reviewed death verdicts in the study period, (2) in each of about 1000 capital-sentencing counties in those states and (3) in nearly 600 capital federal habeas cases decided in the period. Examining factors at multiple levels is useful for several reasons.

  • Some factors operate at only one level and may be missed if only another level is studied.

  • Even if a factor operates at all levels, data on it may be available at only one level. For example, the amount spent on individual trials and counties' aggregate spending on courts is generally not known, but figures on state spending are available. Conversely, states and counties do not rate the "average aggravation levels" of their homicides, but juries make findings on that subject in each capital trial. To see if court spending and the aggravation level of crimes affects reversal rates, it thus is necessary to study both the state and case level.

  • The effect of conditions that operate at one level may be detected at, and misattributed to, another level if only the latter is studied. Error rates may appear to be reacting to county homicide rates, for example, if only county-level effects are studied. But that effect may disappear if the effect of state homicide rates is simultaneously considered and is significant. A way to identify the source and extent of effects that may arise at two levels is to study both.

  • Data for entire counties and states are relatively easy to obtain from public agencies and private researchers, but their aggregate character may obscure subtleties. In contrast, case-level data may reveal a lot about each event but are far more difficult and expensive to gather for all cases. Taking both a wide view of aggregate data blanketing entire jurisdictions and periods, and a close look at hundreds of details about a smaller set of cases, achieves the advantages of both kinds of data.

      a. Why study the effect of state conditions?

It is sensible to expect that conditions at the state level might affect capital error rates because the existence and operation of the death penalty is largely a result of state-level policies:

  • State legislators decide whether to adopt a death penalty law giving local prosecutors and juries the authority they would otherwise lack to seek and impose death sentences.

  • State legislators define the offenses for which death is a legal punishment, including by listing the aggravating and mitigating circumstances that prosecutors may or must consider, and by stating which individuals are and are not subject to the death penalty (retarded individuals?, juvenile offenders?, co-participants who were not directly involved?, etc.). Some states have broad death-sentencing laws; others have narrow ones. Some states give D.A.s and jurors broad discretion over whether to seek and impose death sentences; others give little discretion.

  • The court system that administers the death penalty is by and large a state system. That system uses procedures set by the state constitution and laws, which apply throughout the state.

  • The state's single supreme court has the final say on the meaning of the state's death-penalty law. Its rulings bind prosecutors, judges and capital trials all over the state. The law in many states requires that court to assure statewide uniformity in capital practices and outcomes

  • Some states fund most or all costs of prosecuting and defending capital cases at local trials. Statewide public defenders or capital resource centers represent many or all capital defendants and appellants in a minority of states, and the laws of other states govern who may be appointed to represent capital defendants and how much they can be paid. These states may be contrasted to states that require localities to fund all capital litigation, at least for trials, and that leave it to localities to decide whom to appoint to represent capital defendants and how much to pay them.

  • In many states, the governor, on behalf of the entire state, nominates judges subject (in most states) to voter approval. In many cases—including in the selection of most state supreme court justices—the electorate to whom state judges answer is statewide, not local.

  • State attorneys general set policies that bind or affect decisions by local officials on whether to bring capital cases and how to try them. In most states, the state attorney general defends all capital verdicts on appeal and in state post-conviction and federal habeas proceedings, giving that official considerable power to define acceptable local as well as state capital charging, conviction and sentencing policies, by deciding whether, and how vigorously, local policies will be defended when attacked on appeal.

  • Governors and state boards of pardons and parole establish criteria for executive clemency in capital cases that apply throughout the state.

  • The public and political processes through which state legislators, judges, attorneys general, governors and other state officials make these decisions create a common statewide approach to the death penalty that differentiates each state from every other. State actions and policies lead officials and jurors throughout the state to use the death penalty differently from actors in other states.

As a matter of observable fact, moreover, states vary widely in how they use the death penalty, and in how their death verdicts are perceived by reviewing courts—providing further justification for studying possible state-level explanations for serious capital error. To start with, as Figure 11, p. 121 below shows, states differ greatly in how often they impose death verdicts per 1000 homicides. States also differ greatly in how often serious error led to reversal of their death verdicts during the 23-year study period when their verdicts were reviewed by state supreme courts on direct appeal (see Figure 12, p. 122 below), all state courts combined at both the direct appeal and post-conviction stages (Figure 13, p. 123 below), or federal courts on habeas corpus (Figure 14, p. 124 below).

Figure 11. Death Verdicts per 1000 Homicides, 1973-95*

Figure 12. Percent of Death Verdicts Reversed on State Direct Appeal, 1973-95*

Figure 13. Percent of Death Verdicts Reversed on State Direct Appeal onr State Post-Conviction, 1973-95*

Figure 14. Percent of Death Row Verdicts Reversed on Federal Habeas Review, 1973-95*

As Figure 15, p. 126 below shows, capital verdicts from different states have different rates of serious error even when the same judges inspect them. Federal court of appeals judges are assigned to circuits with responsibility for cases from several adjoining states. Figure 15 compares federal habeas reversal rates for capital verdicts from different states in the same circuit, revealing large differences in the amount of error the same judges find when reviewing death verdicts from different states. Figure 15 supports a point made above: There evidently is something about capital verdicts—and especially about the states that impose them—and not simply something about particular judges that determines how often the judges reverse capital verdicts.290

When all serious, reversible error found at all stages of review completed by at least one death verdict from each state are combined, the resulting overall error rates again vary dramatically from the top to the bottom states, as Figure 16, p. 127 below, illustrates.291 States also vary dramatically in:

  • how long it took them to get from a first death verdict to a first non-consensual execution (Figure 17, p. 128 below), and how long it took on average for an imposed death verdict to be carried out (Figure 18, p. 129 below)Croughly gauging how long capital appeals take;

  • the size of their backlogs of unreviewed death verdicts (Figure 19, p. 130 below);

  • how many people states non-consensually execute for every 1000 homicides—ranging from .02 in California to 4.8 in Delaware (240 times greater) (Figure 20, p. 131 below); and

  • the proportion of each state's death verdicts imposed during the study period that it carried out in the period (Figure 21, p. 132 below).292

Comparing Figure 11 (p. 120 above) to Figures 20 and 21 ( pp. 131-32 below) reveals no evident, positive relationship between states that impose, and ones that carry out, high proportions of death verdicts per homicide. In some cases, the relationship appears to be negative: Arizona, Idaho, Mississippi, Nevada, North Carolina and Pennsylvania are among the most likely to impose death verdicts per homicide and the least likely to carry them out. Louisiana, Missouri, Texas and Virginia imposed relatively few death verdicts per homicide but were among the most likely to carry them out.

Figure 15. Percent of Death Verdicts Reversed on Federal Habeas by Selected Circuits and Their States, 1973-95*

Figure 16. Combined Reversal Rate for Completed Stages of Review, 1973-95*

Figure 17. Years from First Death Verdict to First Nonconsensual Execution, 1973-95*

Figure 18. Average Years form Death Verdict to Execution, 1973-95*

Figure 19. Average Number of Death Verdicts Awaiting Review, 1973-95*

Figure 20. Nonconsensual Executions per 1000 Homicides, 1973-95*

Figure 21. Percen Death Verdicts Carried Out (Nonconsensual Executions)

      b. Why study the effect of county conditions?

Conditions at the local level also affect capital sentencing and may affect capital error:

  • In some states, local prosecutors are free to decide whether to use the death penalty at all. Even where they lack that power, they usually have broad discretion over whether to charge cases capitally and whether to accept plea bargains that avoid the death penalty.293

  • When state statutes and supreme court decisions are ambiguous, local trial judges must interpret the law as best they can—sometimes defining different views of the law in different places. Local judges often are elected locally, which puts different political pressures on them and their interpretations, rulings and instructions to jurors.

  • The quality of the criminal defense bar differs from county to county. So do policies about whom to appoint to represent defendants in states where local officials decide that question.

  • The ultimate decision maker in capital cases is the jury (or, in a minority of states, the trial judge). Local conditions may affect how jurors (and trial judges) approach that decision.

States differ in how they define the areas over which each district attorney and trial court has responsibility, and from which potential jurors in capital cases are drawn. In many states, the area of responsibility for each set of decision makers is different. A district attorney may be responsible for prosecuting capital cases in several adjoining judicial districts, while jurors in resulting trials might come from only a subdivision of one of those districts. The relevant areas change over time, and there is no national or even state repository of information about their boundaries. The closest approximation to these areas of responsibility with stable and well known boundaries are counties. Most districts over which district attorneys and local courts have responsibility, and from which jurors are chosen, are defined in relation to county boundaries (very often being coterminous with counties, and, if not, splitting or combining them). Like states, moreover, counties are good bases for comparison because public and private agencies often organize information they collect by county.

States influence counties and are themselves aggregations of counties. In a number of analyses, therefore, we use analytic techniques that either treat counties as "nested" within states (meaning counties within states are assumed to be more similar to each other than to counties outside the state) or define state conditions as a composite of conditions in all their death-sentencing counties.

      c. Why study the effect of case-level conditions?

Above we explain why understanding capital error requires knowing more than the error found in each case.294 We have just explained why identifying state- and county-level factors related to high capital error rates can help fill out our knowledge of conditions associated with error. Additional information can be provided by traits of particular cases—beyond the error found in each case295—that are significantly related to a lower or higher probability that the verdicts will be reversed. As a result, the trial-level conditions appellate judges describe based on the trial transcripts and other records before them, and the circumstances of the appeals, are worth studying for clues to why errors arise in some cases but not others and, indeed, in some counties, states and periods but not others. Studying traits of particular capital cases supplements state and county analyses in a number of important ways:

  • Case-level analysis may identify explanations for capital error that cannot be tested at the jurisdictional level, because states and counties don't differ much in that respect, but cases do. For example, rates of homicide victimization among women may be similar from state to state, making it difficult at that level to study how the victim's gender affects capital error. But the victim's gender does vary from case to case, so the factor can be tested at the case level.

  • There may be better data about particular conditions at the case as opposed to the jurisdiction level. Not much can be learned about the attributes of appellate lawyers at the county and state level, because no relevant records are kept at those levels. By examining court decisions in each case, however, one can learn the names of the lawyers who represented capital defendants, whether they are private or publicly employed, and where their offices are located.

  • As we note below in discussing racial characteristics related to high capital error rates, a case-level analysis can provide important information about factors that were effectively tested at the jurisdiction level. The racial makeup of a state's population may have an important relationship to high capital error rates, though the racial makeup of the state's death row, or the race of particular defendants or victims, is not significantly related to the probability of error.296 The same may be true for other traits such as the seriousness of crime: The fact that states with high rates of serious crime such as homicide, or at least high rates affecting influential residents, have high capital error rates does not assure that death verdicts imposed for especially serious homicides are more likely to be reversed. That may be true—and, if so, it says something about how serious crime affects error rates. But it also may not be true, or the opposite may be true—that verdicts for more serious offenses have less capital error. Either of the latter findings would say something else about how the seriousness of a jurisdiction's crime problem affects capital error rates.

    2. Explanatory factors studied.

In the 10 years since this study began, we compiled the largest data base ever assembled on factors at the state, county and case level that might lead to the occurrence and discovery of reversible error in capital verdicts. We believe it is the largest data base ever assembled on factors contributing to the existence and discovery of error in any type of judicial proceeding.

Our choice of factors to study was guided by this question: What types of conditions at the state, county and case level might plausibly affect how much serious error trial actors make in imposing capital verdicts and how much error appellate courts discover when they inspect those verdicts? To answer this question, we reviewed the literature on capital sentencing and judicial decision making and consulted widely among lawyers, judges, criminologists and other social scientists. We also examined available sources of data on states, counties and court cases, asking in each case if there were plausible reasons to think that the category of available data might provide clues to the causes of reversible capital error. Below we list the categories of explanatory factors we considered in our state and county-level analyses and in our case-level analysis. For a more comprehensive discussion of the explanatory factors we considered, see Appendix E.

      a. State and county-level factors.

Table 2 lists categories of state and county factors we considered in our state and county analyses. Unless indicated, we considered state and county-level factors of each type. With some exceptions, we were able to collect data in each of these categories on all or nearly all states and counties in each of the 23 study years.


Table 2: Categories of Factors Considered in Analyses
Explaining Differences in State and County Capital Reversal Rates

    I. Capital Judgements and Appeals

    1. Number and Rates (per homicide) of Capital Verdicts
    2. Number and Results of Direct Appeals
    3. Number, Results and Retrial Outcomes of State Post-Conviction Proceedings
    4. Number and Results of Federal Habeas Review
    5. Number and Rates of Undecided Appeals
    6. Presence or Absence of Death Penalty Resource Center (state only)
    7. Post-Reversal Capital Reprosecution Rates
    8. Index of Timing and Frequency of Use of Death Penalty

    II. Death Row Population (state only)

    1. Race (including relative to state population)
    2. Race of Victim (including relative to state population)

    III. Functioning of the Judicial System Generally (state only)

    1. Caseloads
    2. Dispositions
    3. Backlogs
    4. Expenditures

    IV. Political Pressure on the Judiciary (state only)

    1. Selections and Retention Methods
    2. Party and Ideological Influences

    V. State Demographic Characteristics

    1. Population
    2. Race
    3. Population Density and Urbanization
    4. Median Age

Table 2 (cont=d): Categories of Factors Considered in Analyses
Explaining Differences in State and County Capital Reversal Rates

    VI. Crime and Victimization

    1. Crime Rates (all crimes, homicides, violent crimes and FBI Index Crimes)
    2. Victims (racial composition in general and relative to population)
    3. Incarceration (rates and new admissions)

    VII. Political, Ideological and Religious Characteristics (state only)

    1. Relative Strength of Two Major Political Parties
    2. Religious Affiliation of Population
    3. Extent of Use of Criminal Sanctions

    VIII. Social Characteristics

    1. Poverty Rates
    2. Welfare Recipients and Expenditures
    3. Income Distribution
    4. Unemployment
    5. Divorce Rates


      b. Case-level factors.

Table 3 below lists the traits of federal habeas cases examined by Analysis 19 to see if they predict when reversible error is and is not found at that review stage. We could not collect every type of information for every case because most of our data come from published state and federal court decisions reviewing death verdicts, and because reviewing judges do not always know, and their decisions do not always discuss, every trait of the crime, defendant, victim and lawyers in the case, or those of the judges who decided the case.

Table 3: Categories of Factors Considered in Analysis 19
Explaining Outcomes of Capital Federal Habeas Cases

    I. Sentencing State and County

    1. Offense
    2. Trial

    II. Timing

    1. Offense
    2. Arrest
    3. Conviction
    4. Death Sentence
    5. Filing of Each Level of Appeal
    6. Decisions at Each Level of Review
    7. Execution

    III. Offense Characteristics

    1. Offense Charged and Convicted
    2. Location
    3. Circumstances
    4. Weapons Used
    5. Accompanying Offenses
    6. Accomplices and Disposition of Their Cases
    7. Evidence
    8. Defendant's Level of Participation
    9. Aggravating Circumstances
    10. Mitigating Circumstances

      IV. Defendant Characteristics

      1. Age
      2. Race
      3. Gender
      4. Mental Status
      5. Child and Sexual Abuse
      6. Drug and Alcohol Use
      7. Economic Status
      8. Employment History
      9. Criminal Record
      10. Connection to Community
      11. Other Aggravating Circumstances
      12. Other Mitigating Circumstances

    Table 3 (cont=d): Categories of Factors Considered in Analysis 19
    Explaining Outcomes of Capital Federal Habeas Cases

      V. Victim Characteristics

      1. Number
      2. Age
      3. Race
      4. Gender
      5. Traits Indicating Vulnerability
      6. Relationship to Defendant
      7. Economic and Social Status
      8. Connection to Community
      9. Method of Death, Wounds, Suffering

      VI. Defense and State Lawyers at Trial and on Appeal

      1. Public or Private (for defense lawyers)
      2. Appointed or Retained (for defense lawyers)
      3. Whether or Not Employed by Capital Case Resource Center (for defense lawyers)
      4. Out of State or In State (for defense lawyers)
      5. Employer
      6. Location
      7. Local or State Responsibility (state lawyer)
      8. Years of Experience
      9. Education
      10. Specialty

      VII. Judges

      1. Trial Judge
      2. State Direct Appeal Judges
      3. State Post-Conviction Judges
      4. Federal Habeas Judges
        1. District or Circuit
        2. Name and Party of Appointing President
        3. Education
        4. Service as Prosecutor or Other Government Official
        5. Years on Bench
        6. Size and Composition of Panels and Majority
        7. Caseloads

    Table 3 (cont=d): Categories of Factors Considered in Analysis 19
    Explaining Outcomes of Capital Federal Habeas Cases

      VIII. Procedural History

      1. Trial Procedures
        1. Length of Trial
        2. Guilt Determination by Trial or Plea
        3. Jury or Judge Sentencing

      2. State Appellate and Post-Conviction Procedures
        1. Votes and Outcomes
        2. Was Evidentiary Hearing Requested, Held

      3. Federal Procedures
        1. Was Evidentiary Hearing Requested, Held
        2. Lower Court Outcomes
        3. Appeals

      IX Legal Claims and Defenses

      1. Number and Types of Claims Raised by Defendant, and Court's Response
      2. Number and Types of Objections to Relief by State, and Court's Response


        c. Time.

    Time is a potentially important factor considered in all our analyses except four—Analyses 14-17, which explain differences in states' and counties' reversal experiences as a 23-year whole. Studying time may identify particular years in which important events affected reversal rates (e.g., major elections or court decisions) or may suggest that reversal rates have steadily improved or worsened over time. But, as we note above, the bearing of time may be difficult to determine.297 For one thing, time may be a stand-in for other factors that change over time. For example, court backlogs may increase over time. If backlogs affect reversal rates but are not separately studied, the passage of time will be given credit for that effect. This explains why the effect of time may diminish from a baseline analysis in which only the effect of state and time is considered to analyses in which other, more specific factors are considered.

    The effect of time may also be distorted in studies of ongoing processes, where outcomes of interest almost inevitably continue to occur after the study cut-off date. We already have identified two effects of the time-limited nature of our study, which together lead changes in reversal rates over time to be a better indicator of the influence of delay in reviewing death verdicts than of changes in the amount of capital error:

    • The condition most of our analyses explain is the proportion of death verdicts imposed in a given year that later court review found to be seriously flawed. Because the three-stage review process takes time—12 years, on average, as of the last year of our study—it is inevitable that fewer verdicts imposed in later study years will have been reversed by the study's cut-off date than is true of verdicts imposed in earlier study years, because many fewer verdicts will have completed review by then. This creates an artificial downward trend in the most recent years in reversals as a proportion of imposed verdicts, not because there are fewer errors in later death verdicts but because fewer were reviewed for error by the time the study ended.298

    • At the federal habeas stage, seriously flawed death verdicts take more time to be finally reviewed and reversed than do unflawed verdicts.299 Inevitably, therefore, a study cut-off date at a point when some verdicts imposed in the study period were not yet fully reviewed assures that the verdicts whose review outcomes we do count—ones finally reviewed before the cut-off date—include a disproportionately high number of the unflawed death verdicts, while verdicts whose review outcomes we don't count—because they were still under review as of the cut-off date—include a disproportionately high number of the flawed verdicts. Thus, a cut-off date automatically understates the actual rate of error by causing more flawed than unflawed verdicts to go uncounted. Because this problem is especially acute in years with large numbers of verdicts yet to be reviewed as of the cut off date, and because—as Table 4 below shows—the later the study year, the larger the number of verdicts yet to be reviewed as of the cut-off date, the bias against counting seriously flawed verdicts is especially a bias against counting seriously flawed verdicts imposed later in the study period.


    Table 4: Percent of Death Verdicts Still Under Review at Some Review Stage as of the Study Cut-off Date (Sources: DRCen, DADB, SPCDB, HCDB)
     
    Year of Verdict Tot. Number of Verdicts Percent Still Under Review
    1973-1978 694 14%
    1979-1981 704 23%
    1982-1985 1297 45%
    1986-1987 665 57%
    1988-1991 1087 66%
    1992-1993 537 87%
    1994-1995 426 99%


    For these reasons:

    • Where, after taking other factors into account, we find that the passage of time increases rates of reversible error (as is true for death verdicts reviewed at the state direct appeal stage of court review), we almost certainly have understated that effect.

    • Where, after accounting for other factors, we find that the passage of time is not significant (as is true in some of analyses of county reversal rates at all three stages), this may be because our use of a cut-off date has artificially depressed reversal rates for verdicts imposed in later years, neutralizing what otherwise would appear to be an increase in reversal rates over time.

    • Where, after accounting for other factors, we find that reversal rates decrease over time, and where we have measured reversal rates as a proportion of imposed death verdicts—and in all studies of the federal habeas stage of review, where the second effect discussed above occurs—the most we can say is that the actual effect of time is unknown. This is because a drop in reversal rates over time is at least partly the effect of the successively lower number of more recent verdicts that are finally reviewed and thus available to be reversed (or affirmed). Overall, as we note above, the inclusion of time trend in studies such as the ones described here serves to control for the effect on reversal rates of unfinished and delayed appeals, and is of little or no use as a measure of changing error rates over time.300

    E. Sources of Data

    Three data bases original to this study were assembled by the authors from information in court decisions available mainly through the Westlaw and Lexis legal search engines:

    • Direct Appeal Data Base (DADB). This data base contains information on state direct appeals of capital verdicts imposed in all years during the 1973-1995 period in which the relevant state had a valid post-Furman capital statute. The 4546 appeals in this data base include all those that we identified as having been finally decided during the 1973 to 1995 period. Data on each appeal include year of death verdict; year of decision; outcome of decision (whether the verdict was affirmed or reversed); subsequent judicial history on rehearing in the state system and certiorari to the U. S. Supreme Court; and citation. The methods used to identify capital cases that were decided on direct appeal and at other stages of review is described at pp. 14-16 above. This information is used in regression Analyses 1-4, 7-18 of factors associated with differences in state and county capital reversal rates.

    • State Post-Conviction Data Base (SPCDB, reproduced in major part in Appendix C to this Report). Appendix C and SPCDB contain a list of capital verdicts that were imposed during years when the relevant state had a valid post-Furman capital statute and were reversed on state post-conviction review between 1973 and April 2000. For each reversal, information was collected about year of decision; outcome of decision; basis for reversal; stage of trial affected by reversal; outcome on retrial; timing; and citation. A fuller description of the manner in which these data were collected is in Appendix C, pp. C-1 to C-2. In addition, SPCDB includes the county in which the death verdict was imposed and the year of the death verdict. This information is used in regression Analyses 1, 1R, 2, 5, 7-9, 11-18 of factors associated with differences in state and county capital reversal rates.

    • Habeas Corpus Data Base (HCDB). This data base contains information on the 598 final decisions of initial (non-successive) capital federal habeas corpus cases between 1973 and 1995 that reviewed death verdicts imposed during years in the 1973 to 1995 period in which the relevant state had a valid post-Furman capital statute. Included are hundreds of items of information (insofar as it was available in the court decisions reviewed for each case) in each of the nine categories and approximately 75 subcategories listed in Table 3, pp. 138-40 above. This information is used in regression Analyses 1, 1R, 2, 6-9, 11-18 of factors associated with differences in state and county capital reversal rates. It also is used in Analysis 19, a regression study of factors associated with the probability of affirmance or reversal of capital verdicts on federal habeas review.

    Our Death Row Census Data Base (DRCen) is a compilation of death verdicts imposed between 1973 and 1995 organized by state and by the name of the defendant who was sentenced to death. We assembled this data base starting with the information used to produce the NAACP Legal Defense Fund's quarterly death row census, Death Row U.S.A.. We expanded that information to include every death verdict we could identify that was imposed in any year during the study period in which the relevant state had a valid post-Furman capital statute, and to include information on sentence year, county of death verdict, county census designations, and other information. Data include name of all individuals on a state death row between 1973 and 1995; state and county in which death verdict was imposed; year death verdict was imposed. Death Row U.S.A. is also our source of information about executions—when and where they occurred and whether they were consensual or non-consensual. This information is used in all 19 of our regression analyses.

    State and county reversal rates—the condition studied in regression Analyses 1-18Care calculated based on:

    • the number of death verdicts that were imposed in each state and county during the study period (see DRCen), or that were reviewed at each of the relevant capital review stages (DADB, HCDB), or that were available for review at the state post-conviction stage because they had completed direct appeal review and been affirmed (DADB); and

    • the number of death verdicts from the relevant state or county that were reversed at each of the three review stages or at all three of them (DADB, SPCDB, HCDB).

    State and county characteristics—the factors operating at the state level that are examined by regression Analyses 1-6, 8-17, and the factors operating at the county level that are examined by regression Analyses 7-18, to determine whether they are associated with differences in state and county capital reversal rates—are from the following public sources:

    • State population and racial composition are from the United States Census (USCen).

    • Crime data are from the FBI Uniform Crime Reports (UCRDB).

    • State and county homicide and homicide victimization data, including by race, are from the Vital Statistics of the United States or other data sources generated by the Centers for Disease Control and Prevention National Center for Health Statistics (HomVic).

    • Annual state prison population data are from the Source Book of Criminal Justice Statistics (Pris—en).

    • General court caseload data are from State Court Statistics (Ct—aLd).

    • Court expenditure data are from Expenditure and Employment Data for the Criminal Justice System (CtExpen).

    • Welfare recipients and expenditures are from The Statistical Almanac of the United States (FacWelf).

    • Other county-level data are from a data set created by Professor Steven F. Messner and his colleagues at the University of Albany and the University of Illinois, as described at p. 170 & n.371 below. We are extremely grateful for the use of these data.

      Our indexes of the political pressure on state judges from judicial selection techniques are original to this study and based on provisions of the 34 study states' constitutions and codes governing judicial selection, supplemented by information from the National Center for State Courts (PolPres).301

      F. Format for Reporting Results

    The text and notes to this Report discuss the results of our 19 analyses. For each separate analysis in each study, we show the amount of unexplained variance left by the baseline analysis compared to that of each analysis of more specific factors; the fit of the baseline analysis compared to that of each analysis for specific factors; factors that were significantly associated with reversal rates; the factors' significance levels; and the factors' effect size (sometimes given numerically, other times displayed in graphs). Appendix E defines the explanatory factors included in one or more of our analyses, and Appendix F contains correlation matrices of all those factors as well as the minimum, mean and maximum values and standard deviations for each factor. A table with the detailed results of each analysis is collected in Appendix G.

next | previous | table of contents