Race and Imprisonments: Vigilante Violence, Minority Threat, and Racial Politics

Abstract

The effects of lynchings on criminal justice outcomes have seldom been examined. Recent findings also are inconsistent about the effects of race on imprisonments. This study uses a pooled time-series design to assess lynching and racial threat effects on state imprisonments from 1972 to 2000. After controlling for Republican strength, conservatism, and other factors, lynch rates explain the growth in admission rates. The findings also show that increases in black residents produce subsequent expansions in imprisonments that likely are attributable to white reactions to this purported menace. But after the percentage of blacks reaches a substantial threshold—and the potential black vote becomes large enough to begin to reduce these harsh punishments—reductions in prison admissions occur. These results also confirm a political version of racial threat theory by indicating that increased Republican political strength produces additional imprisonments.

NOTES

Notes

¹ Pooled time-series methods offer benefits. Causal assertions must rest on claims about within case changes (Lieberson 1985). But cross-sectional analyses force differences between cases to be used to proxy within case changes. This demands a heroic assumption. “If [we can] assume the values of a dependent variable were all initially the same, only then are there grounds to believe that situations observed cross-sectionally will at a later … time provide a reasonable[picture of]causation” (Lieberson 1985:180, his italics). Pooled time-series designs also typically provide more cases and therefore greater information, decreased collinearity, and increased statistical power than cross-sectional alternatives.

² In this article, we do not assess Blalock's economic version of threat theory because the necessary data are unavailable, but we provide tests of the Blalock's racial threat version of this theory.

³ Significant differences in racial voting began after the passage of the civil rights laws. White southerners responded to these laws by sharply reducing their votes for Democrats (Edsall and Edsall 1991; Lamis 1999), but blacks continued to vote for this party. In some deep southern states only a few whites now support Democrats, but these white votes along with almost monolithic black support often give the Democratic Party influence over policies important to southern blacks.

⁴ Aide John Erhlichman described Nixon's 1968 campaign. “We'll go after the racists. That subliminal appeal to the anti-black voter was always present in Nixon's statements and speeches” (Beckett 1997:42 quoting Erhlichman 1982:233). In 1988, Republicans ran an ad quoting a victim about Bush's opponent “[ex-governor] Dukakis not only opposed the death penalty, he allowed first-degree murderers to have weekend passes from prison” … [as the] clearly black [offender]—Willie Horton stared dully into the camera” (Carter 1996:76–77). In a later debate “29 Republican legislators spoke derisively about midnight basketball … characterizing the program as ‘hugs for thugs.’” (Hurwitz and Peffley 2005:99–100).

⁵ Mendelberg (2001) gives state examples. “In 1991 … little-known Republican Kirk Fordice upset incumbent Democrat Ray Mabus by … attacking [him for] coddling criminals” (p. 184). Then several “Republican governors … ran [successful] ads attacking their opponents for being lax on violent crime. These messages … implicitly referred to violent black criminals” (p. 6).

⁶ We prefer to estimate with fixed effects, but this estimator cannot assess lynch rate effects. The lynch rates differ between states, but they are based on state-specific constant 30-year means and remain constant. Fixed-effects models cannot estimate a variable's effects if it is time invariant.

⁷ Almost all residents in state prisons are sentenced to at least a year. We analyze admissions rather than prison populations because court decrees that strictly limit crowding make prisoner numbers highly dependent on prison space, but space is extremely difficult to measure. Prison space and hence incarcerated populations also are determined by spending on past construction that often occurred before the start of our sample. Specifying correct lags in models that predict inmate numbers is difficult since inmates serve different terms. These considerations make analyses based on admissions considerably more plausible.

⁸ We do not assess determinant sentencing policies as this, and the many other policies that influence imprisonments are part of the dependent variable. At issue are the social and political factors that produce the many laws and informal practices that combine to influence admission rates. But these policies are caused by the social and political influences included in our models. And, to focus on only one of these policy decisions and ignore others that are extremely difficult or impossible to measure would produce biased estimates.

⁹ Johnston (1984) states that exhaustive models yield more accurate estimates. “[I]t is more serious to omit relevant variables than to include irrelevant variables since in the former case the coefficients will be biased, the disturbance variance underestimated, and conventional inference procedures rendered invalid, while in the latter case the coefficients will be unbiased, the disturbance variance properly estimated, and the inference procedures properly estimated” (Johnston 1984:262).

¹⁰ The following variables are significant at the one-tailed .05 but not the two-tailed .05 level: black presence and Republican legislative dominance in model 1 and Republican legislative dominance in model 2. In model 3 Republican legislative dominance again is significant at this level, and median incomes, the violent crime rates, and Republican legislative dominance are significant only at the one-tailed level in model 4. The unreported state specific trends are jointly significant to well beyond the .0001 level and the same is true of the eight unreported coefficients on the regional dummy variables all four models. Note as well that neither the lynching rate nor the black presence effects are attributable to the inclusion of the Republican president dummy variable and/or the state-specific time trends.

¹¹ A multivariate analysis would not be as necessary if the lynching rates were uncorrelated with explanatory variables, but this is not the case. These rates are correlated at .387 with the dependent variable, while the largest correlations between lynching rates and other explanatory variables are −.418 with median income, .281 with ideology, and .275 with violent crime rates. But collinearity is not problematic. The strongest correlation between explanatory variables is .618. A variance inflation factor (VIF) test appropriately conducted without the squared terms gives a maximum statistic of 6.17, or a score well below the threshold of 10 recommended by conservative statisticians as an indication of this problem.