Associations between Racial and Socioeconomic Discrimination and Risk Behaviors among African-American Adolescents and Young Adults: A Latent Class Analysis

Abstract

Purpose.

Discrimination is a common stressor among African Americans and may increase vulnerability to risk behaviors, such as early initiation of substance use, substance use problems, and physical aggression; however, few studies have examined different types of discrimination and their associations with patterns of risk behaviors. This study examines the relationship between experiences of racial and socioeconomic discrimination and risk behaviors in African-American adolescents and young adults.

Methods.

We investigated associations of two discrimination types with risk behavior patterns identified with latent class analysis in a high-risk sample of African Americans (N= 797, M_age=17.9 years, 50.2% female).

Results.

Four distinct classes of risk behaviors were characterized by High Use & Aggression (10%), Moderate Use & Aggression (10%), High Alcohol (17%), and Low Use & Aggression (63%). Classes that exhibit general risk behaviors, including substance use and aggression, were significantly associated with racial and socioeconomic discrimination, even in the fully-adjusted model. Relative to other classes, the High Use & Aggression class demonstrated an elevated likelihood of experiencing both racial and socioeconomic discrimination.

Conclusions.

Findings support a link between racial and socioeconomic discrimination and risk behavior in African-American youth, which may be stronger for socioeconomic discrimination. Understanding the relationship between discrimination and risk behavior can inform future interventions to prevent substance misuse and conduct problems in youth. Further study is needed to elucidate the relationship between discrimination and other risk behaviors.

INTRODUCTION

Many studies suggest that discrimination increases vulnerability to health problems, exacerbating inequity [1, 2]. One meta-analysis found associations between racism and poor mental, physical, and general health [3]. Certain populations, such as African Americans, experience elevated levels of discrimination and may be more vulnerable to negative health outcomes [4]. Data suggest that African Americans experience more discrimination than Whites or other racial/ethnic groups [5, 6], including youth [7]. Studies suggest that >75% of African American adolescents experienced racial discrimination in the past three months [8], and >85% in the past year [7].

Discrimination may influence risky health behaviors in addition to health outcomes. In accordance with the stress-coping model, researchers hypothesize that individuals may engage in heavy substance use and other risk behaviors to cope with the stress of discrimination [9-11]. The model posits that discrimination increases negative affect which in turn leads to risk behaviors as coping mechanisms. While evidence testing specific components of this model is limited, a recent study in a sample of black college students lends support in its observation that elevations in negative affect provoked by discrimination, combined with drinking to cope with negative affect, were linked to more alcohol problems [12]. The literature has provided ample evidence of the associations between racism and many risk behaviors, including use of alcohol [13-15], cannabis [16, 17], tobacco [18], illicit drugs [19], and multiple drugs [10, 11, 20], as well as conduct disorder symptoms [21] and physical aggression [22, 23]. Physical aggression is especially relevant among young African Americans. In the 2017 Youth Risk Behavior Survey, African American adolescents were more likely to report engaging in physical fights in the past year than White or Hispanic youth [24]. In 2016, homicide was the leading cause of death in 12 to 21-year-old African Americans [25]. The high burden of violence evidences the need for understanding how different sociocultural experiences may increase vulnerability to aggression. Furthermore, discrimination and risk behavior are likely related to psychosocial factors, as previous studies found links between racial discrimination and depression [26, 27], suicidality [28, 29], and social anxiety [30, 31].

However, many studies focus on a single risk behavior, as in studies of discrimination and alcohol-related problems [14], stages of alcohol use [32], marijuana use [16], tobacco use [18], and violence/ gang involvement [22]. A potentially useful method for studying discrimination and multiple risk behaviors is latent class analysis (LCA). LCA, a form of data reduction, can be used to identify mutually exclusive behavioral patterns in a series of categorical items. The latent classes identified reflect probabilities of homogeneous groupings of behaviors that may be studied, as opposed to investigating each behavior individually. We chose to use LCA as opposed to aggregate scores because LCA reveals behavioral differences at the individual level that aggregate scoring may obscure. We also used LCA because we were interested in studying subtypes of alcohol and drug users which reflect aggressive and nonaggressive tendencies, given that this is a high-risk population.

LCA has previously been used in studies investigating substance involvement and discrimination experiences. One study used LCA to develop substance use profiles and their associations with a stress inventory, which included racial discrimination, in a sample of Hispanic adolescents [32]. Another study used latent profile analysis, a related strategy, to study discrimination profiles and mood/substance-use disorders in an adult sample of African Americans [33].

However, to our knowledge, no study has utilized LCA to study risk behavior patterns and their relationship with discrimination in African Americans. Moreover, to fully understand the relationship between discrimination experiences and risk behavior in African Americans, it is important to consider other sources of discrimination, such as socioeconomic status (SES), which is “part of the causal pathway by which race affects health” [34]. Socioeconomic discrimination can occur on an institutional level (residential segregation) [35] or an individual level (criminalizing individuals on welfare) [36]. African Americans are overrepresented among low-income families [35, 37, 38]. Although low SES cannot be equated to discrimination, this gap suggests that SES may be a significant source of discrimination in African Americans; however, there is a dearth of studies that consider discrimination other than racial discrimination. Among few that have looked at a range of discrimination sources, one study found that the association between socioeconomic discrimination and depressive symptoms among African American pregnant women was stronger than that for racial discrimination [39]; thus, socioeconomic discrimination merits investigation. A relevant concept is that of intersectionality, wherein individuals with multiple identities experience multiple forms of discrimination [40]. For example, the experiences of low-income African Americans may be the product of intersecting patterns of racism and classism. The present study attempts to address this gap by investigating racial and socioeconomic discrimination.

Individuals who experience discrimination may also experience psychological distress and cope by engaging in risk-seeking behavior. Although our data are cross-sectional and cannot provide evidence of causality, we hypothesized that there may be a close association between discrimination and risk behavior. The present study 1) utilizes LCA to identify risk behavior patterns in African American adolescents/young adults, specifically substance use and conduct problems, and 2) investigates associations between risk behavior classes and racial/socioeconomic discrimination. Given limited prior research on the effect of parental experiences of discrimination on offspring [21, 28, 41], we also consider the potential moderating role of maternal discrimination on offspring risk behavior.

METHODS

The data were from the Missouri Family Study (MOFAM), a high-risk family study of alcoholism that oversampled African-American families. Missouri state birth records from 2003 to 2009 were utilized to identify families with children aged 13, 15, 17, or 19 years and one or two full siblings. Mothers were contacted for telephone surveys to determine family risk status based on paternal history of alcoholism. Families where the mother reported paternal excessive drinking were classified as High-Risk (HR); otherwise the family was categorized as Low-Risk (LR). Families were classified as Very High Risk (VHR) based on driving record evidence of 2+ paternal DUIs. Mothers completed the telephone screen to confirm that the men with multiple DUIs were the children’s biological father [42]. European Americans were excluded due to low prevalence of reported racial discrimination. The final enrollment of African-American families was 151 LR, 150 HR, and 149 VHR families. Of the 450 African-American families enrolled, 406 of these families had at least one child interviewed, with 797 offspring having complete data (out of 806) for the analyses.

After the mother’s interview, permission was obtained to contact offspring for participation. Offspring whose mothers granted permission were invited to participate and only those who also consented to the study were interviewed. Offspring completed a comprehensive psychiatric telephone interview and were sought for follow up every 2 years. Comprehensive psychiatric interviews developed by the Midwest Alcoholism Research Center were based on the Semi-Structured Assessment for the Genetics of Alcoholism [43]. This study utilizes data from baseline offspring interview and maternal interview.

Measures

Racial and Socioeconomic Discrimination

Questions assessing racial and socioeconomic discrimination were adapted from the Experiences of Discrimination measure [4]. For racial discrimination, participants were asked: “Have you ever experienced racial discrimination, that is because of your race or color, been prevented from doing something or been hassled or made to feel inferior; in any of the following situations: at school, getting a job, at work, at home, getting medical care, on the street or in a public setting, and from the police or in the courts.” For socioeconomic discrimination, the introductory phrase was “Because of your social class, that is your social or economic class, have you ever experienced discrimination, been prevented from doing something or been hassled or made to feel inferior in any of the following situations,” followed by the same 7 domains described above. If the participant reported perceived discrimination, they were then asked if they experienced discrimination “often,” “sometimes,” or “rarely.” Analyses revealed that participants reported similar levels of distress due to discrimination, regardless of frequency, consistent with previous results [28].

The Krieger paper suggests counting discriminatory experiences in three or more domains as experience of discrimination; however, because some of the domains were not relevant for many of our young participants, we elected to follow methods adopted in prior publications using these data [28, 44] by counting discrimination in at least one domain as having experienced discrimination. The proportion of respondents who reported discrimination in 1, 2 and 3+ domains were 24%, 12%, and 10% for racial and 8%, 4%, and 4%, for socioeconomic discrimination, respectively. To ensure that our results were not an artifact of setting a too-low domain threshold for offspring discrimination experiences, we coded a three-level variable to encompass discrimination exposures in 2 +, 1, or no domains, and tested the estimates for 2+ versus 1 only to determine, using multinomial regression, whether there were differences between 2+ and 1 domain of exposure.

Risk Behavior

This study focuses on two risk behavior types: 1) substance use and misuse as well as 2) conduct and aggressive behaviors. Substance-related behaviors included early use of alcohol, cannabis, and tobacco, problem use, and heavy use. Early substance use initiation was defined as initiating substance use in the youngest quartile of the population of users: before age 16 for alcohol, before age 15 for cannabis, and before age 13 for tobacco. Problems with alcohol and cannabis were defined as endorsing any one of ten lifetime criteria as defined by the DSM-5 for substance use disorder. The eleventh criterion, craving, was added to the DSM-5 criteria in 2013, long after data collection for the study was complete, thus this item was not included. The literature indicates that craving is a very severe item with negligible impact on AUD prevalence, hence its omission is highly unlikely to affect our results [45]. Tobacco problems were defined as endorsing any one of seven DSM-IV lifetime criteria for Tobacco Dependence; as noted earlier, this study occurred prior to DSM-5 finalization, so new elements for Tobacco Use Disorder were not assessed. We also captured more than minimal exposure to substance use, defined as drinking alcohol on at least 6 separate days [46], using cannabis at least 11 times [47], and using any form of tobacco at least 20 times [48].

Physical aggression was assessed with two questions: initiating a physical fight, and intentionally inflicting harm on another person. Conduct problems were defined as exhibiting three or more symptoms (excluding physical aggression) according to DSM-IV criteria.

Covariates

Maternal Racial and Socioeconomic Discrimination.

We accounted for intergenerational associations with discrimination experiences by including a measure of maternal discrimination using the same instrument as for the offspring. For mothers, discrimination was counted if the experience had occurred in 3+ domains, as recommended by the developers of the instrument.

Offspring Psychosocial Risk Factors.

We accounted for psychosocial factors that have been associated with various forms of discrimination, including suicidality [28, 29], depression [26, 27], and social phobia [31]. Endorsement of any suicide ideation, “Have you ever thought about taking your own life?”; plan, “Did you ever plan a way of taking your own life?”; or attempt, “Have you ever tried to take your own life?” were coded as positive for suicidality. Major depressive disorder (MDD) was defined as meeting at least five major symptoms (excluding suicide symptom) during a two-week period, according to the DSM-IV criteria. Social phobia was defined using DSM-IV criteria.

Sociodemographic Factors.

Other covariates included family risk status, offspring age and gender, mother’s and father’s education (based on maternal report), and maternal report of household income. Parental education was categorized into: less than high school, high school only (including the acquisition of a GED), and more than high school. Household income was categorized into: $29,999 or less, between $30,000 and $74,999, and $75,000 or above. Dummy variables were created for father’s education and income to account for missing values and single-parent households. Dummy variables were also created for family risk levels.

Data Analysis

We utilized LCA to examine patterns of risk behavior based on the substance-related and aggressive behaviors defined above. Latent class analysis assumes that, for a set of categorical variables or symptoms, there is an underlying construct with several mutually exclusive classes such that the observed characteristics within each class are independent (“local independence”). LCA produces, from heterogeneous categorical variables, homogeneous classes that are distinguished by their item-response probabilities within each class. Parameters produced from LCA include overall class prevalence (the probability that an individual will be assigned to a class), and conditional probabilities (the probability of endorsing each item within each class).

We considered multiple factors to assess model fit. Lower Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample-size adjusted BIC (SSABIC) scores and higher entropy values indicate better fit. We also considered interpretability and item-response probabilities when selecting the optimal number of latent classes. The assumption of local independence was also assessed using standardized residuals < ∣3.84∣ [49].

Logistic regression models were used to test the relationship between class membership and offspring racial and socioeconomic discrimination, with discrimination as the outcome variable. Because racial and socioeconomic discrimination were not highly correlated in offspring, they were treated as independent discrimination types. Separate models were run for racial and socioeconomic discrimination. Model 1 included the latent class variables, as well as the covariates of age (continuous), gender, family risk status, maternal education, paternal education, and household income, and dummy variables to reflect missing values and single-parent households. Model 2 added offspring psychopathology (suicidality, depression, and social phobia) to account for potential psychosocial associations with discrimination over and above associations with the risk-behavior classes. Model 3 included all variables from the previous model as well as mother’s experience of racial and socioeconomic discrimination. Because the correlation between mother’s experience of racial and socioeconomic discrimination was high (r > 0.8), separate models were run for each.

The data were prepared in SAS (Version 9.2, Cary, NC: SAS Institute Inc.). Latent class analyses were conducted in Mplus (Version 7.3, Los Angeles, CA: Muthén & Muthén) and logistic regression was performed in Stata (Version 11.2, College Station, TX: StataCorp LP). Prior to analysis, tetrachoric correlations between all variables were computed to examine potential collinearity problems (correlations 0.10–0.79). All analyses controlled for familial clustering.

RESULTS

On average, the offspring included in the present study were 17.9 years of age (median 17 years, range 12-38 years). Of the 797 African American offspring in this study, 400 (50.2%) were female. Nearly half (n = 368; 46.2%) reported racial discrimination (48.7% of males and 43.8% of females), while 125 (15.7%) reported experiencing socioeconomic discrimination (16.9% of males and 14.5% of females). Rates of discrimination in each domain were similar between genders; however, males reported more discrimination from the police or courts for both racial (22.5% compared to 8.2% in females 8.2%) and socioeconomic discrimination (9.1% compared to 3.8% in females 3.8%). A majority (52.3%) were from low-income households. Additionally, 58.9% of mothers, but only 29.5% of fathers, had education beyond high school. As was observed for offspring, mothers reported more racial discrimination (35%) versus socioeconomic discrimination (15.8%).

Latent Class Characteristics

We compared 1- to 6-class solutions based on 11 binary indicator variables. Fit indices suggested that the 4-class solution was optimal. The 4-class solution had the lowest BIC value and a high entropy value, indicating good fit. Moreover, the 4-class solution exhibited good separation, given that average probability of the most likely class membership ranged from 0.90 to 0.95 (Table 2). Finally, because none of the bivariate residuals exceeded ∣3.84∣, which would indicate high item interdependencies, the assumption of local independence was not violated. Endorsement probabilities for classes are displayed in Table 3; Figure 1 graphs the endorsement probabilities by class. Although there are measurement limitations linked to using the most likely categorical class assignment, which ignores the nonzero probabilities of other class memberships, the very high probabilities (means in excess of .9 indicative of largely unambiguous assignments) reduce concern that this approach leads to erroneous interpretation of associations with discrimination.

Table 2.

Model fit indices for 1 to 6 classes.

# of classes	AIC	BIC	SSABIC	Entropy	Average Latent Class Probabilities
1	8567.268	8618.84	8583.909	-	-
2	7093.612	7201.444	7128.406	0.914	.968-.981
3	6878.479	7042.572	6931.427	0.904	.904-.980
4	6715.754	6936.107	6786.855	0.881	.903-.950
5	6689.136	6965.749	6778.391	0.813	.806-.935
6	6676.621	7009.494	6784.029	0.822	.805-.919

Table 3.

Endorsement probabilities for each of the 4 latent classes.

Risk Characteristic	Low Use & Aggression (n = 507, 64%)	High Alcohol (n = 134, 17%)	Moderate Use & Aggression (n = 77, 10%)	High Use & Aggression (n = 79, 10%)
Endorse ≥ 1 symptom of alcohol use disorder	0	0.814	0	1
Drank on ≥ 6 separate days	0.06	0.955	0.273	1
Initiated alcohol use at ≤ 15 years	0.082	0.206	0.341	0.723
Binge drank (consumed ≥ 5 drinks in 24 hours)	0.017	0.738	0.124	0.891
Initiated cannabis use at ≤ 14 years	0.018	0.029	0.49	0.621
Used cannabis ≥ 11 times	0	0.203	0.428	0.803
Initiated tobacco use at ≤ 12 years	0.093	0.102	0.329	0.349
Used a form of tobacco ≥ 20 times	0.021	0.331	0.455	0.741
Has initiated a physical fight	0.239	0.219	0.565	0.661
Have hurt another person on purpose	0.067	0.113	0.262	0.363
Endorse ≥ 3 symptoms of conduct disorder	0.021	0.035	0.323	0.579

Figure 1.

Probabilities of endorsing each indicator variable in each of the 4 latent classes.

The four latent classes are described as follows:

Low substance use and low aggression (“Low Use & Aggression,” 63.6%).

The probabilities of all of the substance use indicators were all <0.1. Furthermore, the endorsement probabilities of conduct problems (0.02) and intentionally harming another (0.07) were very low, while the endorsement probability of fight initiation (0.24) was the lowest among the 4 classes.

High alcohol use and low aggression (“High Alcohol,” 16.8%).

High probabilities were observed for alcohol problems (0.81), regular drinking (0.96), and binge drinking (0.74), but low probability for early initiation (0.21), and much lower probabilities for cannabis or tobacco items. Conduct problems and physical aggression measures also had low endorsement probabilities, similar to the Low Use & Aggression class.

Moderate polysubstance use and moderate aggression (“Moderate Use & Aggression,” 9.7%).

Compared to the High Alcohol class, this class manifests higher endorsement probabilities for early initiation across all 3 substances (item-response probabilities of 0.34, 0.49, and 0.33 for alcohol, cannabis, and tobacco, respectively) and for regular use of cannabis and tobacco, but negligible to zero endorsement probabilities for heavier alcohol involvement such as binge drinking (0.12), and alcohol problems (0). This class also exhibited higher values for conduct problems (0.32), fight initiation (0.57), and intentional harm (0.26) compared to the High Alcohol group.

High polysubstance use and high aggression (“High Use & Aggression,” 9.7%).

This class had very high endorsement probabilities for all substance use items, including problems, regular and heavy use, and early initiation, as well as the highest probabilities of all the classes for intentional harm (0.36), fight initiation (0.66), and conduct problems (0.58).

Logistic Regression for Racial and Socioeconomic Discrimination

Base and fully-adjusted models of racial and socioeconomic discrimination associated with each latent class are reported in Tables 4a and 4b.

Table 4a.

Logistic regression analysis predicting offspring racial discrimination.

	Model 1		Model 2		Model 3a		Model 3b
Description	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
High Use & Aggression	4.41*	2.40-8.11	3.66*	1.98-6.75	3.93*	2.11-7.29	3.65*	1.97-6.74
High Alcohol	1.73*	1.12-2.67	1.60*	1.02-2.50	1.63*	1.04-2.55	1.58*	1.01-2.48
Moderate Use & Aggression	2.12*	1.29-3.49	1.88*	1.14-3.10	1.98*	1.19-3.30	1.89*	1.14-3.12
Low Use & Aggression	REF		REF		REF		REF
Mother’s Racial Discrimination	-	-	-	-	1.38	0.98-1.93	-	-
Mother’s Socioeconomic Discrimination	-	-	-	-	-	-	1.44	0.96-2.17

^*signifies p < 0.05. All models controlled for age, sex, family risk type, maternal education, paternal education, and household income. Model 2 added offspring depression, suicidality, and social phobia. Model 3a added mother’s experience of racial discrimination and Model 3b mother’s experience of socioeconomic discrimination.

Table 4b.

Logistic regression analysis predicting offspring socioeconomic discrimination.

	Model 1		Model 2		Model 3a		Model 3b
Description	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
High Use & Aggression	6.06*	3.21-11.45	4.66*	2.46-8.85	4.78*	2.52-9.06	4.68*	2.49-8.79
High Alcohol	2.33*	1.29-4.20	1.96*	1.04-3.66	1.97*	1.05-3.70	1.93*	1.03-3.63
Moderate Use & Aggression	4.35*	2.38-7.95	3.45*	1.82-6.56	3.52*	1.84-6.74	3.42*	1.78-6.57
Low Use & Aggression	REF		REF		REF		REF
Mother’s Racial Discrimination	-	-	-	-	1.21	0.77-1.91	-	-
Mother’s Socioeconomic Discrimination	-	-	-	-	-	-	1.49	0.87-2.56

^*signifies p < 0.05. All models controlled for age, sex, family risk type, maternal education, paternal education, and household income. Model 2 added offspring depression, suicidality, and social phobia. Model 3a added mother’s experience of racial discrimination and Model 3b added mother’s experience of socioeconomic discrimination.

Racial Discrimination

In the base model, the High Alcohol, Moderate Use & Aggression, and High Use & Aggression classes were significantly associated with elevated likelihoods of experiencing racial discrimination compared to the Low Use & Aggression class. These associations remained significant after adjusting for offspring psychopathology (Model 2) and maternal racial and socioeconomic discrimination (Models 3a and 3b). The associations between offspring racial discrimination and the three risk behavior classes remained strong even after controlling for maternal racial (Model 3a) and socioeconomic (Model 3b) discrimination. Odds ratios were significantly different from the Low Use & Aggression class and ranged from 1.63-3.93 for racial and 1.58-3.65 for socioeconomic discrimination. No significant associations with maternal racial or socioeconomic discrimination were observed. We also tested whether the estimates for the latent classes were statistically different from each other. Estimates for the High Use & Aggression class differed from all other classes in all models, while those for the High Alcohol and Moderate Use & Aggression classes were not different from each other in all models.

Socioeconomic Discrimination

In Model 1, displayed in Table 4b, compared to the Low Use & Aggression class, all other classes demonstrated significant elevated odds of experiencing socioeconomic discrimination. These associations remained significant and strong after controlling for offspring psychopathology (Model 2) and maternal racial (Model 3a) and socioeconomic (Model 3b) discrimination. No association was observed for maternal racial or socioeconomic discrimination and offspring’s experience of socioeconomic discrimination. In Models 3a and 3b, the strength of the association between offspring socioeconomic discrimination and the risk behavior classes remained significant, with odds ratios ranging from 1.93–4.78. However, unlike results for offspring racial discrimination, differences were observed only between the High Alcohol class and High Use & Aggression class (all p’s < .01), with no differences between the Moderate and High Use & Aggression classes.

We also re-ran models using a three-level variable of discrimination exposures which distinguished those who experienced discrimination in 2 or more domains from those who reported discrimination in only one domain. In all models for offspring racial and socioeconomic discrimination, statistical testing revealed that estimates for 2+ domains versus 1 domain were not different from each other and could be combined (Results available upon request).

DISCUSSION

The present study expands existing research on discrimination and risk behavior among African American youth by examining multiple risk behaviors and by examining two sources of discrimination. These findings further our understanding of risk behavior in this population by highlighting the relevance of socioeconomic in addition to racial discrimination, as well as distinct patterns of risk behavior enabled by the study of multiple, rather than single, risk behaviors. This research also informs future targets for risk behavior prevention.

This study is novel in that it examines multiple risk behaviors simultaneously through latent class analysis. Using LCA, we identified four classes which differed in their combination and severity of risk behaviors. In these classes, the probability of early alcohol use initiation was relatively low in the High Alcohol group. Because black women tend to start drinking later in life, classes with a higher proportion of females, such as the High Alcohol class, may exhibit a lower probability of early alcohol use initiation [50]. Importantly, LCA enabled us to identify patterns of risk behavior that were differentially associated with discrimination, which would not have been obvious had we studied substance use and aggression separately. The High Use & Aggression class has an elevated likelihood of experiencing racial and socioeconomic discrimination relative to the Low Use & Aggression and High Alcohol classes, which suggests that higher levels of substance use and aggression are associated with discrimination.

We found evidence that racial discrimination is associated with increased risk behavior, which is generally consistent with prior studies on discrimination and risk behavior. A 2011 study of African American adolescents found that contextual stress (which included racial discrimination) was associated with substance use and aggressive behavior in males and substance use only in females [52]. Our study extends this work by considering specific forms of discrimination rather than a general measure of stress. Furthermore, recent studies have investigated the role of racial socialization in attenuating the influence of racial discrimination on substance use [53, 54]; suggesting a potential target for substance use prevention. The association between discrimination and risk behavior was significant, even after accounting for sociodemographic characteristics, offspring psychosocial factors, and maternal racial and socioeconomic discrimination. Moreover, maternal experiences of discrimination were not associated with offspring discrimination over and above offspring risk behavior, which increases our confidence in these results.

In addition, we found that, even after accounting for demographic, psychosocial, and maternal factors, there is a link between socioeconomic discrimination and increased risk behavior. We found a low correlation between racial and socioeconomic discrimination in offspring, suggesting that the two forms of discrimination are independent at this developmental juncture. These findings highlight the importance of investigating multiple sources of discrimination in the study of discrimination and risk behavior, given that most studies focus on racial discrimination only. To our knowledge, this is one of few studies that address socioeconomic discrimination. Socioeconomic discrimination was associated with all three risk behavior classes and that the effect estimate was significantly higher in the High Use & Aggression class compared to the High Alcohol class. This suggests that socioeconomic discrimination is more associated with the combination of substance use and aggression, as opposed to substance use alone. However, we are hampered in our further explanation of this association by the absence of timing information; that is, we are unable to determine whether those engaging in aggressive behaviors are more likely to elicit discrimination, or, conversely, whether experiences of socioeconomic discrimination provoke aggression.

Furthermore, it is important to note that socioeconomic discrimination and SES are independent constructs. In this study, socioeconomic discrimination was reported at similar levels regardless of family income status or maternal education. This may be partly explained by observing that most families were low-income and there was low variance of income in our sample. Previous studies focused on socioeconomic status rather than socioeconomic discrimination [35, 51]. Our study demonstrates that studying socioeconomic discrimination, in addition to SES, can help elucidate the complex network of socioeconomic factors linked to risk behavior.

One strength of this study is its investigation of intergenerational effects. We did not find that maternal discrimination mediated the association between offspring discrimination and risk behavior (Table 4). While it is important to evaluate potential cross-generational effects, the present study is limited because maternal experiences of discrimination were included as a covariate rather than as an exposure or outcome. Nonetheless, this could affect other studies and has been investigated using the Missouri Family Study previously: a recent study investigated the association between maternal discrimination and alcohol initiation [44], and another study found that the association between suicidality and racial discrimination in young African-American males is partially explained by maternal racial discrimination [28]. Future research may benefit from considering the role of parental experiences of discrimination on the health of offspring. Additionally, investigators could consider how changes in socioeconomic status between generations may interact with experiences of discrimination to influence risk behavior.

Limitations

Several important limitations of this study should be noted. First, our Missouri-based African American sample is enriched for paternal AUD and may not be generalizable to other ethnic groups, low-risk populations, broader geographic samples. Nonetheless, because high-risk populations may be more likely to engage in risk behavior, this study contributes to understanding an important population. Further, the classes reported here may not be obtained in other samples. Recovering similar classes in a general population sample would increase confidence in the class structure validity. Second, the Experiences of Discrimination scale has certain limitations. The sample used to validate the scale included a significant proportion of African Americans, but the scale has not been extensively used with African American youth populations [4]. Additionally, this study measured self-reported discrimination, but structural racism [55], institutional discrimination [56], and internalized racism [57] may also influence health outcomes. Third, participants have limited economic exposure due to their young age, which may affect socioeconomic discrimination. Racial and socioeconomic discrimination were collinear among mothers, but not offspring, which suggests that age is related to discriminatory experiences. Our results may understate the level of socioeconomic discrimination participants will face in the future as they accumulate work and labor market experience. Fourth, it is not possible to determine direction of effect from these cross-sectional data, since timing of the discrimination exposures was not available. Future studies that consider timing would enhance our understanding of discrimination and risk behavior. Although data from this study captured discrimination frequency, we did not include discrimination magnitude in our analyses. Studies suggest that different magnitudes of discrimination may have distinct associations with health outcomes [58]. Fifth, some relevant risk behaviors were not included, such as risky sexual behaviors. Although this study includes questions regarding risky sexual behavior, they were only addressed to participants fifteen years or older, which would have sharply reduced the sample size. However, risky sex would be an important addition to the array of behaviors studied here in future investigations. Finally, we limited our analyses to African American youth in the study, since experiences of racial discrimination were rarely reported among European American offspring. However, because socioeconomic discrimination exposure is not limited to a single race/ethnic group, future work should explore the impact of this understudied stressful exposure in a broad array of race/ethnic groups.

Future Directions

Despite these limitations, this work has several implications for future policy and research. A relationship between discrimination and health evidences a need for health policy that supports culturally-competent care. This need that is particularly relevant for the population studied here, given the history of racial, ethnic, and class prejudice against African Americans. Further study is needed on relationship between discrimination and other risk behaviors, such as risky sex, unhealthy dietary behaviors, inadequate physical activity, and illicit drug use. Moreover, future research should consider multiple sources of discrimination, given evidence that individuals who report multiple sources of discrimination face a “double disadvantage” in health [59]. Overall, understanding the relationship between discrimination and risk behavior can inform practitioners on future interventions to substance and behavioral problems.

Table 1.

Distribution of characteristics in African-American offspring (N=797).

Characteristic	n (%)
Mean age (SD)	17.9 (3.9)
Female	400 (50.2%)
Offspring Risk Behaviors
1+ Alcohol use disorder symptoms	191 (24%)
Drank on 6+ separate days	264 (33.1%)
Initiated alcohol use at ≤ 15 years	157 (19.7%)
Ever had a binge drinking episode	192 (24.1%)
Initiated cannabis use at ≤ 14 years	106 (13.3%)
Used cannabis 11+ times	130 (16.3%)
Initiated tobacco use at ≤ 12 years	117 (14.7%)
Used a form of tobacco 20+times	152 (19.1%)
Has initiated a physical fight	251 (31.5%)
Have hurt another person on purpose	101 (12.7%)
3+ Conduct disorder problems	90 (11.3%)
Offspring Racial Discrimination (any)	368 (46.2%)
At school	197 (24.7%)
Getting a job	77 (9.7%)
At work	72 (9.0%)
At home	7 (0.9%)
Getting medical care	20 (2.5%)
On the street or in a public setting	198 (24.8%)
From the police or in the courts	122 (15.3%)
Socioeconomic Discrimination (any)	125 (15.7%)
At school	72 (9.0%)
Getting a job	30 (3.8%)
At work	21 (2.6%)
At home	5 (0.6%)
Getting medical care	21 (2.6%)
On the street or in a public setting	48 (6.0%)
From the police or in the courts	51 (6.4%)