A Drop in the Ocean: How Priors Anchor Attitudes Toward the American Carceral State

Abstract

That black and white Americans disagree about the carceral state is well established; why this is the case is much less clear. Drawing on group hierarchy theory and the state's role in perpetuating group subordination/domination, we theorize that differences in socialization and contact during emergent adulthood produce divergent priors for racial groups and gender subgroups within race. These different starting points shape how people integrate new information from recent contact into their belief systems. Using a survey of over 11,000 respondents, we find that, instead of all groups integrating information the same way, recent direct contact contributes most to negative attitudes among groups whose contact with government agents is least negatively valenced. While interactions with the American carceral state divide opinions considerably among white Americans and women, adulthood contact for black Americans, especially black men, appears to be but ‘a drop in the ocean’ of political life.

What do Americans know about politics? How do they learn it? Many scholars have argued that, although most US residents can produce few facts and figures about politics, they possess relevant evaluative knowledge gained through day-to-day experiences and interactions with state agents (Cramer and Toff 2017; Lipsky 2010; Weaver, Prowse, and Piston 2019). When someone interacts with a welfare caseworker, for instance, they learn about the responsiveness and quality of government services and their value to the state (Schneider and Ingram 1993; Soss 2002). With individual experiences central to this political learning model, the theory supposes that if people have similar direct experiences with the government, they will develop similar beliefs about its quality (see, for example, Slocum and Wiley 2018; Tyler 2004; Tyler, Fagan, and Geller 2014).

Instead, we argue that groups in the US have unique histories and relationships with institutions, which produce divergent foundations for integrating new information in the form of direct experiences into political views. We demonstrate this process with respect to attitudes about the US carceral state, an intertwined set of state-sponsored institutions with the discretion to punish (Lerman and Weaver 2016). Research indicates childhood legal socialization and carceral state contact during emergent adulthood varies by racial and gender group membership (Brunson 2007; Jones 2014; Lesane-Brown 2006; Stevenson and Arrington 2009; Tyler 2004; Western 2006). The strength and valence of initial attitudes about the carceral state are thus likely to be distinct for these subpopulations. We argue that, as a result, even when Americans have the same punitive interactions with government, attitudes may update in divergent ways based on group membership. Given an experience of similar quality, we anticipate groups with stronger initial beliefs will update their attitudes less. We formalize this logic using a Bayesian learning model (Bartels 2002; Clinton and Grissom 2015; Zechman 1979), which outlines a set of competing hypotheses about the relationship between direct experiences and opinions towards the carceral state.

Our theory anticipates that recent contact with police, courts, and the American prison system will contribute more to attitudes about the quality of the criminal justice system for white Americans compared to black Americans and, within racial groups, women compared to men. Data from an original survey of over 11,000 black and white respondents confirm this finding: recent contact with the American carceral state strongly predicts the attitudes of white Americans and women, producing considerable differences in these groups. However, adulthood contact for many black Americans, and especially black men, is simply ‘a drop in the ocean’ of political life. Examining contact quality produces a similar result – group membership conditions responses to contact of shared perceived quality.

Our findings confirm that people can learn about government – its role, responsiveness, and reform needs – via contact. But some groups carry deep knowledge reservoirs about specific aspects of the political system (Jefferson, Neuner, and Pasek 2021; Weaver, Prowse, and Piston 2019). These reservoirs, our results show, anchor public opinion and attenuate the link between relevant experiences and attitudes. Determining how groups, on average, integrate both attitudinally consistent and inconsistent information into their belief systems helps explain large public opinion gaps across the US electorate and clarifies the avenues for reform and reinstating trust in government.

Information, Attitudes, and Direct Contact with State Agents

In a democracy, information about government is a normative good, identified by many as the bedrock of accountability and participation (Delli Carpini and Keeter 1996; Pérez 2015). Yet, critics have long doubted whether the public meets theorized informational thresholds and whether the average citizen can rationally integrate new information into existing beliefs (Achen and Bartels 2016). One rebuttal proposes that individuals learn about government quality and policies from interactions with state institutions and agents, especially street-level bureaucrats (Lipsky 2010; Soss 2002; Tyler, Fagan, and Geller 2014; Weaver, Prowse, and Piston 2019). Through these experiences, people can project what they learn about one arm of government onto global assessments of the state and efficacy of action. This theoretical framework assumes that individuals' experiences with the state – for better or worse – shift attitudes in predictable and parallel ways regardless of individual characteristics (Slocum and Wiley 2018; Tyler, Fagan, and Geller 2014).

But the value of a new experience to one's attitudes may depend on the information individuals already hold about the state – information that varies systematically across types of people in a society rife with race and gender hierarchies. Many argue that the US is a hierarchical system with group membership along the dimensions of race, gender, and their intersection determining access to power (Sidanius and Pratto 1999). Group membership constrains opportunities and assigns roles, leading to group-level outcomes and motivations (Maltby 2017; Pérez and Vicuña, 2023; Sidanius and Pratto 1999). These categories too provide a ‘prism’ through which different kinds of Americans experience and interpret the world (Bracic, Israel-Trummel, and Shortle 2019; Branton, Carey, and Martinez-Ebers 2021; Burch 2022; Carter and Pérez 2016; Hagan, Shedd, and Payne 2005; Israel-Trummel and Streeter 2022; Masuoka and Junn 2013).

As a result, group membership shapes the introduction of individuals into the political world and their formative political experiences, including with respect to the carceral state. The US criminal justice system affects some groups at systematically higher rates than others, with black Americans in general, and black men in particular, interacting with criminal justice agents disproportionately compared to other groups (Baumgartner, Epp, and Shoub 2018; Forman 2012; Western 2006). Observing this, some have argued that the carceral state is designed to purposely segregate, subjugate, punish, and control along the dimension of race, serving as a ‘new Jim Crow’ (Alexander 2012; Sidanius and Pratto 1999; Wacquant 2001). These patterns are often cited to explain group-related differences in political attitudes toward the carceral state, where black Americans systematically evaluate criminal justice institutions more negatively than white Americans (Peffley and Hurwitz 2010). But there is reason to believe the relationship between hierarchy and punishment in the US means attitudes form much earlier, before adult experiences.

Socialization around criminal justice institutions begins early in life; police officers are one of the first governmental agents children learn about (see, for example, Greenberg 1970), but evidence shows that black and white children – and within racial groups, girls compared to boys – are socialized differently with respect to punitive institutions. Black parents typically make deliberate efforts to teach their children about the dangers of police officers and discrimination in the US legal system (Taylor et al. 1990; Thomas and Blackmon 2015; Thomas and Speight 1999). This is especially true for black sons, who receive more messages focused on stereotypes and structural barriers than their female counterparts (Lesane-Brown 2006; Stevenson et al. 2005). In contrast, many white Americans see the carceral state as protecting and serving people like them, messages conveyed implicitly and explicitly to their children (Abaied and Perry 2021; Anoll, Engelhardt, and Israel-Trummel 2022; Sidanius and Pratto 1999).

Alongside differences in childhood racial socialization, race and gender subgroups also experience different levels of invasive encounters with the carceral state during their youth (Crutchfield et al. 2012; Geller and Fagan 2019; Kerley et al. 2004). Contact with the police and incarceration are disproportionately concentrated in adolescence and early adult years (Baumgartner, Epp, and Shoub 2018; Fagan et al. 2010; Western 2006). Evidence suggests black boys, and men in general, experience more invasive and discretionary forms of contact compared to especially white women (Fagan et al. 2010; Western 2006). Given that the carceral state treats individuals categorically (Lerman and Weaver 2016), people will likely understand these experiences through their group memberships. Members of subordinate groups typically reason their experiences occur because of their group memberships, while those in dominant groups tend to view them as occurring despite these ties (Pérez and Vicuña, 2023). These interpretations can unite otherwise heterogeneous subordinate groups around a motivation to improve their station by reforming offending institutions (Dawson 1994; Pérez 2021).

The combination of divergent childhood racial socialization and experiences with the carceral state during emergent adulthood means that people will likely arrive into their adult political lives with distinct informational baselines connected with typical experiences for their race and gender. These hierarchically-produced baselines shape the value of new information in the form of contact during adulthood. For many black adults, recent direct contact likely provides little new information about the nature of the carceral state and its agents (Hagan, Shedd, and Payne 2005; Justice and Meares 2014; Slocum and Wiley 2018). Rather, their past experiences and socialization lead to well-informed expectations about interactions, which more contact only confirms. In this ‘experience of the expected’, state contact simply reaffirms rather than updates existing beliefs (Dennison and Finkeldey 2021; Hagan, Shedd, and Payne 2005; Slocum and Wiley 2018).¹

In contrast, whites generally have more positive opinions about the carceral state based on less information. This follows from legal enforcement typically benefitting their group and a comparative dearth of firsthand experiences (Alexander 2012; Sidanius and Pratto 1999). White Americans may find direct contact more informative compared to black Americans. Direct experiences may also increase attitude certainty less for whites than blacks because whites see contact as occurring despite their group membership (Pérez and Vicuña, 2023). Lacking a basis for linking experiences to policy change, whites may experience uncertainty about the implications of these experiences for carceral state judgements (Alvarez and Brehm 2002).

Gender differences are also likely to emerge. Black/white boys are more likely to have direct contact with law enforcement during adolescence than black/white girls (Baumgartner et al. 2017; Western 2006), and evidence suggests racial socialization varies too by gender (Lesane-Brown 2006). Given divergent formative experiences, our theory suggests that differently gendered groups likely carry unique prior knowledge that shapes how they interpret new experiences. But the size of gender differences may pale in comparison to those across racial cohorts, even when considering the intersectional experiences discussed above (Crenshaw 1989; Hancock 2007). Public opinion scholars consistently report gender differences in attitudes (for example, Hyde 2005), but these gaps are often small compared to racial differences. Some suggest dynamics of group-based segregation explain this (for example, Sidanius and Pratto 1999). Despite vast differences in lived experience across gender lines, men and women frequently interact with each other. Proximal learning and shared social norms can thus develop, unlike across racial groups, which remain highly segregated at the geographic, institutional, and social levels (Anoll 2022).

A Bayesian Approach to Attitude Differences

We have argued that groups, on average, start with baseline information about the carceral state that varies in both its valence and strength. Some groups have strong, negative attitudes from accumulating many pieces of information over time. Other groups have weak, positive attitudes based on fewer experiences with the carceral state and less explicit socialization. Because of these different starting points, groups will integrate new information in the form of direct contact differently.

This logic linking features of baseline attitudes to the integration of later information nicely matches theoretical assumptions in Bayesian models of public opinion updating (Bartels 2002; Clinton and Grissom 2015; Zechman 1979). We use this literature to clarify how changing the parameters of starting information and the nature of later information relates to subsequent attitude differences, and when different subgroups' attitudes will converge or diverge. Thus, this framework offers rich expectations about the link between group-based experiences, expectations, and attitudes. Moreover, it allows us to identify more precisely potential explanations for group-based attitude differences across differently-positioned groups, uniting work in racial and ethnic politics and political psychology (Pérez 2021).

Bayesian learning models assume people have existing experiences and information, which form baseline attitudes, or priors. They then combine new information with these priors into updated beliefs – posterior attitudes – containing both previous information and the new signal. Within this framework, new experiences matter but are conditioned by starting attitudes, suggesting recent contact may not fully explain group differences.

We assume that because hierarchies produce patterns in legal socialization and contact during emergent adulthood, racial and gender subgroups' baseline attitudes:

• • Vary on average in their valence (μ_{t −1}). Upon entering adulthood, people's carceral state evaluations can vary from positive to negative.

• • Vary on average in their precision, or strength (π_{t −1}). The larger π_{t −1} is, the more confidently individuals hold their existing views. We assume variation here comes from the amount of baseline information people have about the carceral state – the more socializing messages and contact one has, the stronger their priors.

People adjust these initial beliefs as they move through adulthood. Whenever people encounter information about the carceral state (x_t), we argue that they add this new information to their existing views. Like individuals' prior beliefs, new information ranges from positive to negative.² Information can also vary in ‘dosage,' or strength (φ_t). While information may come from any source, we focus on recent, direct experiences with the carceral state. An individual's posterior attitude about the carceral state (μ_t) is thus a precision-weighted average of initial opinions and subsequent experiences:

(1)$$\mu _t = x_t\displaystyle{{\varphi _t} \over {\pi _{t-1} + \varphi _t}} + \mu _{t-1}\displaystyle{{\pi _{t-1}} \over {\pi _{t-1} + \varphi _t}}$$

(2)$$\pi _t = \varphi _t + \pi _{t-1}$$

By showing mathematically how attitudes combine initial opinions (μ_{t −1}) and recent experiences (x_t), equation 1 identifies two ways attitudes update. First, attitudes observably update if new information (x_t) departs markedly from one's priors (μ_t−1). All else being equal, the larger this discrepancy, the larger the difference in final opinions.³ Second, attitudes will differ more with recent contact as the relative strength of the new experience increases (φ_t/(π _t−1 +  φ_t)). People will weigh the new information more heavily if they give little weight to their initial attitudes. The relationship between experiences and opinions is thus a function of the discrepancy between new information and prior opinion, weighted by the relative strength of the new experience: μ _t = (x _t − μ _t−1) (φ_t/(π _t−1 +  φ_t)).

To clarify how this framework helps us explain observed attitude gaps and understand when attitudes might converge, Fig. 1 presents stylized opinion differences given multiple rounds of exposure to new information about the carceral state for two groups. Each figure changes different parameters of the model and assumes this sequence of new experiences is negative. We begin with Fig. 1a, which demonstrates the empirical possibility that new information has no effect (μ_t−μ_{t −1} = 0). Existing opinions may dominate new information, either because the new information comports with existing opinions (x_t = μ_{t −1}) or the strength of someone's initial attitude surpasses the strength of the new information (π_{t −1} > φ_t). In this world, recent contact does not explain or alter gaps between group attitudes; rather, prior attitudes, robust to new experiences, explain differences across groups. This is one empirical possibility.

Figure 1. Stylized patterns of attitude change given group-based differences in priors.

A second possibility is a common effect. Imagine the two groups' average initial attitudes diverge, as in Fig. 1a, but now new information yields parallel attitude differences. This result – where intercepts diverge but the slope of the line is the same – may occur if groups have priors that differ in their valence but not their precision, and then experiences differ in their valence but depart to like degrees from these priors. If so, new information is combined with initial attitudes such that subsequent attitudes differ as much as initial opinions. Alternatively, parallel differences can occur if these groups experience new information differing in valence and also apply a different weight relative to their initial attitudes. In these situations, opinions respond thermostatically – as Fig. 1b illustrates.

These first two figures show parallel effects across groups, but the alternative, depicted in Figs 1c and 1d, is an interaction effect. Figure 1c typifies a relationship that results in divergence. For this pattern to emerge, groups with more negative starting attitudes would need to confront new information that was qualitatively much more negative in valence. Or, they need to place more weight on new experiences that are only somewhat more negative, possibly reflecting weak initial attitudes. A motivated reasoning hypothesis is also consistent with the polarization exhibited in Fig. 1c (Jefferson, Neuner, and Pasek 2021). Racial group membership may result in different types of people viewing the same information through a different lens (Bisgaard 2019; McGowen and Wylie 2020; Peffley and Hurwitz 2010). If people interpret experiences to affirm existing beliefs, as motivated reasoning suggests, then new information, regardless of its nature, will polarize opinions. This will lead groups with initial negative opinions to view experiences more negatively and those with positive initial opinions to downplay negative information.⁴

In contrast, Fig. 1d typifies an interaction effect producing convergence. Here, we observe little attitudinal change among those who hold strong, negative priors even with each new piece of negative information. However, groups with more positive and potentially weaker starting attitudes exhibit large opinion differences with increasing contact. This outcome aligns with our proposed theory and suggests that new negative experiences are, for some groups, simply the ‘experience of the expected’. For these individuals, attitudes differ little, even as recent experiences increase. Similar to Fig. 1c, we observe an interaction between new experiences and group attitudes, but here the group with more positive priors shows the largest attitude updating. Consequently, group attitudes will converge over time.

We expect that because black Americans' priors – especially black men's – are typically negatively valenced and strong, recent experiences with the carceral state will relate to only modest attitude differences for members in these groups compared to white Americans and women. If so, attitude convergence would occur between white Americans with more recent contact and black Americans irrespective of recent experiences. We also expect that even with similar reported contact quality, black Americans' attitudes should change less with experience than white Americans, and men should see smaller differences than women. Rather than producing parallel attitude differences, comparable experiences with government agents and institutions are hypothesized to produce different posterior attitudes.

The Relationship Between Carceral State Contact and Attitudes Varies Across Race and Gender Subgroups

We test our expectations with data from the Race and Carceral State Survey (RCSS). The RCSS is an original online study fielded through Survey Sampling International (now Dynata) in spring 2017 to a nationally diverse sample comprised of 3,073 black and 8,093 white American adults (Anoll and Israel-Trummel 2017).⁵ This large sample size, along with the study's multiple measures of carceral state attitudes and experiences, allows us to look at variations within and across racial groups and by the quality of contact with state agents.⁶

As a first test, we consider the number of recent direct contacts individuals have had with various parts of the carceral state, including the police, courts, and prison or jail. Respondents reported whether, in the last five years, they had been: stopped by the police, and how often (0, 1–2, 3–4, or 5+); summoned to appear in court, and how often (0, 1–2, 3–4, or 5+); and whether they had spent any time in prison or jail (No/Yes).⁷ These questions capture possible direct interactions with three different criminal justice institutions in recent years. First, we score these items to run 0–1 and then add them into an index. We scale this index from 0–1, with 0 indicating no recent direct experiences and 1 representing the max across all items. We refer to this as the direct contact index.

Figure 2 shows the measure's distribution across racial groups. In line with previous work (Lerman and Weaver 2014; Mondak et al. 2017; Walker 2020), black Americans report more experiences with at least one facet of the carceral state in the last five years compared to white Americans (0.15 v. 0.10). Still, a majority of white respondents (57 per cent) and a near majority of black respondents (46 per cent) reported no recent direct experiences.

Figure 2. Frequency of contact with the carceral state. Vertical lines indicate group means.

We also find large variation in contact within racial groups by gender (see also Western 2006). Black men have a uniquely large number of direct contacts with the carceral state (mean = 0.21), roughly 2–3 contacts in recent years. Black women and white men have similar recent contact rates (mean = 0.11). While white women have the fewest recent contacts, the mean is only 0.02 units away from black women and white men, making their experiences much less unique from the other two groups compared to those of black men.

We examine the relationship between the number of recent direct experiences with the carceral state as our independent variable and carceral state attitudes as our dependent variable. We operationalize our outcome variable using attitudinal measures about (1) the carceral state in its current form, (2) support for reform efforts, and (3) support for related movements, combining nine items described in Table 1. We sum these items and then rescale the measure to run 0–1, where higher values signal negative views of current institutions and support for reform (Revelle's ω total: Full = 0.85, Whites = 0.84, Blacks = 0.78).⁸

Table 1. Dependent variable items

^a Participants were randomly assigned to one of two prompts for level of attention. Those in the US condition were uniformly more critical of the police than those in the local area condition. Thus, we retain all responses because our interest is in differences in correlates across groups, not levels of construct endorsement.

Using linear regression, we estimate the relationship between our direct experiences measure and carceral state attitudes. To test if the relationship differs by racial group membership, we interact our direct experiences measure with an indicator for if a respondent identifies as black (Lerman and Weaver 2014; Peffley and Hurwitz 2010).

We note that this cross-sectional approach is not itself Bayesian, which would require testing individual-level change over time. Rather, the Bayesian framework offers testable implications that anticipate a certain relationship between attitudes and experiences at the level of groups. If groups’ prior attitudes differ in strength and valence, then the more informed group should exhibit greater attitudinal similarity, even as direct experiences increase compared to the less informed group. As a result, our interaction term should be negative and statistically significant, indicating a difference in slopes that produces patterns like those in Fig. 1d. Further, we should observe a significant difference at the intercept, with black Americans with no recent experiences exhibiting attitudes that are more negative than white Americans.

We further note that this approach captures average differences across groups. While some may view group membership as a blunt way to capture socializing experiences, we conceptualize it as a measurement of constraint on the development of attitudes across types of people in a society constrained by hierarchical roles and opportunities (Sidanius and Pratto 1999). Variation clearly exists across experiences and characteristics within racial and gender subgroups (Beltrán 2010; Hochschild, Weaver, and Burch 2012), yet patterns by group emerge precisely because the politics of a society forces certain experiences upon individuals because they are part of a group. As a result, considering variation by group is central to understanding the logic of our Bayesian story.

We report the model results visually in Fig. 3 and provide the parameter estimates in the online appendix. The figure shows predicted levels of reform-oriented attitudes as direct experiences increase separately by racial group. In addition, we plot 84 per cent confidence bands for the predictions to show visually where differences across groups are statistically significant at the 0.05-level, more appropriately capturing our between-groups hypothesis test (Schenker and Gentleman 2001).

Figure 3. Predicted effect of direct experiences on carceral state attitudes with 84 per cent confidence bands.

Rather than parallel slopes, group membership conditions the relevance of recent direct experiences. Figure 3a shows that recent direct experiences more strongly relate to the opinions of whites than blacks. A min-max change in recent experiences associates with attitudes about the carceral state that are, on average, 0.27 points more anti-status quo for whites (p < 0.001) and only 0.13 points for blacks (p < 0.001). This 0.14 point difference is reliable (p < 0.001) and is substantively large at over a 100 per cent increase. Likewise, we observe a substantial baseline difference in opinions. Among respondents with no recent experiences, black Americans are on average over 0.20 points more anti-status quo in their opinions compared to white Americans. The figure mirrors the predicted outcomes produced in Fig. 1d, both with respect to intercept differences and the convergence of attitudes across groups.

Figure 3b complements these insights. Here we interact an indicator for whether someone identifies as a man with the direct contact index and look separately at our black and white respondents given the possibility of intersectional effects (Crenshaw 1989; Hancock 2007). We find that the relationship between recent direct experiences and carceral state attitudes is 79 per cent stronger for black women compared to black men (p = 0.002) and 19 per cent stronger for white women compared to white men (p = 0.066). But, as expected, the size of these effects by gender within racial groups are smaller than across racial groups.

We next control for factors potentially associated with group membership, recent direct experiences, and carceral state opinions. Doing so allows us to account for potentially consequential intra-group opinion variability. This offers a more stringent assessment of the difference in relative influence for recent direct experiences across groups because many of these variables are conceptually post-treatment from either group membership or recent direct experiences. Specifically, we control for:

• • Demographic variables: education, income, age, and gender, which vary alongside contact rates (Western 2006).

• • Indirect carceral contact: namely, whether a respondent has peers with felony convictions, which has been shown to shape attitudes and may vary alongside direct contact (Anoll and Israel-Trummel 2019; Mondak et al. 2017; Walker 2020).

• • Partisanship: given the elevation of criminal justice concerns within the political debate, partisanship may covary with contact. We measure it using a seven-point scale.

• • Racial resentment: we use the standard 4-item measure known to structure both black and white attitudes about the carceral state by capturing explanations for black Americans' status (Jefferson, Neuner, and Pasek 2021; Kam and Burge 2019). Low levels of racial resentment correspond with structural explanations for racial inequality and high levels align with individualistic views (Kam and Burge 2018).

All variables run 0–1 or enter as indicators. We interact all predictors with an indicator for whether or not a respondent identifies as black or a man to allow for flexible relationships (Masuoka and Junn 2013). Given our fully interactive specifications and to facilitate again comparisons to our stylized attitude change patterns, we present the results visually in Figs 4 and 5. We account for group differences in all covariates by setting them to group-specific means or modes (Kinder and Winter 2001).

Figure 4. Predicted effect of direct experiences on carceral state attitudes with 84 per cent confidence bands.

Figure 5. Predicted effects of direct experiences on carceral state attitudes by a group with 84 per cent confidence bands.

The relationship in Fig. 4 again supports our contention that group-specific priors anchor carceral state opinions. Even after controlling for several variables potentially post-treatment to race, capturing some of the variation attributable to it (Sen and Wasow 2016), we again find a consequential racial divide in opinion about the carceral state. Black Americans are almost 7 points more anti-status quo in their carceral state evaluations (p < 0.001), reflecting unique categorical experiences. Comparing sample average black and white Americans with no recent direct experiences, this gap expands to 18 points. Recent direct experiences still contribute to carceral state attitudes for both groups even after controlling for plausible correlates – but again, they contribute more to whites' attitudes than to blacks'.⁹ The association with carceral state attitudes is 5 points stronger for whites (p = 0.003), an increase of 44 per cent compared to the relationship among black Americans.

The combination of a stronger relationship between direct experiences and carceral state attitudes for white Americans and the more pervasive anti-status quo opinions among black Americans, even with no contact, means that as recent, direct contact increases, black and white Americans' attitudes begin to converge. Consistent with Fig. 1d, Fig. 4 shows that white Americans with high levels of recent, direct contact hold as anti-status quo attitudes as black Americans with no recent contact. Further, the difference in opinions between blacks and whites decreases by 28 per cent from 18 points to 13 points. This result is consistent with our Bayesian perspective on attitude change and inconsistent with directionally motivated reasoning (Fig. 1c). If motivated reasoning fully explains responses, we should observe divergence because people interpret their experiences differently.

Next, we compare the relationship between direct contact and carceral state attitudes by gender within racial group after including covariates. These results appear in Fig. 5. We find that direct experiences with the carceral state have stronger relationships with negative evaluations of the carceral status quo for women than men, regardless of racial group membership. For black Americans, direct experiences' marginal effect is 55 per cent larger for women than men (p = 0.046).¹⁰ Among white Americans, the difference is similar at 52 per cent (p = 0.001).¹¹

Collectively, our results reflect the attitude convergence pattern Fig. 1d depicts. Recent experiences relate less to the attitudes of black Americans compared to whites and men compared to women. Further, these divides are qualitatively larger by race than by gender (within race), both with and without controls, likely reflecting how experiences spill over across gender but not racial lines (Walker 2020; Walker and García-Castañon 2017). We have established so far that when individuals of different groups face increasing recent criminal justice contact, groups with less precise and more positive starting attitudes exhibit a greater degree of attitude variation compared to those with more precise and more negative attitudes. That is, the same change in the contact rate appears to produce differences in average updating based on group starting position. However, we know from past research that individuals in different race/gender subgroups report differing degrees of quality when they interact with the criminal justice system. We turn next to consider whether the reported quality of experiences also produces divergence in opinions across groups.

Contact Quality's Influence Differs Across Groups

The Bayesian framework expects potential attitude differences related to increasing amounts of new information in the form of recent contact. It also anticipates differences based on the nature of that information, holding constant how much new information people receive (x_t's valence). While we have found across-group variation related to the number of recent direct experiences, we consider potential variation by valence here.

Respondents from the RCSS who reported contact with the police in the preceding five years also reported whether they were treated unfairly (Yes/No) and if they felt endangered (Yes/No).¹² We combine these items into a 3-category measure of quality ranging from 0 (fair and safe) to 2 (unfair and unsafe). We label these categories ‘positive’, ‘mixed’, and ‘negative’. Fig. 6 shows this measure's distribution by racial subgroups – then by gender within race.

Figure 6. Nature of contact with the carceral state by race and gender.

Consistent with previous research (Peffley and Hurwitz 2010), black respondents, on average, report that their experiences with the police were more negatively valenced than white Americans. Of the 47 per cent of black respondents stopped by the police at least once in the last five years, 58 per cent reported their recent experiences were mixed or negative. Only 24 per cent of white Americans felt the same way – with only 38 per cent stopping at least once. This 34 percentage point gap in quality ratings is substantively large and statistically significant (p < 0.001).

Gender differences, too, emerge. Of those stopped by the police,¹³ men report more negative experiences than women on average, but gender differences are smaller than racial differences. Black men report the most negative experiences (38 per cent), followed by black women (25 per cent), white men (11 per cent), and, finally, white women (7 per cent).

Our Bayesian framework suggests that, because of hypothesized differences in average group priors, white people and, within racial groups, women, will show more attitude variation compared to black people and men as reported quality becomes more negative. This is because each new experience's valence departs markedly from the priors of the former groups. Our large sample is particularly well-positioned to enable this test, given that only about half of either white or black Americans report recent contact with the criminal justice system and further variation within contact in quality judgements.

We take multiple steps to make our comparison. Among those with any recent police contact, we use coarsened exact matching (CEM) to create comparable samples that differ only on self-reported contact quality (Iacus, King, and Porro 2012). CEM provides a preprocessing data approach, allowing us to focus on the strata of similar individuals by dropping dissimilar observations. This reduces the influence of outlier observations, an important need for our comparisons given the relative rarity of cases with negative judgements of the police. We thus recover more accurate estimates of differences in views of the carceral state across levels of contact quality than not using this procedure.

We match respondents based on the controls outlined in the prior section and add the frequency of police stops to capture contact frequency. We include the weights produced through the CEM procedure in a linear regression. The outcome variable is, again, carceral state attitudes, and the primary independent variable is the self-reported quality of recent police contact. We dummy out each contact quality category and use ‘positive’ experiences as the baseline. We also include the variables used in the matching procedure as covariates to account for remaining variation across individuals within strata. We interact all variables with an indicator if a respondent is black, then repeat this procedure within the racial group for our gender comparisons.

Figure 7 shows how group attitudes differ on average as reported quality changes, holding constant contact amount and covariates. Taking Fig. 7a first, we see that more negative police encounters are associated with more reform-oriented carceral state attitudes for black and white Americans on average. However, quality of contact produces larger attitude differences among white Americans than among black, consistent with a Bayesian explanation (mixed: p = 0.024; negative: p = 0.081).¹⁴ While we find evidence pointing to differences by gender among black men and black women, these results are statistically insignificant (mixed: p = 0.28; negative: p = 0.61). We find zero difference between white men and women (mixed: p = 0.68; negative: p = 0.38).¹⁵

Figure 7. Predicted effect of contact quality on carceral state attitudes. Eighty-four per cent confidence bands. Ns report the matched sample size for each experience quality category.

These results corroborate the Bayesian framework's prediction that group-based attitude differences can, in part, exist because of differences in priors. Not only do attitude differences connected with contact frequency vary across groups, but so too do differences linked with contact quality. These differences highlight how group-specific expectations condition the influence of recent information.

Discussion

It is well established that black Americans have more frequent and negative experiences with the carceral state than white Americans. Further, public opinion data continually confirms that black and white Americans diverge in their attitudes towards the carceral state. Therefore, intuition may suggest that more negative experiences with the carceral state among black Americans produces this opinion gap. However, this intuition is incomplete. Given black Americans’ childhood socialization, we theorize that recent experiences, even if disproportionately negative, are uninformative about the carceral state when they align with one's prior expectations. Using a Bayesian learning model as our framework, we demonstrate that recent experiences do not uniquely explain the Black-White divide. Indeed, our analyses of survey data show weaker relationships between recent experiences and carceral state attitudes for black than white Americans and men compared to women.

Our results contribute to a growing literature on the importance of racially distinct priors in public opinion formation. For example, Jefferson, Neuner, and Pasek (2021) report that black and white Americans' reactions to descriptions of a police officer's use of force are explained in part by different initial beliefs about police bias and racial attitudes. We extend this insight by highlighting how group-based priors may regulate the incorporation of new information, even when the information is varied in nature. Racially distinct priors contribute to group differences in opinion by affecting information interpretation and incorporation. This insight teaches us that racial differences in the amount and quality of contact during adulthood alone cannot explain the public opinion gap between black and white Americans regarding the carceral state. Nor does motivated reasoning, where blacks and whites interpret experiences to affirm existing attitudes, appear capable of fully explaining group differences in opinions (see also Jefferson, Neuner, and Pasek 2021). Given the role of group-based socialization experiences, we expect this framework may speak to other intra-group variation, including age, nativity, and time in the US (see, for example, Smith 2014; Wakefield and Wildeman 2013).

We have assumed a direct connection between direct experiences and carceral state attitudes. However, other attitudes may mediate these experiences. For instance, experiences could alter racial resentment given that it captures in part explanations for black Americans' status (Kam and Burge 2018). Negative experiences could foster greater belief in structural discrimination by providing relevant information, a possibility existing work implies (Engelhardt 2021). Similarly, negative experiences could instil group empathy (Sirin, Valentino, and Villalobos 2021). These experiences may help create this predisposition to ‘care about the perspectives and emotions of members of other groups’ (Sirin, Valentino, and Villalobos 2021, 24). Indeed, group empathy theory's originators suggest this possibility (Sirin, Valentino, and Villalobos 2021, 70). Results reported in Appendix 5 suggest modest mediation of direct experiences by racial resentment, which we interpret as worth future work investigating.

While we have focused on group-based differences in the link between experiences and attitude valence, the Bayesian framework also offers information on attitude certainty – the strength of people's beliefs. Even if a new experience provides no new information, attitude certainty can increase (cf. Alvarez and Brehm 2002). We report results in Appendix 6 addressing this possibility. We again find group-based differences. Direct experiences increase attitude certainty for black Americans – and more so for black women than men. For white Americans, direct experiences decrease certainty, especially for white men, inconsistent with a strictly Bayesian perspective but consistent with the introduction of ambivalence (Alvarez and Brehm 2002). White Americans begin with unique, positive beliefs and perspectives about the carceral state connected with their hierarchical position (Justice and Meares 2014; Sidanius and Pratto 1999). New information for this group introduces new considerations that make it difficult for them to choose among competing predispositions to judge the carceral state.

The Black-White opinion gap on criminal justice attitudes often leads scholars, pundits, and activists to ask what kinds of policy changes can restore trust in the carceral state among black Americans or, alternatively, induce white support for reform. Our findings have implications for both.

For black Americans, our theory and the framework provided by the Bayesian learning model suggest that decreasing the overall amount of direct contact between black Americans and agents of the carceral state is unlikely to change group-level beliefs – nor will neutral interactions substantially alter negativity. Instead, only explicitly positive interactions will likely change black Americans' average attitudes. Evidence for this point comes from a recent field experiment on community policing where a 10-minute interaction constructed to encourage positive intergroup contact improved residents' views of the police (Peyton, Sierra-Arévalo, and Rand 2019). Consistent with our theory, these positive interventions had the greatest effect among black respondents and others with initial negative views.

Conversely, Reny and Newman (2021) show that white Americans' views of the police changed more than black Americans' following the murder of George Floyd in May 2020. Like direct experiences, the video footage and ensuing protests perhaps offered a significant, negative signal about the police inconsistent with white Americans' priors but in line with black peoples' beliefs. Consistent with our theory, their results suggest that priors condition the influence of new information on attitude change. And yet, our findings show too that even with large doses of negative information obtained through direct experience, white Americans' opinions only begin to approach the attitudes of black Americans without recent contact.¹⁶

Our results raise an important question: considering that our Bayesian model and regressions predict convergence in black and white carceral state attitudes, why do we see large differences across these groups in the real world? The combination of lower contact incidence for white Americans, coupled with, on average, more positive experiences, perpetuates group-based attitude differences, even if equalizing experience rates and nature across groups would see convergence. On average, few white Americans have experiences with the criminal justice system and as a result, rather than occupying the right-hand side of our graphs – such as in Fig. 1d – they fall closer to the left. Average differences across groups emerge because strong priors anchor attitudes and dominate much of attitude expression.

Collectively, our results suggest that simple explanations focused on the nature and amount of contact during adulthood fall short of describing why black and white Americans differ so substantially in their views of the carceral state. We must consider how the racialized nature of American life changes people's information about different issues, the valence of their initial opinions, and the variation in their lived experiences. Our results reinforce the primacy of race in this process: although we observe an interaction effect where women's experience relates more to anti-status quo opinions than men's experience, the size of this effect is much smaller than between racial groups, consistent with previous work on the race/gender divide in public opinion (Bracic, Israel-Trummel, and Shortle 2019; Hyde 2005; Israel-Trummel and Streeter 2022). While people learn about politics through their experiences with governmental institutions, direct contact with street-level bureaucrats and officials does not shift attitudes uniformly. Rather, these experiences are integrated into ways that reflect larger dynamics of hierarchical power relations and the ‘prism of race’ that so centrally defines American life (Masuoka and Junn 2013; Pérez and Vicuña, 2023; Sidanius and Pratto 1999).