Ongoing project using DPI death row data

Testing the hypotheses, three sets of regressions with control variables were performed. The first set, shown in Table 5, uses the analytical sample in its entirety in order to focus on the predictive power of gender and race on active sentences and on resentences. The second and third sets, as displayed in Tables 6 and 7, separate the sample by gender and run the identical variables as identified in Table 4 except for race in which the dichotomous variable was used instead of the five-category race variable. The latter variable is preferable in representing the defendants and it was also a better fit. Table 5 presents the results of two binary logistic regression models predicting the likelihood of still having an active death sentence (at the beginning of 2021) and having receiving a reduced sentence to life or less by seven variables identified through a series of tests for multicollinearity and fit.

Both models in Table 5 suggest statistically reliable results.  Model 1 correctly classified 81.7% of the cases and a test of the full model against the constant-only model showed significant improvement (χ2 (8) = 227.289, p = .000). Model 2 correctly classified 74.3% of the cases and the chi square value of (χ2 (8) = 166.831(8), p = .000). Though the ensuing test values should be interpreted with some caution, the variables in Model 1 explain between 39.1% (Cox % Snell R Square) and 53.2% (Nagelkerke R square) of the variance in active death sentence. Those same variables in Model 2 explain between 30.0% (Cox & Snell R Square) and 40.7% (Nagelkerke R square) of the variance in resentencing to life or less. The Hosmer and Lemeshow Test goodness of fit statistics for both Model 1 and 2 additionally suggest that the models are a good fit (respectively; χ2 (8) = 8.304, p=.404, and χ2 (8) = 8.884, p =.352).

Gender was not a significant predictor in Model 1 but it was in Model 2 suggesting that women are more likely to have their sentences reduced to life or less than men are (OR = 1.845, CI 95% [1.138, 2.993]). Race was insignificant in both models as were percentage white population in a county and the percent in poverty. However, for Model 1, the significance levels (albeit weak), and odds ratios, combined with the results of the univariate logistic regressions (Appendix) suggest that the separate men and women models may yield significant results for these control variables. In Model 1, non-southern states increased the likelihood of an active sentence (OR = 2.470, CI 95% [1.408, 4.332]) but not in Model 2 in which Southern states seemed to increase the likelihood of a resentence. Perhaps the results of particular interest are the effects of women legislators in both models. Model 2, though producing significant probability values, has odds ratios below one. Model 1 does suggest, however, that the percentage of Democratic women and black women legislators increase the probability of a defendant still having an active death sentence (OR = 1.233, 95% CI [1.158, 1.313]. As might be expected a state’s population was significantly and positively related to active death sentences (OR = 1.037, 95% [1.014, 1.060]) but not in Model 2 (OR = 0.946, 95% CI [0.928, 0.9650]).

Table 6 shows the results of two models, disaggregated by gender, that predict whether or not a defendant had an active death sentence as of January 2021. Model 3 explained between 40.8% (Cox & Snell R Square) and 55.1% (Nagelkerke R square) and Model 4 explained 31.6% and 50.3% of the variance in whether or not defendants still had active death sentences. Models 3 and 4 correctly predicted 83.2% and 80.1% of the cases, respectively. A test of the full models against the constant-only models were statistically reliable for Model 3 (χ2 = 162.276(7), p = .000) and Model 4 (χ2 = 67.733(7), p = .000).

Now, for male defendants (Model 3), percent white in the county is weakly significant (OR = 1.031, CI 95% [1.000, 1.062]) and percent of the county living in poverty is significant (OR = 1.120, CI 95% [1.002, 1.252]). Again, non-southern states were positively related to active death sentences but only for men (OR = 2.895, 95%, [1.466, 5.174]). As seen in Model 1, both men (Model 3) and women (Model 4) percent of women legislators who are Democrats, percent of women legislators who are black, and population are positively related. Race was not significant for both men and women. 

Lastly, Table 7 displays the results of Model 5 (of men) and Model 6 (of women) predicting the likelihood someone is resentenced to life or less. The models were comparatively not as robust. Model 5 correctly predicted only 73.9% of the cases explaining between 29.5% (Cox & Snell R Square) and 39.6% (Nagelkerke R square). Model 6 correctly predicted 75.5% of the case and explained 31.6% and 42.3% of the variance in whether or not women defendants still had active death sentences. Still, tests of the full models against the constant-only models were statistically reliable for Model 5 (χ2 = 108.276(7), p=.000) and Model 6 (χ2 = 57.243(7), p = .000). The Hosmer and Lemeshow tests also showed that the models were a good fit (Model 5: χ2 = 8.019(8), p = .432; Model 6: χ2 = 5.572(8), p=.695).

         The notable change from Model 2 (Table 5) when differentiating between men and women was the positive relationship between state population and having one’s death sentence reduced to life or less in Model 6 (OR = 1.305, 95% CI [0.518, 3,286]). That is, for men, the negative relationship between state population and reduced sentence is similar to Model 2 but, for women, increases in population are related to increased probabilities for resentencing. Again, the relationships between women legislators and resentences are significant but in unexpected directions though, now, for men (Model 5), the percentage black women legislators are not significantly related.

The racial distribution of the full dataset (page 2 of Table 1 in Part 1) and the analytical sample are similar except that the percentage of men in two groups did change. In the full dataset, 47% (n = 4022) of the men are white whereas in the analytical dataset the percentage of white men drops to 41.9% (n = 166). Women who are white comprise 64.4% of women sentenced to death. Nine percent (n = 771) of men were Latinx in the full dataset versus 12.1% (n = 48) in the analytical sample (Latinx women comprise 6.2% of women sentenced to death). Given that the most death sentences have come from California and Texas, which sentenced 247 and 237, respectively, (729, 66.3%, of all Latinx men), is not all too surprising in states where approximately 40% of the population identify as Hispanic or Latino (U.S. Census, 2024). Put differently, when considering the total number of men sentenced in these two states black men are overrepresented in both California (34.7% 365 out of 1051 men) and Texas (37.9%, 416 out of 1095 men) where the black populations are about 6.5% and 13.4% (U.S. Census, 2024).

The more pronounced gender and race differences are visible when the categories are examined. Beginning with the largest group, defendants who were resentenced to life or less, the percentage of men drops somewhat from the full dataset to the sample dataset in Table 1 (39.7%, n = 3395 vs. 32.8%, n = 86). By far this is the largest percentage for women with 48.6% (n = 86). Taking a closer look at race, the percentage of white men resentenced to life drops (45.2%, n = 1534 vs. 38%, n = 63). This compared to the 69.8% (n = 60) of women that are white. Black women make up 20.9% (n = 18) of women resentenced compared to black men who are 34.5% (n = 60) of men who were resentenced. Black men’s percentages also dropped from the full dataset (47.5%, n = 1612) whereas Latinx men experienced an increase (5.6%, n = 191 vs. 12.5%, n = 6).

Percentages among men increased for active death sentences (25.3% in the full dataset vs. 31.3% in the sample) and sentences commuted (4.1% vs. 7.8%). The percentage of women with active death sentences are 28.2% (n = 50) and those with commutations is 6.8% (n = 12). The percentage of white men with active death sentences dropped from 42.1% (n = 911) to 24.7% (n = 41) whereas 60% of women are white. The percentage of black men also dropped from 40.6% (n = 879) to 28.2% (n = 49) which is closer to the percentage of black women with active death sentences (22%, n = 11). The percentage of Latinx men experienced the most dramatic increase (14.3% to 60.4%) especially compared to the only 12% (n = 6) of Latinx women with active death sentences.

        This preliminary overview suggests that in regions where women (and men) are sentenced to death may be different than areas in which only men are sentenced to death. The dependent variables examined will be dichotomizations of three categories: active death sentences, resentences to life or less, and executions. The first two are chosen due in part to their size without having to combine too many categories, and in part because the lack of executions implies that a comparison between active sentences and those who were resentenced is most appropriate. Executions are included in the sample because the same logic applies as in areas that are willing to execute women, are likely different than areas that do not. Virginia’s only two contributions to the sample are two executions, both originally sentenced in Pittsylvania County, Teresa Wilson Bean Lewis sentenced in and Ronald Lee Fitzgerald. The county with the largest number of executions for both men and women was Harris County, the most populated county in Texas, with 44.4% (n = 24) of executed men and 18.8% (n = 3) of the executed women in the sample. Appendix C lists the states by region, as well as the number of death sentences by county. Only those, up until January 2021, that have active death sentences, were resentenced, or executed are counted in the final analytic sample of 461.

To capture the historical and socio-political characteristics of when and where the defendants were sentenced to death, men and women were “matched” by the county and year of sentencing. This was accomplished by first identifying the counties and years of the women defendants sentenced to death. Then men that were sentenced in the same year and county were located and included in the sample dataset. Federal cases were excluded altogether. The analytical sample (N= 461) contains 151 women and 310 men who, as of early 2021, had active death sentences, were resentenced to life or less, or were executed. Ninety-eight women were sentenced in the same year and county with at least one male defendant. For thirty women, there was only a year’s difference in sentencing between them and their male counterparts. Twenty-one women’s sentences differed from men’s by three to six years. There were, three different counties in which no men were executed in more than twenty years. For these cases, neighboring counties were found for men sentenced in the same or a couple years apart. Fortunately, men already in the dataset, due to their matches with other women defendants, were already included. For example, Shirley R. Tyler, a woman sentenced to death in 1979 in Pike County, Georgia, was sentenced in a county in which a man, Leeland Braley, was sentenced to death twenty years later. Given that the dataset already contained two men sentenced to death in a nearby county in 1975, Braley’s exclusion and Tyler’s inclusion is justifiable.

Current case outcomes—the dependent variables.

The Death Penalty Census (DPIC, 2021) collects the “current case status,” a twenty-category variable, defined as the “final outcome of the court proceedings related to those charges” (p. 7). The full definitions for each of the twenty types of case statuses are available in the Death Penalty Census codebook (DPIC, 2021). For purposes of analyses here, the twenty categories were collapsed into seven of the existing categories: “active death sentences,” those who “died on death row,” were “executed,” “exonerated,” received a “grant of relief,” were “resentenced to life or less,” or had their “sentence commuted.” Three categories, along with all the cases contained within, were omitted altogether because no woman was “found not guilty by reason of insanity,” had their “retrial barred,” or were “no longer on death row” for unknown reasons.

The DPI distinguishes between whether or not defendants were executed in the same state, a different state, or for a different crime. For purposes of this study, all three were consolidated under one category, “executed.” The categories “died pending retrial or resentencing,” “grant of relief (retrial/resentencing pending),” and “grant of relief (relief not final)” were also combined given that the definition of “died pending retrial or resentencing is; “after receiving a final grant of guilt or penalty relief, the defendant died prior to a retrial or resentencing” (DPI, 2021, p. 8). Nila J. Wacaser, for example, had “died pending retrial or resentencing” upon receiving a “final grant of guilt or penalty relief.” Therefore she, and the twenty-seven men who also died after receiving a grant of penalty relief were categorized as having received a “grant of relief.” While the categories are comparable for statistical convenience, the various circumstances are certainly qualitatively distinctive.1

“Sentence commuted” is an amalgamation of those who were commuted through executive action or through “administrative” action (referring to commutations resulting from a court decision). As with the category “resentenced to life or less,” commuted sentences can range from a reduced sentence to life without parole. Lastly, there are the categories “active death sentence” (referring to defendants still on death row as of January 1, 2021) and the “exonerated” (overturned convictions that later resulted in all charges being dropped or receiving a governor’s pardon). To visualize the distribution of case outcomes, Table 1 is organized around the seven case outcomes and by a defendant’s race and sex. 

The racial distribution of the full dataset  and the analytical sample are similar except for percentages of men in two racial groups. In the full dataset, 47% (n = 4022) of the men are white whereas in the analytical dataset the percentage of white men drops to 41.9% (n = 166). Women who are white comprise 64.4% of women sentenced to death. Nine percent (n = 771) of men were Latinx in the full dataset versus 12.1% (n = 48) in the analytical sample (Latinx women comprise 6.2% of women sentenced to death). Given that the most death sentences have come from California and Texas, which sentenced 247 and 237, respectively, (729, 66.3%, of all Latinx men), is not all too surprising in states where approximately 40% of the population identify as Hispanic or Latino (U.S. Census, 2024). When considering the total number of men sentenced in these two states, black men are overrepresented in both California (34.7% 365 out of 1051 men) and Texas (37.9%, 416 out of 1095 men) where the black populations are about 6.5% and 13.4% (U.S. Census, 2024).

The more pronounced gender and race differences are visible when the categories are examined. Beginning with the largest group, defendants who were resentenced to life or less, the percentage of men drops somewhat from the full dataset to the sample dataset in Table 1 (39.7%, n = 3395 vs. 32.8%, n = 86). By far this is the largest percentage for women with 48.6% (n = 86). Taking a closer look at race, the percentage of white men resentenced to life drops (45.2%, n = 1534 vs. 38%, n = 63). This compared to the 69.8% (n = 60) of women that are white. Black women make up 20.9% (n = 18) of women resentenced compared to black men who are 34.5% (n = 60) of men who were resentenced. Black men’s percentages also dropped from the full dataset (47.5%, n = 1612) whereas Latinx men experienced an increase (5.6%, n = 191 vs. 12.5%, n = 6).

Percentages among men increased for active death sentences (25.3% in the full dataset vs. 31.3% in the sample) and sentences commuted (4.1% vs. 7.8%). The percentage of women with active death sentences are 28.2% (n = 50) and those with commutations is 6.8% (n = 12). The percentage of white men with active death sentences dropped from 42.1% (n = 911) to 24.7% (n = 41) whereas 60% of women are white. The percentage of black men also dropped from 40.6% (n = 879) to 28.2% (n = 49) which is closer to the percentage of black women with active death sentences (22%, n = 11). The percentage of Latinx men experienced the most dramatic increase (14.3% to 60.4%) especially compared to the only 12% (n = 6) of Latinx women with active death sentences.

Death Penalty Information Center. (2021). Death penalty census codebook. Death Penalty Information Center. https://dpic-cdn.org/production/documents/Death-Penalty-Census-Codebook.pdf