Introduction

Punitive public views have been a central feature in accounts of burgeoning punishment systems and the development of mass incarceration in the U.S. (Clear and Frost 2014; Garland 2001; Simon 2007; Tonry 2004). In describing the socio-political context of the buildup of harsh sentencing schemes and practices from the 1970s through the 1990s, theoretical narratives have highlighted the importance of desire among the public for harsh punishment and the utility of that punitiveness for policymakers seeking to “govern through crime” (Simon 2007).

Given the prominence of public views or sensibilities in understanding “the punitive turn,” a substantial body of research has considered the nature and sources of American views on crime and its control (Cullen et al. 2000; Garland 2001; Tonry 2004). In an important contribution, Unnever and Cullen (2010) operationalized and evaluated the strength of three models to explain the social origins of punitive American views. Using data from the American National Election Study (ANES) in 2000, they found that a model for anti-Black racial resentment better explained support for punitive views (i.e., support for the death penalty and more general punitiveness) than did models focusing on concern about escalating crime, mistrust in courts, moral decline, and racial stereotypes about African Americans.

The present study takes up the challenge given by Unnever and Cullen (2010, pp. 117–118) to replicate and expand their analysis to consider alternate theoretical perspectives, alternate data sources, and the nature of punitive views in the post-9/11 era. Using data from the General Social Survey (GSS) for 2000, the three models identified by Unnever and Cullen (2010) are operationalized and reevaluated. These models are also estimated using GSS data for 2014. An additional model is operationalized to consider the relationships between neoliberal (anti-poor; anti-welfare) views and punitiveness, controlling for other factors. A final set of models is estimated to evaluate each of the four perspectives for both 2000 and 2014. This approach allows us to assess the utility of major theoretical perspectives to explain punitive views in the early twenty-first century.

Social Foundations of Punitive Views

The extant literature suggests a variety of perspectives on the social foundations of American punitive views. As Unnever and Cullen (2010, p. 101) note, it would be impossible to empirically evaluate in a single project all available explanations for punitiveness. Punitiveness or punitivity is a complex construct, one that is made up of psychological (Maruna et al. 2004), collective (Jones 2010), and contextual elements (Whitman 2005). It is possible, however, to consider some of the major perspectives available on the determinants and nature of punitive views. The three perspectives identified and operationalized by Unnever and Cullen (2010) are reviewed briefly below. We then consider an additional perspective that focuses on neoliberal animus toward the poor, which prior research suggests is an important correlate to anti-Black sentiment in shaping punitive views, but which has not been evaluated against the other theory-derived models considered here.

Escalating Crime, Moral Decline, Racial Animus, Animus Toward the Poor

Drawing on major late twentieth century treatises on the sociology of punishment, Unnever and Cullen (2010) identify three prominent perspectives on punitive American views. The first, based primarily on the work of Simon (2007) and Garland (2001), they call the “escalating crime-distrust” model. This perspective suggests that punitive views are formed based on the perceptions among Americans that crime rates are increasing (even when they are decreasing, as they have been since the mid-1990s) and that the criminal justice system, and courts in particular, are unable to secure their safety. Since the 1960s, the public, sensing rising risks from crime, has turned away from social solutions to crime and has instead looked to severe sanctions to combat crime and protect victims. Concern about crime and the sense that government (and courts in particular) cannot control it have thus contributed to punitive views (Garland 2001; Simon 2007; Zimring 2001). In their analysis of ANES data, Unnever and Cullen (2010) found significant relationships between punitive views and the perception that crime rates are increasing, but they did not find that lack of trust in the courts or an interaction between lack of trust and the sense that crime is increasing significantly predicted punitive views.

Another perspective on punitive views suggests that punitiveness reflects a symbolic response to social anxiety and a sense of moral decline. The idea that punishment serves expressive and symbolic purposes is, of course, reflected in the work of many prominent thinkers (Durkheim and Simpson 1933; Garland 2001; Tonry 2004). In their particular focus on moral decline, Unnever and Cullen (2010) lean on the work of Tyler and Boeckmann (1997), who found that those who are most unsure of the stability of morals and values in society are most likely to support harsh punishment. Using support for Three Strikes legislation in California as their focus, Tyler and Boeckmann (1997) found that people who believed that the stability of families and social cohesion were being threatened were most supportive of punitive penal policies (Tyler and Boeckmann 1997). Using a scale constructed from four measures from the ANES (with a focus on shifting lifestyles, unsettled morality, and traditional family values), Unnever and Cullen (2010) found that concerns about moral decline are positively related to punitive views overall, but when considering models that control for alternate theoretical constructs, moral decline was not predictive of death penalty support.

A third model operationalized by Unnever and Cullen (2010) considers the influence of anti-Black racial animus on punitive views. Drawing on broad evidence that people who hold racially prejudiced views of Black people are also punitive in their crime control views (Bobo and Johnson 2004; Bobo and Kluegel 1993; Pickett et al. 2014, pp. 385–387; Unnever 2008; Unnever et al. 2008a, b), they constructed scales for both anti-Black racial stereotypes and racial resentment. While racial stereotypes failed to predict punitive views in their study (an issue we consider in supplemental analyses in the present study), racial resentment, a form of symbolic racism focused on opposition to African Americans’ receiving assistance to succeed economically, was the strongest predictor of support for the death penalty and more general punitive views, controlling for models of moral decline and crime control concerns.

A fourth perspective, one not evaluated by Unnever and Cullen, is that American punitiveness is rooted in the neoliberal “war against the poor.” Drawing primarily on the work of Wacquant (2009, 2010) and Piven (2015), an animus towards the poor model suggests that the buildup of the carceral state was part of a larger social shift that pushed the poor further to the margins of society and reduced access to welfare support. The expansion of American prisons is framed as not only hiding away criminal offenders, but also removing from sight those who challenge American ideals of individualistic achievement and prosperity: the poor. As Wacquant explains, “…punitive containment offers relief not to the poor but from the poor, by forcibly ‘disappearing’ the most disruptive of them…” (Wacquant 2009, p. 295).

Neoliberal animus toward the poor is also marked by a desire to restrict access to and make more punitive welfare or public assistance programs. As is often noted, the development of increasingly severe penal sanctions has tracked with disinvestment in welfare programs (Beckett and Western 2001; Davies and Worden 2009). Cuts to welfare, often justified by neoliberal arguments about “workfare” and privatization of government resources, have co-occurred as prison populations boomed in the 1970s and through the 1990s (Piven 2015; Wacquant 2010). Increasing punitiveness in the penal system, manifested by longer sentences and greater propensity to incarcerate (Clear and Frost 2014), was mirrored by increasingly restrictive welfare policies throughout the 1990s (Wacquant 2010). Not only has funding for welfare programs been cut, but welfare requirements and processes have increasingly come to resemble extra-carceral punishment of the poor.

Prior research indicates that support for welfare and support for punitive sanctions are inversely related (Beckett and Western 2001; Hogan et al. 2005; Pickett et al. 2013; Rubin 2011). Research also suggests that support for economic individualism, the belief that individuals are responsible for their own welfare and economic success, and punitive sanctions are positively related (Kornhauser 2015). In addition, in prior research based on the same ANES dataset from 2000 used in their 2010 study, Unnever et al. (2008a) found that people who reject egalitarianism, the desire for equitable distribution of rights and opportunities, tend to endorse punitive views (and oppose addressing the structural causes of crime) (Unnever et al. 2008a).

While prior studies have conceptualized anti-poor beliefs and ideas in different ways, the story told by the findings is consistent: Americans who hold beliefs that privilege individualism and who oppose policies that assist the poor tend to favor more severe punishment than those who do not. In addition, it is clear that race is infused in public typifications of both crime and welfare (Chiricos et al. 2004; Gilens 1999; Neubeck and Cazenave 2001; Pickett and Chiricos 2012). Moreover, negative racial stereotypes (e.g., Black people are lazy) are notably “merged” and co-occur with anti-poor resentments (e.g., not wanting welfare assistance given to “undeserving” minorities) (Wheelock et al. 2012). In particular, as Unnever et al. (2008a) point out, Americans with anti-Black racial animosity often support punitive sanctions and oppose welfare policies because they see each position as an appropriate response to people, typified as African American males, who have been socially and politically constructed as both “lazy” and “criminal.”

What is unclear from prior research is the explanatory power of animus toward the poor when other theoretical models are considered. As Kornhauser recently pointed out (2015, p. 43), research has yet to consider the influence of anti-poor views on punitive views in relation to other theory-based factors. While Unnever et al. (2008a) considered both racial animus and inegalitarianism in their models (and included a “feeling thermometer” about people on welfare as a control variable), we fill a gap in the literature by incorporating animus toward the poor into a study that controls for a broader range of well-established and previously considered theoretical models.

The Current Research

Research Goals

The goals of this research are to attempt to replicate Unnever and Cullen’s (2010) previously examined models for Americans’ punitive views using an alternate dataset for 2000, and to extend that research to consider an alternate model that addresses animus toward the poor. As noted, by including a model for animus toward the poor, this research expands upon prior research that has examined relationships between welfare views and punitive views but has failed to control for the alternate explanations operationalized here. In addition, this research extends prior research by considering data from 2014. Unnever and Cullen (2010, p. 118) encouraged this kind of replication and extension by noting that research should consider alternate theoretical perspectives on punitive views, replicate their approach with alternate datasets, and consider the determinants of punitive views in the post-9/11 era.

Beyond the value of scientific replication for the accumulation of reliable findings, it is particularly important to consider models of punitive views in America using updated data because evidence suggests recent shifts in public opinion. The early twenty-first century saw a decline in American punitiveness across a number of measures (Ramirez 2013). Further, research suggests that some determinants of punitive views may have shifted in recent decades. For example, while Unnever and Cullen (2012) found that racial prejudice (a form of symbolic racism focused on interracial marriage and neighborhood integration) was a consistent predictor of support for the death penalty in both 1990 and 2000, they also found that anti-Black stereotypes were predictive of support for the death penalty among White Americans in 1990 but not in 2000 (Unnever and Cullen 2012, p. 536).Footnote 1 In explaining this finding, Unnever and Cullen point to the importance of considering shifts in the determinants of punitive views: “…their impact on punitiveness are social products, malleable to some extent, and thus potential targets for change” (Unnever and Cullen 2012, p. 537).

There are many reasons to suspect that the determinants of punitiveness may have shifted in the early twenty-first century. For example, the salience of crime and its control in public and political discourse has declined as terrorism became a central focus after 9/11 (Altheide 2006; Pew Research Center, n.d.; Simon 2007). After decades in which severe sentencing policies were justified in response to widespread fear of crime and punitive public views, new trends are emerging: Crime rates are as low as they have been since the 1950s (Comey 2014); incarceration rates have stabilized and are dipping in some places (Greene and Mauer 2010); there is talk of “justice reinvestment” (Clear and Frost 2014; Listwan et al. 2008), “smart on crime” rather than “tough on crime” policies (Green 2013, 2015; The Smart on Crime Coalition 2011), and opportunities for decarceration (Doob and Webster 2014; Tonry 2014). As Unnever and Cullen (2010, p. 118) note, “It is not clear that [recent shifts] transformed the sources of punitiveness, but such a possibility exists and should be evaluated.” By considering data for 2014, we explore that possibility.

Analytical Approach

For the replication portions of this analysis, each of Unnever and Cullen’s (2010) three original models are operationalized with measures from the 2000 GSS that very closely approximate their original models. For our expansion analysis, a fourth model, for animus toward the poor, is operationalized. Models are also evaluated using GSS data from 2014. For both of the years considered, 2000 and 2014, we follow Unnever and Cullen (2010) lead in estimating regression analyses for each theoretical perspective individually, and then evaluating all of the theories against one another in a combined model. For the sake of parsimony, we present here only the findings for our combined models (along with a reduced form test of animus toward the poor). We also conduct supplemental analyses to consider possible mediation effects of racial stereotypes on punitiveness via racial resentment and animus toward the poor. Our measure of support for the death penalty is dichotomous, so we use logistic regression for those models. Our measure of punitive views (the perception that sentences are not harsh enough) is ordinal, so we use generalized ordered logistic regression for those models.

Data

This analysis uses nationally representative survey data from the U.S. General Social Survey (GSS) for 2000 and 2014. GSS data are collected on a biennial basis by the National Opinion Research Center (Smith et al. 2015). Data are derived from in-person interviews with American adults (non-institutionalized persons over 18 years of age) using a multistage cluster design approach. The sampling strategy employed in the GSS creates nationally representative samples. At the time of this writing, the most recently released data are for 2014. Unlike the ANES, which included general punitiveness measures only on the surveys in 2000, the GSS contains relevant variables that are consistently measured over time.

Dependent Variables

Two dependent variables are considered in this research: support for the death penalty and punitive views (as indicated by believing criminal sentences are not harsh enough). In the GSS, support for the death penalty is measured by a question that asks, “Do you favor or oppose the death penalty for persons convicted of murder?” In 2000, 69 % of Americans indicated support for the death penalty; in 2014, that figure dropped to 65 %.Footnote 2 Unlike the ANES, the GSS did not ask respondents about the intensity of their death penalty views in either 2000 or 2014. This variable was coded so that support for the death penalty was coded as 1, while not supporting the death penalty was coded as 0.

Our second dependent variable measures punitive views. In their analysis, Unnever and Cullen (2010) measured punitive attitudes by constructing a scale that combined four measures. Each measure asked respondents, in slightly different ways, whether they prefer social solutions to crime (improving schools, addressing poverty), or punitive approaches (ensuring criminals are punished). In contrast, we use a single measure assessing public views on the severity of courts in dealing with criminals with the question, “In general, do you think the courts in this area deal too harshly or not harshly enough with criminals?”Footnote 3 In 2000, 75 % said that courts are not harsh enough; that figure dropped to 63 % in 2014 (not shown). As respondents could indicate three potential levels of punitiveness (too harsh, about right, not harsh enough), this variable was coded from 1 to 3 to indicate increasing punitiveness; those who said that courts are not harsh enough are considered the most punitive (3).

Crime Concerns and Distrust in Courts

In assessing distrust in courts, Unnever and Cullen (2010) used a feeling thermometer that asked respondents how they feel toward the U.S. Supreme Court on a scale from 0 to 100 (unfavorable and cold to favorable and warm). In our replication, we measure distrust of courts using a measure of confidence in the U.S. Supreme Court. In the GSS, respondents are asked, “I am going to name some institutions in this country. As far as the people running these institutions are concerned, would you say you have a great deal of confidence, only some confidence, or hardly any confidence at all in them? U.S. Supreme Court.” This measure was coded so that 1 indicates a great deal of confidence, while 3 indicates hardly any confidence.

To assess public concern about crime rates, we used a combination of measures on spending for “rising” crime and law enforcement. The GSS samples for 2000 and 2014 used split samples for these measures, with random assignment to either the first or second item. The wording for the first item is: “I’m going to name some … problems [in this country], and for each one I’d like you to tell me whether you think we’re spending too much money on it, too little money, or about the right amount. Halting the rising crime rate.” It should be noted that this question wording, which indicates rising crime rates, was used even in years when crime rates were falling. This is useful for our purposes in that it taps into widespread misconceptions about crime rates (Roberts et al. 2003). This measure differs from the measure used by Unnever and Cullen (2010), which asked people whether they believe crime rates are getting worse, better, or staying the same. The second item used the same set up of the question, but substituted “law enforcement” in place of rising crime rate. The combined measure ranged from 1 to 3, with 1 indicating less concern (i.e., spending too much), and 3 indicating more concern (i.e., spending too little). To ensure the question wording did not unduly influence results, a dichotomous variable indicating which version of the question a respondent received was evaluated in each model that utilized the combined public concern variable. As this indicator variable was non-significant in all models (analyses not shown), and no other results significantly or substantively changed, it was dropped from the final models.Footnote 4

Social Anxiety

In Unnever and Cullen’s (2010, p. 110) analysis, they operationalized a measure of perceived moral decline by combining together ANES measures of public views on traditional family values, the social threat of new lifestyles, and moral shifts. The GSS does not include directly comparable measures. However, our reading of the relevant literature suggests that in shaping punitive views, social anxiety occupies the same conceptual space as concern about moral decline.Footnote 5 In particular, research suggests that punitive views can be interpreted as expressive or symbolic responses to social pessimism and insecurity (Garland 2001; Tyler and Boeckmann 1997). In this analysis, we operationalize social anxiety.

In order to assess social anxiety, we used a standardized scale that combined three GSS measures. The three measures ask about trustworthiness of people: (1) “Would you say that most people can be trusted or that you can’t be too careful in life?” [Three point scale]; (2) Helpfulness of people: “Would you say that most of the time people try to be helpful, or that they are mostly just looking out for themselves?” [Three point scale]; and (3) Whether people are fair: “Do you think most people would try to take advantage of you if they got a chance, or would they try to be fair?” [Three point scale].Footnote 6 These measures were standardized and combined into an averaged scale with an alpha coefficient of .66 in 2000 and .64 in 2014. Higher values indicate more social anxiety.

Racial Animus

Following Unnever and Cullen’s (2010) lead, we constructed two standardized scales to assess anti-Black racial animus: one measures racial resentment and the other racial stereotypes.Footnote 7 The racial resentment scale was created using five GSS items: (1) “Some people say that because of past discrimination, blacks should be given preference in hiring and promotion. Others say that such preference in hiring and promotion of blacks is wrong because it discriminates against whites … Are you for or against preferential hiring and promotion of blacks? If favors: Do you favor preference in hiring and promotion strongly or not strongly? If opposes: Do you oppose preference in hiring and promotion strongly or not strongly?” [Four point scale]; (2) “I’m going to name some … problems [in this country], and for each one I’d like you to tell me whether you think we’re spending too much money on it, too little money, or about the right amount. Improving the conditions of Blacks.” [Three point scale]; (3) “I’m going to name some … problems [in this country], and for each one I’d like you to tell me whether you think we’re spending too much money on it, too little money, or about the right amount. Assistance to blacks.” [Three point scale]; (4) Respondents were asked to indicate their agreement with the following statement on a scale from strongly agree to strongly disagree: “Irish, Italians, Jewish and many other minorities overcame prejudice and worked their way up. Blacks should do the same without special favors.” [Five point scale]; (5) “Some people think that (Blacks/negroes/African-Americans) have been discriminated against for so long that the government has a special obligation to help improve their living standards; they are at point 1. Others believe that the government should not be giving special treatment to (Blacks/negroes/African-Americans); they are at point 5. Where would you place yourself on this scale, or haven’t you made up your mind on this?” [Five point scale].

The two questions on “Improving the conditions of Blacks” and “Assistance to Blacks” were asked of randomly split samples similar to the rising crime and spending on law enforcement questions. Prior to creating the final scale, we followed the same method for combining these two variables together as we used with the crime control questions. Similar to the crime control questions, we evaluated the wording differences using an indicator variable in these models. Results indicated the question wording was non-significant in these models, and no other results changed substantively or significantly (results not shown); thus this indicator variable is not included in the final models. Higher values on this scale indicate more racial resentment, and the alpha coefficient is .72 for both 2000 and 2014.

Our scale for racial stereotypes was constructed from two items: (1) “The second set of characteristics asks if people in the group tend to be hard-working or if they tend to be lazy. Blacks” [1 = hardworking; 7 = lazy]; and (2) “Do people in these groups tend to be unintelligent or tend to be intelligent? Blacks.” [1 = unintelligent; 7 = intelligent]. Variables were recoded so that higher values on the scale indicate more negative racial stereotypes (lazy and unintelligent). The alpha coefficient for this scale is .61 in 2000 and .40 in 2014.Footnote 8 The correlation between the two racial animus scales is .30 in 2000 and .29 in 2014.

Animus Toward the Poor

A standardized scale was created to assess animus toward the poor and rejection of welfare. Four items were used: (1) “Some people think that the government in Washington should do everything possible to improve the standard of living of all poor Americans; they are at point 1. Other people think it is not the government’s responsibility, and that each person should take care of himself; they are at point 5. a. Where would you place yourself on this scale, or haven’t you made up your mind on this?” [Five point scale]; (2) “Some people think that the government in Washington ought to reduce the income differences between the rich and the poor, perhaps by raising the taxes of wealthy families or by giving income assistance to the poor … Think of a score of 1 as meaning that the government ought to reduce the income differences between rich and poor, and a score of 7 meaning that the government should not concern itself with reducing income differences. What score between 1 and 7 comes closest to the way you feel?” [Seven point scale]; (3) “I’m going to name some … problems [in this country], and for each one I’d like you to tell me whether you think we’re spending too much money on it, too little money, or about the right amount. Assistance to the poor” [Three point scale]; and (4) “I’m going to name some … problems [in this country], and for each one I’d like you to tell me whether you think we’re spending too much money on it, too little money, or about the right amount. Welfare.” [Three point scale].

The two questions on “assistance to the poor” and “welfare” were asked of randomly split samples similar to the rising crime and spending on law enforcement questions. Prior to creating the final scale, we followed the same method for combining these two variables together as we used with the crime control questions. The scale variable was recoded so that higher values on the scale indicate more animus toward the poor, and the alpha coefficient for the scale is .57 in 2000 and .65 in 2014.Footnote 9 Similar to the crime control questions, we evaluated the wording differences using an indicator variable in these models. Results indicated the question wording was non-significant in these models, and no other results changed substantively or significantly (results not shown); thus this indicator variable is not included in the final models.Footnote 10

Covariates

For our replication and expansion, we used the same nine covariates included in Unnever and Cullen’s (2010) research: religiosity, authoritarianism, political orientation, age, race, sex, education, region (South), and urban, and each has been found in prior research to predict punitive views.Footnote 11 For religiosity, we created a standardized scale with five items: (1) Fundamentalism/liberalism of respondent’s religion [1 = liberal; 3 = fundamentalist]; (2) “How often do you attend religious services?” [0 = never; 8 = more than once a week]; (3) “Would you call yourself a strong [religious preference] or a not very strong [religious preference]?” [1 = no religion; 4 = strong]; (4) “About how often do you pray?” [1 = never; 6 = several times a day]; and (5) “Which of these statements comes closest to describing your feelings about the Bible? The Bible is the actual word of God and is to be taken literally, word for word, or the Bible is the inspired word of God but not everything in it should be taken literally, word for word, or the Bible is an ancient book of fables, legends, history, and moral precepts recorded by men” [1 = fables; 3 = word of God]. These variables closely match those used by Unnever and Cullen (2010). Responses to variables were recoded so that higher values indicate more religiosity. The alpha coefficient for the scale is .79 and 2000 and .83 in 2014.

Authoritarianism was also measured using a standardized scale. Three items were included: (1) “If you had to choose, which thing on this list would you pick as the most important for a child to learn to prepare him or her for life? a. Which comes next in important? b. Which comes third? c. Which comes fourth? To obey” [1 = least important; 5 = most important]; (2) “If you had to choose, which thing on this list would you pick as the most important for a child to learn to prepare him or her for life? a. Which comes next in importance? b. Which comes third? c. Which comes fourth? To think for himself or herself” [1 = most important; 5 = least important]; (3) “Do you strongly agree, agree, disagree, or strongly disagree that it is sometimes necessary to discipline a child with a good, hard, spanking?” [1 = strongly disagree; 4 = strongly agree]. Responses to variables were recoded so that higher values indicate more authoritarianism. The alpha coefficient for the scale is .61 in 2000 and .62 in 2014.Footnote 12

Political orientation was measured by one GSS question that closely matches the ANES variable used by Unnever and Cullen (2010): “We hear a lot of talk these days about liberals and conservatives. I’m going to show you a seven-point scale on which the political views that people might hold are arranged from extremely liberal—point 1—to extremely conservative—point 7. Where would you place yourself on this scale?” Age was measured in years, and ranged from 18 to 89. Race and sex were measured as dichotomous variables [0 = not Black; 1 = Black]Footnote 13 [0 = female; 1 = male], education was measured by number of years completed in school [from 0 to 20]; region was measured as South or non-South [0 = not South; 1 = South];Footnote 14 and urban residence was coded if respondents lived in cities of more than 50,000 persons, the legal lower limit for an “urbanized area” according to the U.S. Census Bureau (U.S. Census Bureau 2015) [0 = non-urban; 1 = urban].

We handled missing data in the independent variables using the multiple imputation command in Stata 14 (mi impute) prior to scale creation.Footnote 15 This process involved generating 20 iterations of imputed data per respondent for each missing value, estimated based on the non-missing dependent and independent variables for that respondent. All descriptive statistics and model results hereafter report the combined imputation estimates after survey weights were applied.Footnote 16 Descriptive statistics are provided in Table 1 by model outcome (i.e., death penalty, punitive views) and year.

Table 1 Descriptive statistics by model and year
Full size table

Plan of Analysis

The first step of the analysis involved the multiple imputation corrections noted earlier, followed by the creation of all standardized scales. In all models, multiple imputation corrections were used and weights were applied using the weight variable compwt supplied in the GSS datasets available at the ‘SDA: Survey Documentation and Analysis’ website.Footnote 17 Following this, for each set of theoretical models being considered, we estimated two different types of models, with results reported as both unstandardized coefficients and odds ratios.

For the first set of models, logistic regression models were estimated to predict the dichotomous support for the death penalty variable for both 2000 and 2014 data. For support for the death penalty, odds ratios can be interpreted as the increase in odds of supporting the death penalty given a one-unit increase in the independent variable.

For the second set of models, partial proportional odds models using generalized ordinal logistic regression were estimated to predict punitive views (i.e., harshness of the courts) for both 2000 and 2014 data. The specific Stata command used to estimate these models was gologit2, which is a user-written command (Williams 2006).Footnote 18 Post-hoc analyses of the Brant test (1990) indicated violation of the parallel lines (proportional odds) assumption, mainly due to a small subset of variables in each model. This was also confirmed via analysis of visual graphs of coefficient estimates across levels of the dependent variable (not shown). The use of partial proportional odds models account for violations of the parallel lines (proportional odds) assumption by specific variables (as identified initially through the use of the autofit option on the non-survey weighted data), as these models allow the coefficients of those variables to not be constrained by the parallel lines assumption. Thus, the unconstrained variables will have coefficients that vary at different categories of the dependent variable. Coefficients for variables that do meet the parallel lines assumption are constrained so that they do not vary across categories of the dependent variable. Due to the use of imputed data and survey weights with the partial proportional odds models, tests of overall model fit (e.g., likelihood-ratio or Wald χ2) are not available (Williams 2006, p. 68).Footnote 19

For the punitive view models, coefficients for the constrained variables can be interpreted as standard ordinal logistic regression coefficients (Institute for Digital Research and Education 2015). For the unconstrained variables, coefficients are considered at different levels of the dependent variable. In this case, a comparison is first made between the effects of the unconstrained independent variables for the two more punitive categories (for those who said court sentencing is “about right” or “not harsh enough”) compared with the least punitive category (for people who say courts are “too harsh”). A second comparison is made for the effects of unconstrained variables in the most punitive category compared with the least and less punitive categories (for those who believe courts are “too harsh” or “about right”). In each of these two comparisons, positive coefficients are interpreted to mean that higher values on the independent variable (e.g., the expression of more social anxiety) are associated with a greater likelihood of being in a more punitive category. Negative coefficients indicate that higher values of the independent variable are associated with a lower likelihood of being in a more punitive category (Williams 2006, p. 63).

Uncentered variance inflation factors (not shown) indicated some potential multicollinearity among measures within the models. In particular, and as noted in the results section, the measures for anti-Black racial resentment and racial stereotypes appeared to be collinear. However, when alternate models were estimated with racial resentment and racial stereotypes isolated, there was little change in the strength, direction, and significance of the observed effects. Given the lack of change in these isolated models, we decided to leave both measures in the final models.

Results

Three Original Models

We begin our review of the findings by considering the results for models assessing the three perspectives operationalized by Unnever and Cullen (2010) using GSS data for both 2000 and 2014. For 2000, social anxiety and anti-Black racial resentment are significant predictors of death penalty support (Table 2, Model 1). These results mirror those of Unnever and Cullen (2010), with the exception of a few covariates: urban and authoritarianism are non-significant in these models, while being male is significant.

Table 2 Logistic estimates for support for the death penalty, all 3 original models
Full size table

For 2014, racial resentment is the only theoretical factor that predicts support for the death penalty when the three original perspectives are considered together (Table 2, Model 2). This may indicate increasing racialization of support for the death penalty (Unnever and Cullen 2012). It is important to note that social anxiety is non-significant in 2014 despite being significant in 2000, but follow-up analyses (not shown) of the interactions between each variable and a dummy variable representing the year indicate the only coefficient that is significantly different between the 2000 and 2014 models is conservatism. This suggests that the underlying sources of support for capital punishment were notably stable between 2000 and 2014.

When considering general punitive views (i.e., the belief that courts are not harsh enough), our findings show that, as was the case with ANES data, concern about rising crime and racial resentment are significant predictors in 2000 (Table 3, Model 1). In addition, having estimated partial proportional odds models, we find that higher levels of distrust in the Supreme Court is associated with a lower likelihood of being in the two more punitive categories compared with the least punitive category. Social anxiety is associated with a higher likelihood of being in the most punitive category compared with the two less punitive categories. Future research may be needed to consider the dynamics of distrust in courts and social anxiety in explaining views on punishment among the least and most punitive Americans.Footnote 20

Table 3 Generalized ordered logit estimates for punitive view, all 3 original models
Full size table

In 2014, social anxiety and racial resentment were both significantly and positively predictive of punitive views related to the severity of courts across all levels of the dependent variable (Table 3, Model 2). For the unconstrained variables, concern about crime consistently predicted increased punitivity. In particular, the positive coefficients indicate that in 2014 people who expressed more concern about crime were more likely to be in a more punitive category across all categories. Conversely, coefficients for being Black and being male were not consistent across levels of the dependent variable. That is, being Black was associated with a lower likelihood of being in one of the two more punitive categories, while being male was only associated with a lower likelihood of being in the most punitive category. In comparison to the 2000 model, the 2014 model suggests that while crime concerns and racial resentment remained important predictors of punitive views, distrust of courts became less important, while social anxiety became more consistently important.Footnote 21

Animus Toward the Poor

We next estimated models to test the Wacquantian idea that neo-liberal animus towards the poor and rejection of welfarist ideals predict punitivity (Tables 4, 5). For both 2000 and 2014, and for each of our dependent variables, animus toward the poor was a significant and positive predictor of punitiveness, controlling for covariates. Follow-up analyses (not shown) indicated that the difference between the 2000 and 2014 coefficients for animus toward the poor was non-significant in the models for support of the death penalty. Further, the coefficients for animus toward the poor are notably similar for the 2000 and 2014 punitive view models (1.63 vs. 1.59, respectively). We also note that race (being Black) was our only unconstrained variable for the punitive view models incorporating animus toward the poor. Coefficients for that variable in both 2000 and 2014 indicate that being Black was associated with a lower likelihood of being in the two more punitive categories compared with the least punitive category.Footnote 22

Table 4 Logistic estimates for support for the death penalty, animus toward the poor models
Full size table
Table 5 Generalized ordered logit estimates for punitive view, animus toward the poor
Full size table

Original Models and Animus Toward the Poor

Tables 6 and 7 present the findings from our final set of models. In these models, we include all four perspectives in order to evaluate which among them best explain the social foundations of punitiveness, controlling for the others. The results of these models show that anti-Black racial resentment and animus toward the poor are the most powerful and consistent predictors of punitiveness.

Table 6 All 3 original theories and animus toward the poor
Full size table
Table 7 Generalized Ordered Logit Estimates for Punitive View, All 3 Original Models and Animus Toward the Poor
Full size table

Considering support for the death penalty, social anxiety, racial resentment, and animus toward the poor were all significant in 2000 and 2014 (Table 6). It is notable that follow-up analyses (not shown) indicated no differences in the predictive power of those variables between the years we considered.Footnote 23 Overall, we find strong evidence for the sustained power of social anxiety, racial resentment against Blacks, and animus toward the poor in shaping support for the death penalty in the twenty-first century. To further evaluate these relationships, we conducted follow-up analyses to examine the salience of the theoretical predictors across all biennial GSS surveys between 2000 and 2014. Figure 1 depicts the supplemental analyses predicting support for the death penalty. Covariates (not shown) were included in all models. As Fig. 1 indicates, while the other theory-derived measures shifted in direction from 2000 to 2014, racial resentment and poverty animus were consistently predictive of punitiveness.

Fig. 1
figure 1

Support for the death penalty, all 3 original models and animus toward the poor

Full size image

For our measure of more general punitive views, racial resentment and animus toward the poor were each positive and significant in both years (Table 7). The effects of those variables also did not vary significantly across levels of punitiveness. The effects of concern about increasing crime were consistent across levels of the dependent variable in 2000 but were varied in 2014. In models for both years, however, coefficients for crime concerns were significant and indicated that increased concern was associated with increased punitivity. Additionally, we find that distrust of courts and social anxiety varied in their effects across levels of the dependent variable in 2000. In particular, greater distrust of courts was associated with a lower likelihood of being in the two more punitive categories compared with the least punitive category. More social anxiety was associated with a greater likelihood of being in the most punitive category compared with the two less punitive groups. Social anxiety was a constrained and significant factor in 2014, and distrust of courts was unconstrained but not significant.

Overall, these findings suggest that the belief that courts are not harsh enough has a slightly wider social basis than does support for the death penalty, despite some changes between 2000 and 2014. These findings also confirm the importance of anti-Black racial resentment (but not anti-Black racial stereotypes) in shaping punitive views, and suggest that animus toward the poor is an important independent contributor to American punitiveness in both 2000 and 2014. Indeed, supplemental analyses (see Figs. 2, 3) demonstrate that animus toward the poor is a consistent predictor of both death penalty support and general punitiveness for Americans in the twenty-first century.Footnote 24

Fig. 2
figure 2

Support for the death penalty, animus toward the poor model

Full size image
Fig. 3
figure 3

Punitive view, animus toward the poor model

Full size image

Supplemental Analyses of Racial Stereotypes

The failure of racial stereotypes to significantly predict punitiveness in this research (and in Unnever and Cullen’s 2010 analysis) warrants additional consideration. Given a broad body of literature on the importance of anti-Black racial stereotypes in typifications of criminal offenders and responses to them (Chiricos et al. 2004; Drakulich 2015; Unnever and Cullen 2012), it is important to consider possible mediating effects of our measures of stereotypes on punitiveness via racial resentment and animus toward the poor. In supplemental analyses (not shown), we examined these possible effects for both of our dependent variables. For the 2000 models, results of Sobel–Goodman mediation tests and individual model runs suggest that there were mediation effects between racial stereotypes and our two dependent variables via both racial resentment and animus toward the poor. As such, racial stereotypes are positively associated with both racial resentment and animus toward the poor, and both of these are in turn positively associated with support for the death penalty and general punitiveness.

For the 2014 models, when considering animus towards the poor as a mediator, there were no significant direct or total effects between racial stereotypes and either dependent variable. However, we did find mediation, or significant indirect effects, for racial stereotypes in the death penalty model. As such, racial stereotypes are positively associated with racial resentment, which in turn is positively associated with support for the death penalty. In the 2014 general punitiveness model, there was a significant mediating relationship for racial stereotypes, with racial resentment mediating about a third of the total effect between racial stereotypes and the belief that courts are not harsh enough.

These findings suggests that explicit, anti-Black views do shape punitiveness, but that they do so primarily through the expression of more symbolic and veiled anti-Black sentiment. This finding is significant in that it points to the persistent, if sometimes obscured, power of anti-Black racial stereotypes in shaping American punitiveness despite the tendency in research to fail to find direct effects.

Discussion

This study contributes to research on punitive American views in the twenty-first century by examining the explanatory utility of four models for the social determinants of those views. Three of the perspectives considered here were operationalized and evaluated by Unnever and Cullen (2010) using ANES data for the year 2000. Our goals in the present research were to replicate their models using an alternate data source, to extend their analysis to consider the explanatory power of animus toward the poor controlling for other factors, and to consider the social foundations of punitive views in 2014. We focus here on the findings in Tables 6 and 7, which shed light on the explanatory usefulness of each perspective, controlling for the others.

Overall, we have replicated the core findings from Unnever and Cullen’s (2010) important theory-testing contribution using GSS data. In addition, we found significant indirect effects of anti-Black racial stereotypes on punitiveness via both anti-Black racial resentment and animus toward the poor in 2000. In order to extend Unnever and Cullen’s (2010) analyses, we first examined data for 2014. In short, our results for 2014 suggest that racial resentment is the only significant predictor of support for the death penalty when considering the three original theories. These findings suggest that while overall support for the death penalty has slightly decreased nationwide (from 69 % in 2000 to 65 % in 2014 according to GSS data), for those who continue to support it, race matters. Our findings for 2014 indicate a relatively wide social foundation for general punitiveness: social anxiety, racial resentment, and concern about increasing crime rates predicted the belief that courts are not harsh enough. We also found a significant indirect effect for racial stereotypes via racial resentment in predicting support for the death penalty and a mediating effect via racial resentment in predicting general punitiveness.

In our extension, we found that in both 2000 and 2014 animus toward the poor was a significant predictor of each of our dependent variables, controlling for other factors. When added to a final model along with the three original theories, animus toward the poor continued to be a significant contributor to punitiveness in both years. Indeed, of all the factors examined, only racial resentment and animus toward the poor were significant across all models.

Given prior evidence that both welfare and crime control beliefs are racially-charged (Garland 2001; Gilens 1999; Pickett and Chiricos 2012), it is not surprising that punitive views are inherently linked to both anti-Black racial resentment and animus toward the poor. While the tendency to “otherize” racial minorities (Loury 2008), failure of “empathic identification” (Unnever and Cullen 2009), and inegalitarianism (Unnever et al. 2008a) have been identified as contributors to support for harsh punishment, we find that overt animus toward the poor predicts punitiveness, even when controlling for other key factors. Our findings thus not only affirm that racism should be at the center of our understanding of punitive American views (Unnever and Cullen 2010, p. 121), but that animus toward the poor must also be at the heart of models of American punitiveness.

A particularly important contribution of these findings is that both racial resentment and animus toward the poor are powerful predictors of punitive views, despite recent decreases in racial intolerance and the overall prevalence of punitive views (Ramirez 2013; Unnever and Cullen 2012, p. 537). These findings lend support to prior research which has found that racial resentment and other forms of symbolic racism are notably stable contributors to punitive views (Unnever and Cullen 2012) and point to the continued salience of race and poverty in conceptions of the threatening “other” (Garland 2001, p. 196; Loury 2008).

The present research tackled some of the acknowledged limitations of Unnever and Cullen’s (2010) previous work, but there are certainly opportunities for further expansion. By considering data for 2000 and 2014, we were able to view broad developments in American punitiveness. It would be useful, however, for future research to consider additional sources of contemporary punitive views. In particular, given the rise of “crimmigration,” an approach to social control in which criminal justice and immigration policies have become fused, and recent evidence suggesting that anti-Latino sentiment is linked with support for aggressive policing, future research should consider the power of anti-Latino animus as a contributor to twenty-first century American punitive views (Pickett 2015).

In addition, future research might investigate the social foundations of public opinion across segments of the American public. Certainly, public opinion research suggests that some segments among the population (e.g., those who are politically active and those “at the margins”) are critical to the success or failure of particular policy approaches (Erikson et al. 2002; Stimson 2004). Additional analysis of these dynamics may clarify the implications that stasis or movement in the determinants of punitive views might have for contemporary policy reform efforts.

In conclusion, this analysis replicated Unnever and Cullen’s (2010) findings on the social sources of punitive views in 2000 by using GSS data. In particular, we find that anti-Black racial resentment is a critical and consistent foundation of punitiveness, while concern about increasing crime and social anxiety are less consistent social sources of punitiveness. In addition, we find that the social sources of punitive views have not shifted fundamentally in the last 15 years. By operationalizing animus toward the poor and including it in our models, we find that both racial resentment and animus toward the poor have been and remain powerful predictors of both measures of punitivity, controlling for other perspectives. These findings affirm the importance of putting anti-Black racial resentment at the core of models of punitivity, and further affirm that animus toward the poor is a powerful social foundation of contemporary American punitiveness.