The United States is one of the most punitive countries in the world, with minorities—and Black people, in particular—bearing the heaviest burden of state punishment. Consistently undergirding much of the late 20th century punitive outcomes have been high levels of public support for a variety of punitive policies (Gottschalk Reference Gottschalk2006).Footnote 1 But the United States has not always been such an outlier. Instead, the literature has consistently argued that the mid-century United States witnessed a dramatic punitive turn in both attitudes and punishment outcomes (Beckett and Francis Reference Beckett and Francis2020).
Race and context—in the case of the latter, crime and its perception—are theorized to be dominant forces in this turn in support for state punishment (Beckett and Francis Reference Beckett and Francis2020). While most studies acknowledge that there was a generalized increase in crime that took place in the mid-century (e.g., Klebba (Reference Klebba1975)), accounts differ as to how much this contextual change mattered relative to, and in combination with, significant political and social mediating forces—for instance, the media’s portrayal of crime (Enns Reference Enns2016); the political gains of Black people igniting racial threat (Duxbury Reference Duxbury2023; Eubank and Fresh Reference Eubank and Fresh2022); partisan competition spearheading new national policy discussions of law and order (Lopez Reference Lopez2015; Weaver Reference Weaver2007); and dissatisfaction with the ability and normative consequence of a robust welfare state to solve crime through redistribution (Clegg and Usmani Reference Clegg and Usmani2019; Flamm Reference Flamm2005; Gillens Reference Gillens1999). Despite this extant theory, systematic empirical evidence on punitive attitudes and their racial geography through the purported turn has, to date, been missing from the literature.
In this article, I empirically consider the nature of this supposed turn in punitive public opinion using Gallup and General Social Survey (GSS) data on support for the death penalty spanning 1953–1985. The temporal scope of this polling data is key. A significant challenge in characterizing the punitive turn is the necessity of data that not only captures the period of growing punitiveness, but prior trends as well. While there are numerous frequently used surveys post-1975 that ask a range of punitive questions, Gallup data asking about the death penalty offers the only pre-1975 source of nationally representative individual-level survey data on punitive attitudes. While later surveys can establish corroborative trends, or cross-sectional relationships, they do not speak directly to the turn in attitudes.
My empirical focus on death penalty support is motivated by two considerations. First, the death penalty is the state’s most violent, coercive, and irreversible method of punishment. As such, it is of substantial interest in and of itself, as illustrated by the significant literature dedicated to its study (see, e.g., Steiker and Steiker (Reference Steiker and Steiker2020)). Second, it is the only form of state punishment for which consistent questions have been asked in national surveys going back as far as the 1950s. As such, measures of support for the death penalty—while only one facet of what is often conceived of as a multifaceted construct of punitiveness (Adriaenssen and Aertsen Reference Adriaenssen and Aertsen2015); one that is subject to its own unique history in the public and political consciousness of the United States (Banner Reference Banner2002)—are uniquely able to provide insight into the turn in attitudes that occurred between the 1950s and 1970s.
With this data, I establish three sets of empirical facts about changing attitudes during this period. First, I establish that White respondents became differentially supportive of the death penalty in the mid-1960s by approximately 12 percentage points. Although White respondents were already more supportive of the death penalty in the preturn period, this post-turn increase represents more than a 50% increase in their average support.
Second, I identify additional geographic heterogeneity in death penalty support. While the South was initially less supportive of the death penalty (relative to the combined non-South) in the preturn period, the region experienced the largest increase in death penalty support after the turn—by approximately 7 percentage points—nearly entirely closing the pre-1967 regional difference. Under an alternative set of statistical assumptions, randomization inference indicates that it is extremely unlikely that this post-1967 Southern divergence arose by chance. Furthermore, White death penalty support in the South grew more after the punitive turn relative to Black respondents in that region. This difference was driven by changes in White attitudes, as I find no evidence that Black respondents’ support for the death penalty changed after the 1960s in regionally defined ways. The inclusion of a time-varying control for region-specific crime, and evidence that Black people were far more likely to experience violent crime than White people further suggests that this divergence in these attitudes was not a simple reaction to the phenomenon of rising crime.
Third and finally, to better understand the racial and regional patterns, I examine how these patterns relate to partisan identification and regional violent crime trends. I find that among White respondents in the South, there is suggestive evidence that partisan identification became increasingly connected to death penalty support after the punitive turn and after racial-partisan realignment. Given the dominant role of racial resentment in partisan realignment (Kuziemko and Washington Reference Kuziemko and Washington2018), I interpret these findings as suggestive—though unfortunately not conclusive—of the critical mediating role of racial resentment in changing attitudes toward the most coercive punishment of the state during this period. Finally, by contrast, White respondents were not differentially more likely to respond in their death penalty attitudes to regional violent crime relative to Black respondents. Nor were respondents in the South of either race more sensitive to regional violent crime than respondents in other regions.
These findings contribute to the literature that has considered punitive attitudes by offering important new descriptive characterizations of historical patterns in racial attitudes often considered only via cross-sectional correlations and experimental work from contemporary periods (Bobo and Johnson Reference Bobo and Johnson2004; Unnever and Cullen Reference Unnever and Cullen2010). Extant findings are important, but they only indirectly address questions about the historical evolution of individual punitive attitudes. Where the literature on race and punitiveness has considered trends in punitive attitudes over time (Beckett Reference Beckett1997; Duxbury Reference Duxbury2020b; Enns Reference Enns2016; Jacobs and Carmichael Reference Jacobs and Carmichael2002; Ramirez Reference Ramirez2013b; Stinchcombe et al. Reference Stinchcombe, Adams, Heimer, Lane, Smith and Taylor1980), my findings contribute by considering trends from before the supposed punitive turn, and by examining unevaluated covariation in racial and geographic opinion. For example, in the existing literature, the (separate) racial and regional subgroup analysis in Ramirez (Reference Ramirez2013b) began only in 1972, and the question asked by Ramirez is distinct. Where Ramirez asks what time-varying factors contribute to different punitive sentiment by race, I ask whether punitiveness among respondents of different races underwent a particular transformation in the mid-century (requiring earlier data) and whether that break had distinct geographic heterogeneity. Authors that use only the GSS (e.g., Stinchcombe et al. (Reference Stinchcombe, Adams, Heimer, Lane, Smith and Taylor1980), Anderson, Lytle and Schwadel (Reference Anderson, Lytle and Schwadel2017)) necessarily begin their analysis in 1972, as does Duxbury (Reference Duxbury2020b)’s analysis. Beckett (Reference Beckett1997)’s analysis began somewhat earlier, 1965, but pairs this single data point with data from only one later year, 1988, which limits our ability to understand trends prior to and through the intervening period. Baumgartner, Boef and Boydstun (Reference Baumgartner, De Boef and Boydstun2008) examined responses to questions about punitive attitudes as early as 1952, but the authors’ research questions are distinct and do not focus on racial or geographic variation in the rise of punitiveness.
In summary, existing work does not estimate when death penalty support among different racial groups changed from the 1950s onward, nor where those changes were most pronounced, after accounting for other individual-level factors that might account for attitudes. Finally, while aggregate trends can tell us about broad correlations between attitudes on different related topics—e.g., racial attitudes, or welfare attitudes—evaluating multipart heterogeneity requires a different estimation strategy like the one employed in this paper.
These findings contribute to a continued and still unsettled debate about whether and why crime and race mattered in the punitive turn and the subsequent late 20th-century phenomenon of mass incarceration (for recent entries in that debate see Murakawa (Reference Murakawa2014); Enns (Reference Enns2016); Hinton (Reference Hinton2016); Clegg and Usmani (Reference Clegg and Usmani2018); Clegg and Usmani (Reference Clegg and Usmani2019); Norton and Stein (Reference Norton and Stein2020); Beckett and Francis (Reference Beckett and Francis2020); Clegg and Usmani (Reference Clegg and Usmani2021)). This literature largely asks about the relationship of crime and race to end-line carceral outcomes. This paper, by contrast, does not ask about incarceration itself, but rather addresses one potential link in the mass incarceration causal story (see, e.g., Enns (Reference Enns2016) and Duxbury (Reference Duxbury2020a)): the punitive attitudes, as measured by death penalty support, of a democratic public. My findings accord with past scholarship that discounts the sustained importance of crime in producing mass incarceration (e.g., Beckett (Reference Beckett1997)), indicating instead that insofar as punitive public opinion and incarceration are linked, the explanatory story of incarceration should account for the mediating forces of race and geography (see e.g., Duxbury (Reference Duxbury2020a)).
These findings are not, it must be noted, a causally identified test of a single causal factor explaining why the punitive turn occurred, nor an empirical accounting of punitive attitudes comprehensively construed. It remains difficult to source variation that meets inferential requirements to make such a causal claim; while systematic public opinion data that captures attitudes to other related aspects of the theoretical construct of punitiveness are not available. However, these results nevertheless comprise new descriptive evidence of the important role played by race and geography in the so-called punitive turn. The question answered by this paper is not how death penalty attitudes, a core component of punitive attitudes broadly, grew from the late mid-century—something well-documented in the literature above—but how that trajectory changed in the mid-century in racially and geographically specific ways heretofore undocumented in the literature.
The Making of Punitive Public Opinion
Punitive attitudes (or “sentiment” or “mood”), broadly construed, comprise support for a variety of state policies that punish (Ramirez Reference Ramirez2013b), though precise theoretical conceptualizations and related empirical measures remain debated (see, e.g., Adriaenssen and Aertsen (Reference Adriaenssen and Aertsen2015)).
Across different conceptualizations and measures, scholars consistently identify race as a key factor in individual punitive attitude formation, with White people differentially supporting punitive policies (more so when they exhibit racial animus), and holding beliefs consistent with Black people being those to whom punitive policies will be applied (Bobo and Johnson Reference Bobo and Johnson2004; Hurwitz and Peffley Reference Hurwitz and Peffley1997; Pollak and Kubrin Reference Pollak and Kubrin2007; Ramirez Reference Ramirez2021; Unnever and Cullen Reference Unnever and Cullen2012, Reference Unnever and Cullen2010). This difference especially pronounced when considering the death penalty as a specific case of state punishment (Unnever and Cullen Reference Unnever and Cullen2007; Unnever, Cullen and Jonson Reference Unnever, Cullen, Jonson and Tonry2008).Footnote 2 By contrast, scholars argue that Black attitudes are formed distinct from those of White people as a consequence of anxiety about crime, perceptions of criminal justice system bias, and fear that state repression will be directed toward them (Johnson Reference Johnson2006). As a consequence, these studies find that White people are consistently more punitive than Black people.
Of course, neither the construction nor salience of race are temporally constant. Thus, how people conceive of their own race, and the race to whom punishment will apply are not time-invariant. While the perceived association between race and criminality runs deep in American history (see, e.g., Muhammad (Reference Muhammad2011)), the mid-20th century also drew new linkages between them, arguably producing a turn in punitive attitudes (Beckett and Francis Reference Beckett and Francis2020). These linkages were built on a number of foundations. First, scholars argue that they were built on generalized negative White reaction to the social and political gains secured by Black people as a result of the Civil Rights movement (Beckett and Francis Reference Beckett and Francis2020; Duxbury Reference Duxbury2023; Eubank and Fresh Reference Eubank and Fresh2022). That movement made important incremental gains that ultimately culminated in the 1964 Civil Rights Act and 1965 Voting Rights Act that formally institutionalized—and de facto produced—desegregation and voting rights for Black people (Fresh Reference Fresh2018).
Second, the status threat induced by those gains in the general public provided a rich breeding ground for new forms of racial appeals, something that political entrepreneurs capitalized on in an attempt to secure partisan gain (Blalock Reference Blalock1967; Duxbury Reference Duxbury2020b; Weaver Reference Weaver2007). To do so, political entrepreneurs, namely on the ideological Right, attacked the social welfare policies of New Deal Liberalism and President Johnson’s Great Society program, casting Black welfare recipients as undeserving of the state’s help (Flamm Reference Flamm2005; Gillens Reference Gillens1999). Simultaneously, law and order emerged as a new and powerful national policy dimension; situating itself as an alternative and highly punitive means of social control for a now ascendant minority (Alexander Reference Alexander2012; Weaver Reference Weaver2007). By criticizing welfare support directed at minorities, and championing of law and order applied to minorities, policy discussions of punishment became increasingly layered with racially coded language (Lopez Reference Lopez2015; Mendelberg Reference Mendelberg2001). These appeals proved quite successful. Disaffected White voters, particularly in the South, exited the Democratic party, completing a long process of racial-partisan realignment (Kuziemko and Washington Reference Kuziemko and Washington2018; Schickler Reference Schickler2016).
The civil disobedience of the Civil Rights movement provided ample fodder to those who sought to portray Black people as disorderly law breakers, deserving of punishment rather than state support. This further coupled attitudes about race and the welfare state in the context of crime (Flamm Reference Flamm2005; Kellstedt Reference Kellstedt2003).Footnote 3 The media in turn developed and enhanced these linkages between race, crime, welfare, and punishment (Enns Reference Enns2016; Kellstedt Reference Kellstedt2003).Footnote 4 In this way, punishment was not a necessary response to rising crime, but reflected an increasing view that a liberal social welfare solution had decidedly failed (Beckett Reference Beckett1997; Flamm Reference Flamm2005).
Although these race-laden discussions of punishment were directed primarily at White people, particularly in the South (Duxbury Reference Duxbury2020b; Lopez Reference Lopez2015; Mendelberg Reference Mendelberg2001; Weaver Reference Weaver2007), Black people were not immune from political entrepreneurs heralding law and order. Indeed, controlling crime in Black neighborhoods, where it was highest, was argued by Black contemporaries as essential in order to protect the gains of the Civil Rights movement (Forman Reference Forman2017). But Black people may have responded less punitively to law and order rhetoric as a function of legacies of race-based state repression (Johnson Reference Johnson2006). Moreover, punitive policy is not the only understood remedy to rising crime. Indeed, Black communities frequently asked for crime control coupled with social programs to address the root causes of crime, the very programs under attack from the 1960s on (Hinton, Kohler-Hausmann and Weaver Reference Hinton, Kohler-Hausmann and Weaver2016).
Independent from race, scholars going back to at least Wilson (Reference Wilson1975) consistently theorize context, and specifically, crime, as a key factor in punitive attitude formation. Insofar as punishment is viewed as a deterrent, increasing crime may generate a rational demand for policies that punish (e.g., Beckett (Reference Beckett1997)’s democracy-at-work model).Footnote 5 Even if state punishment is not viewed as a deterrent, crime is associated with affective responses—namely, fear and anger—that are also theorized to find their outlet in greater punitiveness, independent of race (Frieberg Reference Frieberg2001; Hale Reference Hale1996). This is because punitiveness is a way for society to express its disapproval of certain actions (Garland Reference Garland2002).
And indeed, crime grew significantly during the mid-century—potentially interacting with race to produce punitive attitudes, as described above—but also potentially exerting its own independent effect. To illustrate the backdrop of crime during the mid-century, Figure 1 presents trends in homicide victimization rates by race and region. Although most studies utilize FBI Uniform Crime Report data (UCR) to measure crime (Appendix H), I also collect homicide victimization rates because they are disaggregated both by race and geographic region, and available well prior to the purported punitive turn.Footnote 6 The patterns clearly indicate that non-White people had the highest homicide rate and the largest trend growth from 1960 onward. From the mid-1970s, homicide rates declined, markedly among non-Whites.
It is worth noting, however, that as significant as the growth in crime was in between, approximately, 1960 and 1975, scholarship is also clear that the importance of an individual’s actual experience of crime is often less important for attitude formation than the individual’s perception of it—experienced through mechanisms of media and political framing (Beckett Reference Beckett1997; Duxbury Reference Duxbury2020a; Enns Reference Enns2016; Hale Reference Hale1996; Kleck and Jackson Reference Kleck and Jackson2017). Indeed, survey respondents consistently overestimate their personal victimization risk, and nearly always think crime is increasing—perceptions that link to attitude-formation beyond actual risk and objective trends (Quillian and Pager Reference Quillian and Pager2010; Ramirez Reference Ramirez2013a). Expressions of death penalty support, for instance, tend to be emotive rather than rational, such that increased factual information about the administration or effects of the death penalty fail to (re-)shape opinion (Ellsworth and Gross Reference Ellsworth and Gross1994).
Ultimately, the relationship between punitive attitudes and crime may have been temporally heterogenous. Crime may have provided the initial spark that made it easier for politicians and the media to link race and punishment than it (counterfactually) otherwise would have been. Once the fire of punitive attitude growth was roaring, however, crime may have been unnecessary to sustain it.
The Empirical Expectations
The extant literature indicates that the 1960s and 1970s witnessed a changing context of crime and a dramatic reorganization of race relations and racial associations with crime and punishment in the United States: i) crime increased, ii) Black people made significant civil and voting rights gains, and iii) the parties differentiated their policy platforms with respect to two issues linked to crime and race—law and order, and welfare. Together, these factors have led scholars to identify a turn in punitive attitudes occurring in the 1960s (Beckett Reference Beckett2020; Clegg and Usmani Reference Clegg and Usmani2018; Ramirez Reference Ramirez2013a).
The existing literature also warrants a close examination of the trends in punitive attitudes held by different racial groups in different regions during the mid-century. All else equal, in the case of a pure explanatory story of rising crime, we would expect Black people to increase their punitive attitudes more than White people because of their growing victimization rates (H1). The mechanism underlying this relationship takes punitive attitudes as a simple rational function of the true risk and experience of crime. Importantly, this prediction assumes that a marginal increase in crime has the same effect on punitiveness for any baseline level of crime (i.e., constant elasticity). If it is the case that a change in crime increases punitiveness more at low levels of crime, then punitiveness might change more among White people than Black people because White people’s baseline experience and victimization risk were lower. The logic could be because of (H1a) ceiling effects for Black people—i.e., high crime levels had already produced high punitiveness—or because (H1b) the equivalent change at low baseline is more meaningful given that it is a larger share of one’s experience of crime (i.e., people respond to growth rates).
Additional factors may have mediated the relationship of crime, however. Despite lower experiences of crime, White people may have received the dog-whistle messages targeted at them linking status threat and undeservingness of welfare to law and order, thus augmenting any genuine fear for their safety, and thus increasing their punitiveness.Footnote 7 In this case we would expect the punitive attitudes of White people to increase by more than those of Black people (H2). However, a larger punitive increase for White people relative to Black people might also reflect that, among Black people, genuine fear for safety is tempered by concern about the repressive nature of law enforcement (drawing on a history of problematic relations between the state and the Black community). Therefore, a larger mid-century increase in punitiveness among White people relative to Black people could reflect one or a combination of two linked processes: (H2a) a differential augmentation of White punitiveness, or (H2b) a differential suppression of Black punitiveness.Footnote 8 Though I note the existence of these two mechanistic channels, the subsequent empirical analyses cannot separate them.
Regional variation offers one means of better understanding observed changes in punitiveness by race. There are a number of reasons to expect that a change in punitiveness—and specifically, punitiveness by race—may display important regional variation. If differential changes in punitive attitudes by racial groups are related to racial advancements after landmark civil and voting rights legislation, we should expect those changes to be larger in those places where the advancements were more threatening in both status and instrumental terms. This is arguably the South, where the intensity of segregationist and disenfranchising policies—and by virtue, the intensity of integration and enfranchisement that resulted after 1965—were strongest (Key Reference Key1950).Footnote 9 Relatedly, the receptiveness of White populations to law and order dog whistles and to racially laden welfare criticism may have also been centered in the South where, again, racial hierarchy was more central to political, social and economic life (Kuziemko and Washington Reference Kuziemko and Washington2018; Weaver Reference Weaver2007). Differential regional legacies of racial hierarchy may have also depressed the punitiveness of Black people in the South for opposite reasons. Thus, the expectation is for regional location in the South to positively moderate the role of race in the punitive turn (H3).
There are, of course, other features of the South that are unique beyond the social centrality of racial hierarchy. But to account for punitive change differentially experienced in the South, those features must either theoretically link to activation as a consequence of changing crime, partisan politics, or civil rights; or themselves undergo change during the punitive turn.Footnote 10
Two alternative geographic predictions exist prominently in the literature. The first is that the Second Great Migration produced a Northern urban under-class without employment prospects, who were plagued by crime,Footnote 11 and in turn had significant demands on the welfare state that were not being met (Clegg and Usmani Reference Clegg and Usmani2019). Denied other political remedies, Northerners increased their punitiveness. In this case, the expectation would be that White people in the North and Mid-West—not the South—would be more likely to increase their punitiveness (H4). Second, scholars have argued that the South is part of the broader context of the Sunbelt, which combines the South and Southwest (Campbell and Schoenfeld Reference Campbell and Schoenfeld2013; Lynch Reference Lynch2010).Footnote 12 The increasing role of the Sunbelt in national politics writ large, was mirrored in the elevation of those states’ carceral policies to the national stage, fueling the spread of mass incarceration. Implicit in this argumentation is that those policies were backed by significant public punitiveness in those states, which arose from a long tradition of fiscal conservatism coupled with racial animosity. This argument predicts that the Sunbelt—beyond simply the South—was the region with the most pronounced punitive turn (H5).
Data and Measurement
To understand when, where and among which racial groups punitive opinion changed in the mid-century United States, I collect individual-level data on punitive attitudes from 16 Gallup polls (1953–1985) and 11 GSS polls (1972–1985) comprising repeated cross-sections of $\sim $ 40,000 individuals.Footnote 13 I restrict the analysis to pre-1985, allowing a long mid-century window to examine the punitive turn, but also limiting the potential contamination from later events, namely the “innocence turn” that other scholars argue emerged from the 1990s (Baumgartner, Boef and Boydstun Reference Baumgartner, De Boef and Boydstun2008; Ramirez Reference Ramirez2013a).Footnote 14
The theoretical construct of interest is punitive attitudes. I operationalize this construct by using responses to the question “Are you in favor of the death penalty for a person convicted of murder?” focusing on support or oppose responses with additional analyses incorporating don’t know (Appendix J).Footnote 15 I use this question because it is the only question on a punitive attitude asked frequently and with consistent wording throughout the period of study. In order to understand the existence of a punitive turn, it is necessary to know the nature of punitive attitudes consistently before and after any purported turn. Thus, it is essential to use a question asked as early as possible—here 1953.Footnote 16 The frequency of the question is also essential in order to engage in subgroup analysis while minimizing (random) sampling error. Finally, the consistency of the wording ensures that changes in observed responses over time are not a function of changes in how the question was asked.
In order to ensure any changes I document are attributable to changes in attitudes and not question inclusion, I focus only on support for the death penalty rather than integrate punitive attitude questions introduced almost entirely in later surveys.Footnote 17 For this inferential clarity, I trade the ability to capture a more comprehensive version of punitiveness as a theoretical construct. Yet, the salience of the death penalty during the period (Gottschalk Reference Gottschalk2006; Steiker and Steiker Reference Steiker and Steiker2020), and the correlation established in later years when multiple punitive questions are available between death penalty attitudes and other punishment questions leads me to believe that this question is still deeply relevant for understanding punitiveness.Footnote 18 Enns (Reference Enns2016) found that the favor the death penalty question correlates substantially with his latent measure of punitiveness (factor loading $ = .941 - .971$ , depending on the precise poll source) (p. 37). The only question exhibiting a higher correlation among the thirty-three Enns uses is belief that the death penalty deters crime ( $.983$ ).Footnote 19 The death penalty question has the advantage of asking directly about attitudes toward state-administered punishment as distinct from questions about adjacent, but arguably distinct, theoretical constructs (e.g., trust in police). Finally, that this is the most extreme punishment that the state in the U.S. administers suggests that relying on the death penalty question alone may underestimate levels of support, but insofar as it moves with other punitive questions in later periods suggests that in terms of change (arguably the most relevant, see Enns (Reference Enns2016)), death penalty support is a valid measure of an important component of punitiveness.Footnote 20
To correct for sampling issues, particularly relevant in the early surveys, all subsequent analyses are conducted on population-weighted data where weights are constructed via poststratification cell-weighting by race, region, gender, and education using census data (see Appendix E which includes results for unweighted data).Footnote 21
The Empirical Analysis
In this section, I answer the questions of when, where, and among which racial groups death penalty support changed between 1953 and 1985. Figure 2 presents aggregate trends in death penalty support, demonstrating an inflection point in support in the mid-to-late 1960s. Based on additional examination of these trends in Appendix C, I utilize 1967 to define the preturn and post-turn periods in the subsequent analyses.
Racial Differences in the Punitive Turn
First, I ask how this turn in punitive attitudes as measured by death penalty support differed by racial groups. I present the same average trends as before (Figure 3), splitting the sample into White and Black respondents. The trends in Figure 3 are consistent with the well-documented finding that, in levels, Black people consistently hold less punitive attitudes. Consequently, this does not support H1a that Black attitudes could not increase as much as other the attitudes of other racial groups because of ceiling effects.
To more formally consider differential trajectories by race in death penalty support, I use ordinary least squares (OLS) to estimate:
for $i$ respondent in $t$ survey. The parameter ${\gamma _t}$ is a vector of survey indicators that capture level differences in death-penalty support common to all respondents in a given survey, including, but is not limited to crime trends, media portrayals of crime, national political rhetoric, and so on; $white$ is an indicator equal to one for respondents who identify as White (indexed by $t$ because the data is structured as repeated cross-sections); post is an indicator equal to one in any year after 1967; X is a matrix of $z$ covariates, including geographic region indicatorsFootnote 22 finally, $\varepsilon $ is the idiosyncratic error term, which I estimate accounting for heteroskedasticity and clustering for just 8 regions, which is highly restrictive. The parameter ${\beta _1}$ in model 1 estimates the pre-to-post 1967 (punitive turn) difference in death penalty support between White and Black respondents. I treat this as a descriptive parameter.Footnote 23
Given the theoretical importance of not only race, but also context in producing punitive attitudes, in the model above, I include region-specific variation in homicide victimization rates per 100,000 for both White and non-White people in $\boldsymbol X$ . If attitudes vary importantly by region, it may be because individuals are more sensitive—via factors like personal experience, media exposure, or information diffusion in local networks—to crime occurring in their region, in addition to, or in place of the national level. Indeed, there are distinct regional patterns in homicide rates across regions (Appendix H).Footnote 24 I collected these data from vital statistics reports published by the Centers for Disease Control (Appendix H).Footnote 25 Homicide rates have three advantages: i) they capture the crime most closely related to the death penalty, ii) they are available from the earliest period of the attitude data, and iii) their disaggregation by race allows us to understand respondent attitudes in relation to in-group and out-group crime. In Appendix H, I demonstrate that the results are robust to the homicide growth rate, and to measuring crime more generally, using violent and property crime rates from the UCR system of the FBI (Kaplan Reference Kaplan2021); however, these latter data are only available from 1960 and are not race-specific. I do not separately include national crime, nor any national-level time-varying controls, as these are co-linear with the survey fixed effects ( ${\gamma _t}$ ). Thus, this model assumes death penalty support is a function of national crime rates (via ${\gamma _t}$ ), and a function of crime in the respondent’s region, but asks (and estimates) how the punitive turn was heterogeneously experienced by different racial groups. This is different from the question asked by much of the literature about which national factors move with the aggregate trend in punitiveness for which groups.
I find that White support for the death penalty grew differentially at the punitive turn as compared to that of Black respondents (Table 1). While White support was, on average, higher in the preperiod by 16 percentage points (estimate on $white$ ), in the post-1967 period, this difference grew by 12 percentage points more than that of Black respondents (estimate on $white \times post$ ) (supporting H2 relative to H1). The post-1967 period was to increase by three quarters the existing differential White support for the death penalty. Regional homicide rates are neither substantively nor statistically predictive of death penalty attitudes.
Notes: The above table presents estimates of equation 1. Standard errors are clustered at the region. Data are weighted using poststratification census weights (Appendix E). Average death penalty support over the above analysis period is 0.64 (sd = .48). See Appendix O for covariate estimates.
*p<.1; **p<.05; ***p<.01.
Race and Geography at the Punitive Turn
Next, I turn to asking where death penalty support grew the most, and whether any region-specific growth was differentially experienced by different racial groups. As motivated by the theoretical framework, my interest is first and foremost in the extent to which the South is an outlier in the punitive turn.
I begin with graphical evidence. Figure 4 presents the trends by race and region (plot a). Prior to 1967, trends in death penalty support were modestly declining for all subpopulations. After 1967, however, support among Southern Whites rapidly increased, matching levels of support among non-Southern Whites by 1972 and moving in near lock-step from that point forward. By contrast, support for the death penalty among Southern Blacks—which had exhibited similar levels and patterns to Southern Whites in the pre-1965 period—did not exhibit a sharp increase. Instead, it converged toward the level and trend of non-Southern Black people whose support increased, but did so more slowly and to a lower absolute level than Whites.
Figure 4. (plot b) summarizes these trends by presenting the difference in means between Black and White respondents in each of the two regions and each of the two time periods. On average, growth in the South’s racial difference in death penalty support from pre-to-post 1967 is pronounced, and larger than the growth in the non-South.
To estimate these differences more precisely, I adjust equation 1 to consider regional heterogeneity:
I use a binary indicator for South to ease interpretation.Footnote 26 In addition, I consider this equation for the sample of White respondents alone, asking whether White people in the South, specifically, experienced the punitive turn differentially from White respondents outside of the South. Third and finally, I combine the racial variation in equation 1 with the regional variation in equation 2 to estimate a triple difference specification on the full sample of respondents that compares differences in death penalty support: (1) between Black and White respondents, (2) before-to-after 1967, and (3) between the South and non-South (via the addition of $\beta \left({sout{h_{it}} \times whit{e_{it}} \times pos{t_{it}}}\right)$ ). This evaluates whether the difference between White and Black death penalty support grew by more in the South relative to the non-South from before to after 1967. Given that the growth in support might be driven in part by non-national time-varying factors, the triple difference uses the growth of other races and in other regions as way to benchmark how much White Southern support might have (counterfactually) grown if not for the factors unique to the South and to White people. That said, the identifying assumption that the attitudes of different races represent the true counterfactual for one another is extremely difficult to defend and I therefore treat the triple interaction as a descriptive parameter.
Table 2 presents the results. Model 1 indicates that, although the South was less supportive of the death penalty on average pre-1967 (estimate on $south$ ), death penalty support grew by approximately 5 percentage points more after 1967 as compared to the preturn period (estimate on $south \times post$ ). Death penalty support was not higher than other regions on average, post-1967. Rather, the South made up approximately 75% of the gap in death penalty support exhibited in the pre-1967 period between it and other regions. Model 2 (and all even numbered columns) adds the time-varying controls for regional homicide rates and estimates a similarly sized parameter on the interactions (robust to other crime measures in Appendix H).
Notes: The above table presents OLS estimates of equation 2. Standard errors are heteroskedastic robust to regional clustering. Data are weighted using poststratification census weights (Appendix E). Average death penalty support over the above analysis period is .64 (sd = .48). See Appendix O for covariate estimates.
* p<.1;
** p<.05;
*** p<.01.
In models 3 and 4, I restrict the sample to White respondents. I find that Whites in the South in the pre-1967 period had lower levels of support for the death penalty—by approximately 7 percentage points—than did their non-Southern counterparts (estimate on $south$ ). However, Southern Whites differentially increased their death penalty support after 1967 by approximately 7 percentage points (estimate on $south \times post$ ), making up the full difference in average pre-1967 difference between Southern and non-Southern Whites. Appendix P demonstrates that the Central South subregions experienced the largest changes among White people. Appendix I shows that in the post-1967 period, there is no evidence that Black attitudes became punitive in the same regionally defined way. Black people in the South had lower pre-1967 death penalty support despite higher homicide rates. Post-1967 death penalty support among Black respondents in the South declined by approximately 1–2 percentage points relative to the pre-1967 period, and I cannot distinguish that estimate from zero.
The results from the triple difference in model 5 more formally compare changes in White attitudes in the South to Black attitudes in the South pre-to-post 1967. Substantively, they suggest larger growth in death penalty support pre-to-post 1967 between White and Black respondents in the South as compared to the non-South ( $\sim $ 10 percentage points), but I cannot statistically distinguish the estimate from zero.Footnote 27 Nevertheless because White (models 3–4) and Black (Appendix I) respondents’ attitudes in the South were moving in different directions, the lack of statistical significance is likely to be an issue of power. Together, these results support H3.
I consider the magnitude of these differences in the punitive turn for the South relative to other regions using an alternative set of assumptions. Specifically, I ask: conditional on the sample, was the South exceptional in its experience of the punitive turn, or could these results have arisen by chance? To understand this, first, I estimate the models presented in columns 2 and 4 of Table 2 separately for the six other permutations of a single region as the region of hypothesized difference (“treatment”).Footnote 28 In all cases (Figure 5 plot a), I find that the estimates for other regions are less than or statistically indistinguishable from zero, consistent with the South growing fast in its death penalty support and being contained within the 0 of the indicator of other regions (lack of support for H4).
Furthermore, in Appendix P, I find little evidence that the Sunbelt, as an alternative regional grouping to the South, produces stronger results (lack of support for H5). Although it is the case that the Southwest region of the Sunbelt was more supportive of the death penalty than the South or North/Mid-West prior to 1967, the Southwest did not experience a change in its death penalty support in the mid-1960s in the way that the South did, relative to the remainder of the country. Insofar as the Southwest region of the Sunbelt contributed to national punitive policy, this paper’s results suggest that it was because of its relatively high (pre-existing) punitive levels, as measured by death penalty support, rather than a differential attitudinal change the region underwent during the 1960s.Footnote 29
I more formally use randomization inference—a Fisher exact test—to understand the probability of observing a coefficient on $south \times post$ as large as those in models 1 and 3 under the (sharp) null of no relationship for any respondents: ${H_0}:{\beta _{1,i}} = 0,\forall i$ , where ${\beta _{1,i}}$ is defined as the difference in death penalty support for respondent $i$ in the post-1967 period in the South ( $south = 1$ ) as compared to not being in the South ( $south = 0$ .Footnote 30 To conduct this inference, I assume that my sample comprises the full population.Footnote 31 I permute 2,000 random draws of respondent “assignment” to $south$ (maintaining region clustering), and evaluate in how many of these permutations of regional assignment I observe a coefficient as large as that estimated on the actual data ( ${\beta _1}$ ). I reject the null hypothesis at the 1% level (p-value $ = .0091$ ).Footnote 32 Figure 5 (plot b) shows the distribution of the parameter estimates from the permutations and the estimate from the actual data (vertical line). Thus, under an alternative set of assumptions, my results indicate that the differential post-1967 growth in death penalty support in the South was extremely unlikely to have arisen by chance.
In summary, these results suggest that white people, and those in the South, increased their support for the death penalty differentially in the mid-1960s. Given the regional and racial distribution of crime, and the inclusion of time-varying regional measures of crime, the results suggest that genuine concern for personal safety is unlikely to constitute the full explanation for this turn in attitudes. Although Black people were far more likely to experience crime as measured by homicide victimization rates, White people increased their punitiveness, as measured by death penalty support, more. And although Black people in the South experienced the highest crime rates, it was White people in the South whose death penalty support grew the most after 1967. An alternative theoretical model—directly empirically evaluated in Appendix H using homicide growth rates—that assumes higher elasticities of death penalty support increase at lower levels of crime would predict that White people in the non-South would have had a larger post-turn increase in death penalty support than those in the South; the opposite of what is found. Instead, the results can be interpreted such that insofar as experience with crime may have been a first-order activation of support for the state’s most punitive policy, that activation was filtered through respondent race and geographic context in systematic ways.
Toward Understanding the Role of Race and Region
Having established the role of race and region in the punitive turn in death penalty support, here I examine two possible explanations for why they played that role. First, I consider whether White people in the South became more responsive to crime from before to after the punitive turn relative to respondents in other regions. Second, I consider whether the patterns of partisan support for the death penalty are consistent with Southern White exodus from the Democratic party occurring among those whose attitudes changed at the turn. Given data availability, these empirical analyses take the form of better understanding dynamic moderated relationships.Footnote 33 Understanding this moderation requires data spanning the full-time frame, a significant challenge, particularly preturn, as many implications of from the theoretical framework are simply unmeasurable in a way that can be linked to nationally representative individual-level attitudes about punishment.
Moderation in the Relationship Between Crime and Death Penalty Support
I first consider whether region-specific and race-specific violent crime shaped death penalty support in the South differently from before to after the punitive turn. The previous analyses already account for time-varying region-specific crime as a factor directly affecting attitude formation—that is, the previous analyses account for the fact that death penalty support may have changed differentially in the South because the South differentially experienced more/growing crime.
Here, I ask whether the punitive turn was characterized by White people in the South becoming newly or differentially responsive to crime relative to Black people, or relative to people in other regions. If such temporal moderation occurs differently in the South, this would be consistent with the contention that the transformation of death penalty support in the South at the punitive turn was, at its root, a function of violent crime. Though the media or political actors may have served as mechanisms, drawing the public’s attention to crime (at the regional level), attitude formation was fundamentally rooted in the objective occurrence of crime (H1).Footnote 34
As above, I measure violent crime by region using homicide victimization rates. This approach proposes that individuals’ perceptions of their personal risk, the risk to their communities, or even to society more broadly are a function of not only what is happening nationally (the survey fixed effects), but what is also happening more locally. Of course, although there is significant regional variation in homicide rates across this period (Appendix H), perceptions may be shaped by even more local (e.g., county, city) occurrences, or by one’s personal experience of crime that may not be captured with this measure.
In Appendix H, I demonstrate that reducing within-region variation by restricting the sample to urban residents—those most likely to have close personal experience with violent crime (Miczek, Reiss and Roth Reference Miczek, Reiss and Roth1994)—does not change the results. It’s worth emphasizing that homicide victimization rates by race do not directly capture perpetrator race, only victim race. Although violent crime, particularly homicide, occurs predominantly within-race (Miczek, Reiss and Roth Reference Miczek, Reiss and Roth1994).Footnote 35
With this data, I analyze whether there is differential post-turn responsiveness to regional violent crime—that is, whether death penalty support became more strongly positively correlated with violent crime after the punitive turn, as compared to before; and whether that correlation varies by region. To do so, I augment equation 1 with interactions between regional homicide rates by race and indicators for the post-1967 period. I present estimates on samples restricted by respondents’ race in models in Table 3 for interpretability. I find that support for the death penalty responded differently pre-to-post 1967 to the White homicide rate among White respondents (estimate on $homicide rate (W) \times post$ in model 1), but that this responsiveness among White people was no different between the South and non-South (estimate on $homicide rate (W) \times south \times post$ in model 3). Thus, White respondents became more supportive of the death penalty for increases in the homicide rate from before to after the punitive turn; however, that change in support was no different between the South and non-South. Moreover, the post-turn change served to attenuate the preturn relationship (estimate on $homicide rate (W) in model 1). I consider this evidence inconsistent with H1. If anything, Black respondents’ in the South became slightly more supportive of the death penalty (by $ \lt 1$ percentage point) in response to the non-White homicide rate, but overall there were no regionally distinct changes in death penalty support among Black respondents (Appendix I). In Appendix H, I further demonstrate that this result holds when considering the homicide growth rate (lack of support for H1b), and when restricting the sample only to large urban areas. I interpret these results as indicating that changes in death penalty support among White respondents in the South were not a function of a changes in regional crime responsiveness, namely, co-racial homicide rates. Though it remains possible that more local experiences of crime (e.g., within one’s city rather than broader region) or highly individual victimization risk still shaped attitudes.Footnote 36
Notes: The above table presents OLS estimates of equation 1 with additional interactions of race-by-region homicide rate per 100,000 of the race relevant population. Standard errors are heteroskedastic robust to regional clustering. Data are weighted using poststratification census weights (Appendix E). Average death penalty support over the above analysis period is .64 (sd = .48).
* p<.1;
** p<.05;
*** p<.01.
Southern Racial-Partisan Realignment
It is well known that mid-century saw Southern Whites resorted from the Democratic to the Republican party as the Democratic party embraced Black people and their political concerns (Schickler Reference Schickler2016). While other work has identified racial attitudes as an important driver of White exodus from the Democratic party (Kuziemko and Washington Reference Kuziemko and Washington2018), this process has not been systematically linked to punitive attitudes.
Here, I analyze whether there is dynamic moderation in the relationship between partisan identification among White respondents in the South and their punitive attitudes, as measured by death penalty support. In particular, I analyze whether White Southern Democrats and Republicans were relatively equally supportive of the death penalty in the preturn period, and whether White Southern Republicans were more supportive in the post-turn period. I focus on White respondents in the South because their process of realignment is the best understood in the existing literature.
In the preperiod, a Southern White Democrat identified with a party that had constructed Jim Crow and fought against civil rights for Black people. After realignment and then the punitive turn, a White Southern Democratic was instead identifying with a far more ideologically liberal party that was more supportive of Black civil rights and a larger welfare state. If there is a positive or neutral relationship between Democratic identification and death penalty support before the punitive turn, and a negative relationship afterward, this would suggest that racially resentful Whites who overwhelmingly exited the Democratic party were also becoming differentially more punitive (H2). Theoretically, this resentment may have operated through the appeal of Republican’s law and order policy, social welfare policy, or both. On this last point, unfortunately, the data are not comprehensive enough to say.
Ideally, I would be able to analyze a true panel in order to assess heterogeneity in partisan responses among those White people who I observe switch from the Democratic to Republican parties, holding individual time-invariant factors fixed. Unfortunately, to my knowledge, no such panel exists over this period. With repeated cross-sections, therefore, we must be mindful that we cannot account for cross-sectional samples capturing stable partisan identifiers rather than those who switched their identification. Furthermore, while I would ideally replicate Kuziemko and Washington (Reference Kuziemko and Washington2018)’s strategy of measuring prejudice and partisanship, the prejudice questions co-occur with the death penalty question only after the punitive turn, making it impossible to use them to evaluate dynamic moderation.Footnote 37
Though noisy, graphical evidence is suggestive of this temporal relationship (Figure 6). Before 1963, Southern White Democrats were, on average, more supportive of the death penalty than Southern Whites who identified with other parties. After 1967, this relationship reversed. I formally estimate this change by considering my baseline specification with a partisan indicator ( $Democrat$ ) including weak or leaning support, and an interaction between that indicator and post-1967 ( $Democrat \times post$ ).Footnote 38 Table 4 presents the results. Model 1 considers all survey years, while Model 2 excludes the 1965 and 1966 survey which occurred after Kuziemko and Washington (Reference Kuziemko and Washington2018) identified the significant inflection point in partisan alignment (1963) and before the inflection point in punitive opinion (1967). Perhaps understandably, given the sample restrictions to White respondents in the South, the results are noisy. But they are at least suggestive of a link between death penalty support and partisan realignment. After the punitive turn, White respondents in the South identifying as Democrats were less likely to support the death penalty by approximately 2–7 percentage points, though I cannot statistically distinguish the estimate from zero on the full sample. Although I cannot estimate more precisely the racial attitudes of those in my sample who identified with each party, the implication from Kuziemko and Washington (Reference Kuziemko and Washington2018)’s evidence is that this switch was undertaken predominantly by racially resentful White people, suggesting that prejudice played in an important role in the punitive turn (consistent with H2).
Conclusion
In this article, I study the nature of changing death penalty support by evaluating racial and geographic heterogeneity in the mid-20th century United States using a long time-series of public opinion data.
I find that the punitive turn—or at least the turn in the component of punitiveness captured by support for the death penalty—was in fact a re-turn to prior levels of punitive support, followed by subsequent expansion. Before-to-after the punitive turn, White people increased their death penalty support by more than Black people; respondents in the South increased their support by more than those in the North; and White Southerners increased their support the most. The pattern of moderation of race-regional relationships by partisan identification is suggestive of racial-partisan realignment—known to be driven in significant part by racial resentment—differentially shaping attitudes from before-to-after the turn.
At the same time, there is no evidence that Southerners, nor Southern White respondents, increased their death penalty support because they became differentially more sensitive to changing violent crime in their region. I interpret these findings as suggesting that the experience of regional crime was insufficient on its own to produce the documented changes in death penalty support. Instead, race and additional contextual factors—here encapsulated by the unique political and cultural history of the South—provided substantial mediation of the experience of regional crime. Importantly, however, this does not preclude that differential sensitivity occurred in response to even more geographically localized experienced of crime and risk, rather than trends in the region more broadly.
The results further suggest that theories of the punitive turn that emphasize Northern, Mid-Western, or Sunbelt attitudinal change as driving forces of the turn are unlikely to explain the turn without revision. While the Southwest region of the Sunbelt was more supportive of the death penalty on average prior to 1967 as compared to the South, attitudes in the Southwest did not undergo a transformation in the 1960s, while those in the South did.
While these results speak to the dynamics of public opinion by race and region, they have important limitations that offer opportunities for future research. Although studying death penalty support offers a means of understanding punitive attitudes given its correlation with other attitudes, the multifaceted nature of most conceptualizations of punitiveness, and the unique history of the death penalty, mean that care must be taken in generalizing the results. The results here speak to attitudes about a specific extreme form of penal sanction, rather than the intensity of sanctions or the goals of punishment. In addition, they capture attitudes through just one mode of elicitation (Gelb Reference Gelb2008). That the death penalty is arguably the most extreme punishment that the state can administer might suggest that levels of support for other punishment forms may differ, though as to whether support for those forms differed over time, by region and by group requires data beyond what is available here. The results do not and cannot offer an explanation of initial level differences by race and region, nor the decline in death penalty support—either broadly or in the South, specifically—from the 1950s. These are deeply interesting questions that would more thoroughly contextualize the findings here, but ones which require different theory, data and research design to answer. That the punitive turn in death penalty attitudes was a re-turn to earlier death penalty support followed by subsequent expansion suggests that one of the features of the mid-century was to de-couple the relationship between this punitive attitude and the more objective experience of crime, given that violent crime was declining from the 1970s while death penalty support, and punitive attitudes more generally, continued to grow.
Furthermore, these results are not a definitive test that separates mechanisms proposed in extant theories. By way of example, I cannot evaluate whether White people, nor those in particular regions, differentially consumed particular types of violent media before-to-after the punitive turn (Enns Reference Enns2016). Nor can I evaluate whether partisan realignment is linked to punitiveness because of law and order policy, welfare policy, both or some other divergent partisan policy (Flamm Reference Flamm2005; Weaver Reference Weaver2007). Novel data collection could aim to address those questions in the future. Though evidence suggests that support for rehabilitation declined from the 1960s to the 1980s as punitiveness was increasing (Allen Reference Allen1981; Flanagan and Caulfield Reference Flanagan and Caulfield1984), it’s not known how relative support for rehabilitation changed before the 1960s. The relationship between trends in both of those attitudes would provide a clearer understanding of what precisely changed during the mid-century.
The results are in no way the last word on the punitive puzzle, but they document a racial and geographic pattern in punitive attitudes previously missing from the literature. Given the importance of punitive opinion in making punitive policy (Baumgartner, Boef and Boydstun Reference Baumgartner, De Boef and Boydstun2008; Enns Reference Enns2016), when punitive opinion changes, and for whom it changes, has important implications for the exercise of state punishment. Beyond what might be considered a case-specific question in American politics, these results emphasize the way in which genuine contextual concerns—e.g., crime—can be mediated and augmented by socially constructed attributes—e.g., race—and experienced in geographically diverse ways. Furthermore, the legacies of race and punishment and their link to a particular region in the U.S. case suggest that the identification of relevant attributes and geographies in other studies is likely to be highly context-specific.
Supplementary material
For supplementary material accompanying this paper visit https://doi.org/10.1017/rep.2024.26
Acknowledgments
Thank you to John Aldrich, Josh Clinton, Nick Eubank, Sunshine Hillygus, Matt Grossman, and participants at the inaugural Conference in American Political Economy (CAPE) for helpful feedback.
Funding statement
This research did not receive any specific financial support.
Competing interests
None.