Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 May 4.
Published in final edited form as: Anxiety Stress Coping. 2018 Oct 29;32(1):109–123. doi: 10.1080/10615806.2018.1539964

“R U Mad?”: Computerized text analysis of affect in social media relates to stress and substance use among ethnic minority emerging adult males

Alethea Desrosiers a, Vera Vine b, Trace Kershaw a
PMCID: PMC8094922  NIHMSID: NIHMS1511062  PMID: 30373396

Abstract

Objective:

This study examined the interactive role of emotions and stress in substance use severity among ethnic minority, emerging adult males, using linguistic indicators of emotion obtained through social media.

Method:

Participants were 119 emerging adult, ethnic minority males (ages 18–25) who provided access to their mobile phone text messaging and Facebook activity for 6-months. Computerized text analysis (LIWC2015) was used to obtain linguistic indices of positive and negative affect from texts and Facebook posts. The Perceived Stress Scale was used to measure stress, and items from the Drug Abuse Screening Test were used to measure substance use severity.

Results:

Generalized estimating equations showed that higher negative emotion in texts was associated with greater substance use severity. Stress moderated the relationship between positive emotion expressed in Facebook posts and substance use such that higher positive affect in Facebook posts was associated with less substance use at higher stress and greater substance use at lower stress.

Conclusions:

Findings highlight the complexities of interactions between stress and affectivity. Findings could inform development of substance use interventions for young males that employ social technologies.

Keywords: Substance use, Affect, Social Media, Stress, Emerging Adults, Males

Substance use remains widespread among emerging adults (ages 18–25) in the United States (Substance Abuse and Mental Health Services [SAMSHA], 2015). Research on substance use among emerging adults is particularly important because rates of substance abuse are typically highest during this developmental period (Arnett, 2005; Stone et al., 2012), and earlier initiation of substance use is associated with greater odds of developing drug dependence (Hser & Anglin, 2010). Substance use may be particularly problematic for ethnic minority populations of emerging adult males. In comparison to Whites, Blacks and Hispanics are more likely to develop alcohol dependence (Mulia et al., 2009; Witbrodt et al., 2014) and are more likely to report relapse to substance use following treatment (Wahler & Otis, 2014). Likewise, among marijuana users, Blacks and Hispanics have higher rates of marijuana abuse and dependence than Whites (Wu, Zhu, & Swartz, 2016), and Blacks are more likely than Whites to develop early patterns of illicit drug use (Dean, Cole, & Bauer, 2014). Ethnic minority populations are also more likely to experience adverse social consequences from substance use (Witbrodt et al., 2014; Zemore et al., 2016). It is therefore highly pertinent to better understand factors that might contribute to substance use in young, ethnic minority males. As we explain below, considering the role of emotions at varying levels of stress is important to better understand substance use in this vulnerable population.

Emotion and Substance Use in Emerging Adult Males

Emerging adults experience a number of developmental changes (e.g., physical, cognitive, social) that impact emotional experience and engagement in risky behaviors, such as substance use. During this period, affective responses are heightened, motivation for positive arousal increases, and cognitive control of emotion and behavior is not fully developed (Somerville, Jones, & Casey, 2010; Steinberg, 2008). This combination of factors might create a propensity among some young adults to use substances to cope with negative emotions or to amplify positive emotions.

Theories positing that substance use is motivated by desires for tension-reduction or stress-response dampening (e.g., the self-medication or coping motivation hypothesis) have received empirical support and serve as a useful framework for understanding initiation and escalation of substance use behaviors (see Colder et al., 2010). High levels of negative emotion have been associated with initiation and frequency of alcohol, cigarette, and marijuana use in adolescents and emerging adults (Bonn-Miller et al., 2008; Cheetham et al., 2010; Fucito & Juliano, 2009). Regarding positive emotion, previous researchers have theorized that some individuals might use substances to enhance positive emotions (for review see Cheetham et al., 2010). Enhancement-motivated substance use is conceptualized as an appetitive or hedonic construct distinct from coping-motivated drinking, with distinct associations with positive emotion (Cooper et al., 2016; Cooper et al., 1995). Enhancement coping motives tend to cluster with social contexts that promote positive emotions, such as parties and celebrations (Colder & O’Connor, 2002). As such, substance use vulnerabilities associated with positive emotion are considered particularly relevant in the early stages of substance use disorder in emerging adults, namely onset and risky use (Cheetham et al., 2010; Zapolsky, Cyders, & Smith, 2009).

The Contextual Role of Stress in Substance Use of Ethnic Minority Males

There is increasing recognition of the need for multifactorial modeling of affective predictors of maladaptive behavior and psychopathology, because such models are well equipped to capture complex constellations of risk factors at play in different populations (e.g., Aldao, 2013; Nolen-Hoeksema & Watkins, 2012). Substance use behaviors in particular have been associated with complex interactions of myriad intrapersonal processes and stress-related contextual factors (see Sudhinaraset, Wigglesworth, & Takeuchi, 2016).

Stress has been consistently related to initiation of alcohol and drug use (DeHart et al., 2009; Hatzenbeuler, Nolen-Hoeksema, & Erikson, 2008) as well as the development of substance use disorders (Sinha, 2008). Ethnic minority emerging adults may be especially prone to stress-related substance use. Among Black emerging adults, stress exposure has accounted for more variance in alcohol and substance use than among their White counterparts (McNeil Smith & Taylor, 2015). Ethnic minority emerging adults may also experience heightened stress due to greater exposure to social and environmental stressors, such as discrimination, neighborhood violence, and poverty (Karriker-Jaffe et al., 2012; Krivo, Peterson & Kuhl, 2009; Lauristen & White, 2001). For these reasons, the experience of stress is a compelling potential moderator of emotional traits in the prediction of substance use severity in emerging adults. Furthermore, given the variety of motivations for substance use in this vulnerable population (e.g., Cooper et al., 2016), it is important to consider the moderating role of perceived stress in substance use that is both coping-related (i.e., predominantly negative affect-related), and enhancement-related (i.e., predominantly positive affect-related).

The stress-buffering hypothesis, commonly guiding the prediction of coping-related substance use, posits that individual differences in coping-relevant factors predict substance use more strongly in the context of stressful circumstances (Cohen & Wills, 1985). This hypothesis was first articulated in reference to social support as the buffering factor (Cohen & Wills, 1985), but recent variations have identified an array of internal traits involved in the experience and management of difficult experiences (e.g., Creswell & Lindsay, 2014; Graham, Calloway, & Roemer, 2015; Wills, Pokhrel, Morehouse, & Fenster, 2011). In this vein, trait negative affect has been viewed as especially predictive of maladaptive behavioral outcomes in the context of higher, as opposed to lower, stress (Clark, Kochanska, & Ready, 2000), whereas high positive affect has appeared to be most adaptive in higher stress contexts (Ong, Bergman, Bisconti, & Wallace, 2006). In sum, based on the stress-buffering perspective, negative affectivity is expected to be more strongly predictive of substance use severity, and positive affectivity is expected to be more protective among those who report higher levels of stress.

Estimating Affect via Social Media Communications

In order to adequately assess the interactive effects of stress and emotionality among emerging adult minority males, however, it is essential to obtain a meaningful index of affectivity. While self-report measures of emotion are valuable (e.g., Barrett, 2007; Scherer, 2005), they pose several problems. First, they rely on the capacity to identify and describe emotions, which is often limited (e.g., Vine & Aldao, 2014) and carries the risk of reporting biases. Second, correlations between stress and affectivity appear to become artificially inflated when both constructs are measured via self-report, a problem that can prevent the detection of potential interaction effects (see Clark et al., 2000). Third, the utility of self-reports of affect among emerging adult ethnic minority males might also be limited because affective self-expression may contradict traditional masculine norms and display rules for this population. According to theories of masculine norms, masculinity is defined by socially constructed expectations of acceptable male attitudes and behaviors (Pleck, 1995). In this paradigm, men are socialized to restrict or inhibit expression of affect because such expressiveness is viewed as a feminine characteristic (Connell, 2005). Previous research has also shown that ethnic minority males are more likely to ascribe to traditional masculine norms that prize restricted affectivity and detachment (Harris, Palmer & Struve, 2011; McGuire et al., 2014).

If ethnic minority males are inclined towards restricting affective expression, using alternatives to self-report measures of emotion could provide more accurate indices of trait affectivity. Analysis of language samples collected from participants is thought to offer more implicit-like markers of individual differences (e.g., Mehl & Pennebaker, 2003), based on the notion that, although we tend to be aware of the content or meaning of the words we say, we monitor far less the specific words we use in each utterance (Mehl & Pennebaker, 2003; Pennebaker, 2011). Thus, obtaining information about affective traits through linguistic analysis, such as linguistic analysis of communications in text messages and social networking sites (SNS), might be one way to bypass social norms regarding emotion disclosure among young urban males and gain a less filtered picture of their traits. Note, however, that because of this implicit-like strategy of measurement, linguistic indices are not considered indicators of subjectively felt emotion experience, but rather reflect the extent to which themes, including affective themes, are cognitively activated or accessed with or without awareness (see Pennebaker, 2011). Using text analysis to assess affective traits also has the benefit of reducing both self-report biases and method variance with self-reports of other constructs.

Language samples can be easily extracted from many modern forms of communication, including text messages and social media. Mobile phone use is nearly ubiquitous among youth in the United States, with 85% of emerging adults reporting that they own a smartphone (Smith, 2015). Emerging adults are more likely than other age groups to access social networking sites (SNS) on their mobile phones (Duggan et al., 2015), and African-American and Hispanic emerging adults have reported going online with greater frequency than those who are White (Lenhart, 2015). With this increased access to social networks through social media, emerging adults can connect with others instantaneously and often. Studies investigating associations between linguistic indicators of affect from non-online sources, such as diaries and expressive writing tasks, have shown that affect-related words are significantly correlated with symptoms of depression, anxiety, and stress (Hofmann et al., 2012; Rodriguez, Holleran, & Mehl, 2010; Tov et al., 2013). A few studies have also explored relationships between linguistic indicators of affect from online sources and emotional well-being, but results have been inconsistent (Settanni & Marengo, 2015; Wang et al., 2012). To date, no studies have conducted a text analysis of affect words from SNS and text messages and explored whether emotion-related words are associated with substance use.

Current Study

To address gaps in existing research, the current study seeks to investigate the relationships between affectivity (positive and negative), stress and substance abuse, in an at-risk and understudied population: ethnic minority, emerging adult males. The current study also attempts to assess positive and negative emotion in a novel way to reduce the possible role of self-presentation biases in this population, through linguistic indicators of positive and negative affect themes conveyed via texts message and Facebook. The aims of the current study are to: a) explore correlations among linguistic indices of positive and negative affect in social media (e.g., text messages and Facebook posts), stress, and substance use severity; b) examine whether linguistic indices of affect relate to substance use severity; and c) to examine whether stress moderates these associations. Based on previous research, we hypothesize that positive affect language (e.g., love, nice, sweet) will be inversely related to substance use severity, and negative affect language (e.g., ugly, hate, kill) will be positively related to substance use severity. We also predict, as discussed above, that among individuals reporting high (vs. low) stress, negative affect language will be more strongly associated with substance use severity, and positive affect language will be more inversely related to substance use severity. Study findings have the potential to shed light on underlying risk factors for substance abuse among ethnic minority, emerging adult males and to inform development of culturally relevant substance use interventions that employ social technology.

Methods

Participants

The study includes 119 emerging adult men participating in a larger study of social networks, cell phones, and health behavior. Participants were recruited in a small urban area in the Northeast using a time-location based effort to recruit initial network members based on ethnography and epidemiological assessments of the city to determine places where emerging adult men congregated. Within this small urban setting, recruitment efforts targeted areas that had high risk (e.g., high crime, STI rates, poverty). Trained outreach workers visited locations frequented by emerging adult men and publicized the study. Interested and eligible participants were informed about the study, provided informed consent and enrolled. Snowball sampling was then used to recruit friends in the participant’s social network (i.e., they listed individuals who they regularly spent time with and/or talked to about important things). Inclusion criteria included: 1) male gender; 2) age 18–25; 3) English-speaking; 4) heterosexual; 5) ownership of a cell phone with texting capabilities; and 6) ability to maintain cell phone service. We recruited an average of 10 participants per month, achieving an 80% participation rate of index recruits, and 50% recruitment of eligible network members.

The final sample identified as Black (80.0%), LatinX (16.8%), and Native American (3.2%) males with a mean (SD) age of 20.65 (1.86). For employment status, 49.5% were unemployed, 32.6% were employed part-time, and 17.9% were employed full time. 15.8% completed 11th grade or lower, 38.9% completed high school or GED, and 36.8% completed some college, 4.2% graduated college, and 4.3% completed some graduate or professional training. 49.5% were currently in school. Annual household income was less than $10,000 (40.0%), $10,000-$14,999 (11.6%), $15,000-$19,999 (6.3%), $20,000-$24,999 (4.2%), $25,000-$34,999 (7.4%), $35,000-$49,999 (8.4%), $50,000 or more (10.5%); 11.6% did not report household income.

Procedures

After consenting to participate in the study, participants loaded Mobilespy software onto their mobile phones, which extracted text message data to and from their network members who were participating in the study. No text messages were collected between participants and individuals not participating in our study, due to ethical concerns of collecting interactions of people who had not consented to be part of the study. Participants were also asked to accept a “follow” request from a dummy profile account on their SNS profile(s). Individual dummy accounts were generated for each participant to ensure confidentiality. Dummy accounts were assigned a pseudonym (e.g. Johnny B) and did not include any profile text or images, which ensured that the generated account could not be associated with the study by persons other than the participant and investigators. Facebook posts were obtained by using the extraction feature in NVivo, which extracts all posts from participants from a certain date, including the post, number of likes, and any comments. All information was de-identified so as not to include user names. All study procedures were thoroughly detailed in the informed consent and participants were allowed to opt in to all aspects of the study (e.g., text monitoring, Facebook monitoring).

Participants then completed an Audio Computer-Assisted Self Interview (ACASI) that collected data on substance use behaviors and attitudes and demographic characteristics. The ACASI method was used because it has the advantage of reducing self-presentation bias. Participants were compensated $75 for completing the interview and accepting the “follow” request(s). The University’s Institutional Review Board approved all study procedures. Participants’ SNS posts and text messaging were followed for 6 months. Self-report data was collected at baseline, 3-month follow-up, and 6-month follow-up. At 6-month follow-up, retention rates were 87%. For the purposes of the present study, our outcomes were self-report data on stress and substance use severity obtained at 6-month follow-up because text and Facebook data were collected during the entire 6-month period. This allowed us to assess the relationship between affect in texts and social media posts and subsequent substance use severity, controlling for initial baseline substance use severity.

Measures

Substance use.

Substance use severity was assessed at baseline and 6-month follow-up using a scale developed and validated for the purposes of the larger social network study (Gibson et al., 2015). Lifetime use of any drugs was assessed with a single, dichotomous item (yes=1, no=0) in which participants were asked if they have ever used drugs (i.e., marijuana, cocaine, glue, prescription drugs not prescribed by a doctor, heroin, ecstasy, methamphetamines, LSD, mushrooms, steroids). Participants who reported that they had used drugs in their lifetime also responded to 17 dichotomous items (yes=1, no=0) from the Drug Abuse Screening Test (DAST; Gavin, Ross, & Skinner, 1989) assessing severity of problems over the past 3 months associated with substance use (e.g., “Have you gone to anyone for help with drug problems”, “Have you lost friends because of your drug use”, “Do you ever feel bad or guilty about your drug use”). The DAST has good evidence of reliability and validity. Previous studies have shown that the DAST relates to independent DSM substance abuse diagnoses, providing strong evidence for concurrent validity that DAST is predictive of substance use severity (Gavin, Ross & Skinner, 1989; Yudko, Lozhkina, & Fouts, 2007). A count variable including all items was calculated to reflect overall substance use severity as measured by degree of substance use related problems.

Stress.

Perceived stress was measured at 6-month follow-up with the Perceived Stress Scale (PSS; Cohen &Williamson, 1988), which includes 10 items assessing perceived levels of stress (how unpredictable, uncontrollable, and overloaded individuals perceive their lives to be). Participants indicated how often in the past month they experienced stressful feelings and thoughts (e.g., “(un)able to control irritations in your life, unable to cope with all of the things you had to do, that difficulties were piling up so high that you couldn’t overcome them”). Responses were scored on a 5-point Likert scale ranging from 0 (never) to 4 (very often). The PSS has demonstrated good internal consistency in both adolescents (Devereaux, Weigel, Ballard-Reisch, Leigh, & Cahoon, 2009) and adults. Internal consistency in the current study was high (α =.75).

Estimated positive and negative affect.

To allow for indirect assessment of individual differences in affect, textual data from mobile phone text messages and Facebook posts was collected over a 6-month period and analyzed using the text processing software Linguistic Inquiry and Word Count (LIWC; Pennebaker et al., 2015). This program analyzes text samples by identifying words and word stems and categorizing them in linguistic and thematic categories. This version of LIWC can also read emoticons, commonly used in text messages and Facebook posts, for instance, classifying “:)” in the positive affect category. In the current study, we performed analyses using data from the affective LIWC categories called posemo, capturing positively valenced words (e.g., love, nice, sweet), and negemo, capturing negatively valenced words (e.g., ugly, hate, kill), as well as the three specific negative subcategories, anger, sadness, and anxiety. LIWC scores reflect the frequency of words in each category, expressed as a percentage of total word count. Average LIWC scores across a variety of English language text corpora (e.g., novels, natural speech, Twitter) are 3.67 for posemo and 1.84 for negemo (i.e., 3.67% and 1.84% of total words, respectively; Pennebaker et al., 2015). Rates of these affective words in tape-recorded speech samples have shown test-retest reliability of approximately .5 across a variety of everyday situations, considered substantial given this methodology (Mehl & Pennebaker, 2003). Affect LIWC scores correlate robustly with traditional self-report measures of trait affectivity, leading them to be considered as a proxy or verbal “signature” of individual differences in affect (e.g., Cohen, Minor, Baillie, & Dahir, 2008; Pennebaker, 2011; Pennebaker & King, 1999). Although validity of these LIWC categories has not been reported for the exact population and type of language sample used in the present study, previous reports comparing European American and African American participants show no group differences in LIWC dictionary capture (Pasupathi, Henry, & Carstensen, 2002) or affective word use (Cohen et al., 2008; Ong, Burrow, & Fuller-Rowell, 2012; Pasupathi et al, 2002).

It should be noted that the text message data was collected only among network members participating in the study. Therefore, the text message communications represent texts to a small sample of friends, and not all texts to all contacts. This was done for practical and ethical reasons, and because we were most interested in the way close friends communicate. Therefore, our text data represents communication to a sample of close friends; whereas, the Facebook data represents posts of participants to all Facebook friends.

Data Analysis Strategy

Descriptive statistics were performed for the entire sample. Bivariate correlations were performed to explore associations between linguistic indices of positive and negative affect in texts and Facebook, stress, and substance use. Next, generalized estimating equations (GEE) were used to construct a series of models to examine a) whether linguistic indices of positive and negative affect predicted substance use severity at 6-month follow-up, controlling for baseline substance use severity, and b) whether stress moderated these associations. GEE is appropriate because individuals are nested within social networks. GEE models the dependence in the data by adjusting for the correlations between individuals in the same social network (Zorn, 2001). Because the data is a count variable, and non-normally distributed, we used a negative binomial link function (Zorn, 2001). Separate models were constructed for positive and negative affect in text messages and Facebook posts. For interpretation of significant interactions, we calculated simple effects. Model parameters (negative affect and positive affect) were assessed at varying levels of stress because interactions involved continuous variables. We used standard practice by assessing parameters at the mean, one standard deviation below the mean, and one standard deviation above the mean (See Preacher, Curran & Bauer, 2006) to understand how the effects of positive and negative affect words vary across the continuous measure of stress. For models in which negative affect words significantly predicted substance use severity, GEE was conducted to examine whether any negative affect subcategories (anger, anxiety, sadness) predicted substance use severity. Control variables included age, race/ethnicity (Black=reference group), employment, and education because these variables have been associated with substance use previously. Control variables that were significantly related to substance use severity in the current study were retained in all models. Social network and baseline substance use severity were also controlled for in all models.

Results

Descriptive Statistics

The most frequently reported substances used were marijuana (80.0%) and alcohol (76.9%); participants also reported using ecstasy (18.5%) prescription drugs (10.8%), cocaine (3.1%), mushrooms (3.1%) and LSD (3.1%). The mean (SD) score for substance use severity was 1.52 (2.25) and for stress it was 19.25 (6.26).

Between the baseline and 6-month follow-up interviews, 95 participants had provided text message and Facebook data. On average, the LIWC dictionary was able to categorize at least 70% of participants’ words. The mean (SD) LIWC word count for texts was 6.53 (2.53), and Facebook posts was 9.65 (4.92). Together, participants’ text message and Facebook data provided a high total word count for linguistic analysis for each individual (M=2,177.57; SD=3,749.12), well exceeding published recommendations (Boyd, 2017).

Bivariate Correlations

Bivariate correlations and means scores for affect scores, stress, and substance use severity are presented in Table 1. Negative affect in texts was significantly positively related to substance use severity at 6-month follow-up (r=.433, p<.05). Stress was also significantly positively related to substance use severity at 6-month follow-up (r=.417, p<.05). Linguistic indices of affect were only modestly related to stress at 6-month follow-up (rs −.009 to .151).

Table 1:

Intercorrelations, means, and standard deviations for LIWC scores, stress, and substance use

 1.  2.  3.  4.  5.  6.
1. Substance Use severity            
2. Stress  .417**          
3. Posemo Texts −.105 −.009        
4. Negemo in Texts  .433**  .046 −.158      
5. Posemo in Facebook Posts  .112  .151  .284  .123    
6. Negemo in Facebook Posts −.065  .003 −.282 −.071 −.097  ---
Mean  2.32  17.30  8.21  2.25  11.40  3.64
Standard Deviation  2.22  5.60  8.89  1.97  9.13  3.21
Open in a new tab

Note.

*

p<.05

**

p<.005.

Note. Posemo = LIWC score reflecting presence of words with positive affect valence as a percentage of total word count. Negemo = LIWC score reflecting presence of words with negative affect valence as a percentage of total word count.

GEE Models Examining Effects of Affect on Substance Use Severity

Results of GEE examining associations between positive and negative affect scores and substance use severity at 6-months follow-up are displayed in Table 2 (See Model 1). All models controlled for age and race/ethnicity, as well as baseline substance use severity. We present unstandardized coefficients given the correlated nature of the data within networks, as is recommended for correlated data structures (Gelman & Hill, 2014; Kashy & Kenny, 2000). In the first series of models, more negative affect in texts was significantly associated with greater substance use severity (B=.253, p<.05). No significant associations were found between substance use severity and negative affect in Facebook posts, and positive affect in either texts or Facebook posts. Results of GEE examining negative affect sub-categories (anger, sadness, and anxiety) for texts showed that greater substance use severity was associated with higher presence of anger-themed words (B=.351, p<.01). No significant associations were found between substance use severity and anxiety and sadness themes in texts.

Table 2:

GEE Models Examining Effects of Stress, Positive and Negative Affect in Texts and Facebook Posts on Substance Use Severity in Young Men

 Model 1
  Model 2
 B  SE  B  SE
Texts
Positive Affect in Texts −.021  (.013) −.019  (.013)
Negative Affect in Texts  .253*  (.089)  .257  (.101)
Stress      .089*  (.028)
Positive Affect in Texts x Stress      .001  (.002)
Negative Affect in Texts x Stress      .001  (.011)
Facebook (FB) Posts
Positive Affect in FB Posts −.030  (.023) −.136*  (.045)
Negative Affect in FB Posts  .007  (.034) −.120  (.086)
Stress     −.037  (.044)
Positive Affect in FB Posts x Stress      .009**  (.003)
Negative Affect in FB Posts x Stress      .009  (.006)
Open in a new tab

Note.

*

p< .05

**

p<.005.

All models controlled for race/ethnicity, age, social network, and baseline substance use severity.

Note. Substance use severity at 6-months is the predicted outcome variable in all models. Values are unstandardized estimates.

GEE Models Examining Stress as a Moderator

In a second series of models, we examined stress as a possible moderator of the relationships between substance use severity and affect scores (See Table 1, Model 2). Stress significantly moderated associations between substance use severity and positive affect Facebook posts (B=.009, p<.05). Results of simple effects for significant interactions are displayed in Figure 1. Among individuals reporting higher levels of stress, positive affect in Facebook posts was significantly associated with lower substance use severity (B=−.033, p<.05). Among individuals reporting lower stress, positive affect in Facebook posts was significantly associated with greater substance use severity (B=.087, p<.05). The interaction between stress and negative affect in Facebook posts was not significant, nor was the interaction between stress and positive and negative affect in texts.

Figure 1:

Figure 1:

Open in a new tab

Simples slopes for positive affect in Facebook posts, stress, and substance use severity. High stress is 1 SD above the mean and low stress is 1SD below the mean.

Discussion

As predicted, we found that individual differences in negative affectivity, indexed by the frequency of negatively valenced words in text messages during a 6-month sampling period, were related to 6-month substance use severity, controlling for baseline severity. Future studies should investigate the direction of this relationship, as the nature of our data does not support inferences as to whether negative affect language preceded increases in substance use, or vice versa.

It is interesting that the relationship between affectivity and substance use was driven by presence of anger themes in communications, but not sadness or anxiety. This may be consistent with accounts of emotion socialization among emerging adult males (Harris et al., 2011; McGuire et al., 2014). Future research would need to disentangle whether the emotional expressions associated with risk for substance use are confined to anger, or whether other negative emotions are also associated with substance use vulnerability in emerging adult samples. Perhaps males in our sample experienced a wider array of emotions but only verbally expressed enough anger words to serve as a useful linguistic marker of associations between negative affectivity and substance use.

Regarding the effects of positive affect and stress in predicting substance use, we found a significant interaction between positive affect, indexed by the relative frequency of positively valenced words in Facebook posts, and stress. Among individuals reporting higher stress, higher positive affect scores were associated with lower substance use severity. This pattern is consistent with past findings of greater benefits associated with positive emotions in high- vs. low-stress contexts (Ong et al., 2006), and more broadly with the long-standing view that protective, coping-relevant traits are more strongly related to outcomes under high stress (e.g., Cohen & Wills, 1985). Expressing positive emotion in times of stress is believed to help individuals retain social support, with important implications for coping (Papa & Bonanno, 2008). This leads us to speculate that perhaps at high levels of stress, positive affectivity in communications via Facebook reflected individuals’ abilities to maintain social bonds in ways that ultimately reduced risk for increases in substance use. However, given the small size of our effect, our findings remain in need of replication, and future work is needed to unpack the social coping implications of positive emotion expression via Facebook.

Interestingly, our interaction results suggest that substance use may have functioned differently among the low-stress individuals in our sample, such that higher positive affect language co-occurred with elevated substance use. Although we did not measure motives for substance use, this pattern is consistent with enhancement motives attributed to some substance use behaviors, (Cheetham, 2010; Cooper et al., 2016; Zapolsky et al., 2009). Enhancement-motivated drinking is more prevalent in social contexts promoting positive feelings (e.g., parties; Colder & O’Connor, 2002). It is possible that the combination of lower-than-average perceived stress, positive affect language, and elevated substance use in our sample was a proxy for participation in such social contexts. Overall, while preliminary, our findings highlight the complexity of stress-emotionality interactions and the need for further investigation of the varying socio-emotional functions of substance use behavior in this population. Given the elevated substance use severity we found associated with positive affectivity estimates in low-stress individuals, it may be especially instructive to assess enhancement- as well as coping-related motivations for substance use in future research.

This study has several strengths, including sampling from an understudied and underserved urban population, longitudinal sampling, and collection of objective social media behavior. It can also be considered an overall strength of our study that, instead of relying on self-report of affect, we used a more objective approach involving the computerized analysis of text samples from texts with friends and social media use. Text analysis is able to detect subtle patterns in word use of which individuals are most likely unaware (Pennebaker, 2011). For instance, two utterances could have virtually the same meaning, such as I’m not sure if he’ll come and I’m worried he won’t come, and speakers may not be acutely aware of their word choice. Yet, repeated recruitment of negatively valenced words (e.g., worried) might indicate that the second individual is meaningfully different from the first. Even if an utterance is not explicitly about the self (e.g., from our sample, “R U mad?”), from the perspective of computerized text analysis, the presence of the word “mad” indicates that a negative affect theme is cognitively activated (consciously or not), or that this is likely an individual with ready mental access to that theme. Thus, text samples aggregated over several measurement instances can yield valid indicators of relative presence of certain themes or concerns between individuals, and can therefore be considered proxies or “signatures” of personality traits (e.g., Cohen et al., 2008; Mehl & Pennebaker, 2003; Pennebaker, 2011; Pennebaker & King, 1999). This linguistic method for assessing emotionality might also be preferable to self-report in our sample because of the heightened role of display rules and other social norms in restricting the affect expression and disclosure in young, ethnic minority men (Harris, Palmer & Struve, 2011; McGuire et al., 2014), which could impact how self-report measures are answered.

At the same time, using linguistic analysis to estimate individual differences in affectivity requires careful consideration of some caveats. The LIWC approach relies on word-level analysis and is not sensitive to multiple word meanings or to the speaker’s intended meaning, which can be a potential problem, depending on the nature of the research question. While some newer software programs (e.g., tidytext in R; Silge & Robinson, 2018) can parse text into larger units of analysis (e.g., longer word sequence or sentences), ultimately, coding by human raters (or self-report) is required to more closely infer participant’s felt experience. The word-level limitation of LIWC does not pose a serious problem, however, when the focus is on covertly estimating overall trait affectivity, as in the present study. The LIWC approach was designed to recognize high frequency, colloquial words, and thus the percentage of word capture varies depending on the nature of the text (for example, approx. 83% of words in novels vs. 54% of words in scientific articles are captured; Pennebaker, Chung, Gonzalez, & Booth, 2007). In our sample, on average 70% of words were recognized by the software, leaving another 30% of words that could have revealed more idiosyncratic, but important insights about individuals that might have improved the precision of affect estimates. Thus, the LIWC program presents a trade-off in that it is a somewhat blunt, but efficient, well established, and reliable tool for comparing individuals within a given sample to each other on the presence of psychologically meaningful themes in speech. Additional work is needed contrasting digital vs. human coding of social media communications by ethnic minority, emerging adults in order to fully evaluate the precision of the LIWC affect scores in this population.

Linguistic behavior is emitted in a social context, so it is important to consider the potential social demands that may impact word use on Facebook and in text messages. Individuals are believed to cultivate their Facebook presences in order to present themselves in a favorable light (e.g., Rosenberg & Egbert, 2011; Seidman, 2013). Consistent with this notion, participants displayed a much higher proportion of positive (to negative) words in their Facebook posts than in their text messages with friends (see Table 1). And it is interesting that we found a similar, positive association between substance use and negative affect in texts and positive affect on Facebook (among low-stress individuals), suggesting that the meaning of each linguistic index might vary subtly depending on social media context and who the message is going to (e.g., close friends, wider group of individuals). Perhaps the more candid nature of text message data makes it better suited to detect substance use risk related to negative affect, whereas Facebook data may be more useful for detecting substance use risk related to positive affect enhancement motives. Note again that we did not assess substance use motivations per se; rather, these interpretations are offered tentatively based on results and their compatibility with stress buffering and biphasic accounts of substance use. Future research could better assess whether affectivity indexed via language in Facebook and other forms of social media differentially reveal substance use motivations.

Our results must be interpreted in the context of some limitations. The sample size was relatively small, which creates a possibility that significant affect-by-stress interactions were spurious. At the same time, effects using linguistic analysis tend to be smaller in size than effects using traditional self-report, so it is promising that we found a theoretically compelling pattern of results in a sample of this size. Our per-communication word counts were also on the low side, but this is not surprising given that short phrases are the typical mode of communication in social media and text messages. Future studies should continue to investigate use of LIWC2015 for linguistic analysis of social media communications. Another limitation is that we had text data only between a sample of friends and not all of the texts sent by the participant. However, we were most interested in how communications between close friends may impact substance use. Future studies could compare texts to individuals within a person’s social network and outside a person’s social network. The highly specific nature of the sample, which consisted of emerging adult males that were largely ethnic minorities from a single urban area in the Northeastern US, might also be viewed as a limitation. As such, our results may not generalize to other emerging adult populations or those with higher rates of illicit drug use.

Additionally, we did not include self-report of positive and negative affect, which might be useful to compare with linguistic indicators; while this comparison is well established in general (e.g., Pennebaker, 2011), it has not been explored in the context of stress and substance use in a sample such as ours. Future studies might benefit from including more thorough assessments of substance use behavior, including motivations for use or expectancies about the effects of use, in larger populations. Because our data on affectivity was collected over a 6-month period, our affect estimates capture more trait-level than state-level differences. Using physiological assessments of stress (e.g., heart rate variability) in addition to self-report and employing methods that more effectively tracked daily behavior in discrete time intervals (e.g., ecological momentary assessment) would also strengthen future research in this area. We also are not certain about the directionality of the relationship between texts and substance use. More sophisticated longitudinal techniques could be used in the future to better model directionality. Finally, because we used a somewhat novel measure of substance use severity in attempt to better capture psychosocial problems related to substance use, results should be interpreted with this in mind.

With replication and eventual extension, findings from the current study might inform substance use prevention and screening efforts in novel ways for emerging adult ethnic minority males. For example, data on the frequency of linguistic indicators of affect in social media might help identify those at higher risk of escalating substance use, especially in the case of negative affect indicators and elevated stress. Prevention models using text messages or cell phone applications might also consider incorporating techniques for self-monitoring of positive and negative affect as well as stress levels. These interventions could provide feedback through daily frequency reports of linguistic indices of affect, with suggestions for coping adaptively. For emerging adult males who may be less likely to seek services to manage stress or who may have limited access to resources, mobile-based interventions could be especially beneficial.

Conclusion

In closing, the current study presents a unique approach to assessing the interplay of affect and stress in predicting substance use severity among ethnic minority emerging adult males. Overall, these initial findings provide some support for the view that substance use is complexly, multiply determined, and that the psychological function of this behavior may vary depending on affective traits and social context (Sudhinaraset, Wigglesworth, & Takeuchi, 2016). Considering that ethnic minority males experience more social problems related to substance use (Witbrodt et al., 2014; Zemore et al., 2016), it is highly pertinent to better understand contributing factors, such as experiences of positive and negative affect. Given high exposure to external stressors among this population, it is also important to continue to elucidate how levels of stress might impact associations between affectivity and substance use. Future research should continue to explore ways to use social technologies to assess risk behavior and deliver interventions among at risk and underserved populations.

Acknowledgements:

This study was supported by a National Institute of Drug Abuse grant R21DA031146 (PI: Kershaw).

Footnotes

Disclosure Statement: The authors report no conflicts of interest.

References

  1. Aldao A (2013). The future of emotion regulation research: Capturing context. Perspectives on Psychological Science, 8, 155–172. doi: 10.1177/1745691612459518 [DOI] [PubMed] [Google Scholar]
  2. Arnett JJ (2005). The developmental context of substance use in emerging adulthood. Journal of Drug Issues, 35, 235–244. doi: 10.1016/j.addbeh.2013.08.001 [DOI] [Google Scholar]
  3. Barrett LF, Mesquita B, Ochsner KN, & Gross JJ (2007). The experience of emotion. Annual Review of Psychology, 58, 373–403. doi: 10.1146/annurev.psych.58.110405.085709 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bonn-Miller MO, Vujanovic AA, & Zvolensky MJ (2008). Emotion dysregulation: Association with coping-oriented marijuana use motives among current marijuana users. Substance Use & Misuse, 43, 1656–1668. doi: 10.1080/10826080802241292 [DOI] [PubMed] [Google Scholar]
  5. Clark LA, Kochanska G, Ready R (2000). Mothers’ personality and its interaction with child temperament as predictors of parenting behavior. Journal of Personality and Social Psychology, 79(2), 274–285. [DOI] [PubMed] [Google Scholar]
  6. Cheetham A, Allen NB, Yucel M, & Lubman DI (2010). The role of affective dysregulation in drug addiction. Clinical Psychology Review, 30, 621–634. doi: 10.1016/j.cpr.2010.04.005 [DOI] [PubMed] [Google Scholar]
  7. Colder CR, Chassin L, Lee MR, & Villalta IK (2010). Developmental perspectives: Affect and adolescent substance use. In Kassell JD (ed.) Substance Abuse an Emotion (pp 109–135). American Psychological Association: Washington, D. C. [Google Scholar]
  8. Colder CR, & O’Connor R (2002). Attention bias and disinhibited behavior as predictors of alcohol use and enhancement reasons for drinking. Psychology of Addictive Behaviors, 16, 325–332. doi: 10.1037/0893-164X.164.325 [DOI] [PubMed] [Google Scholar]
  9. Cohen AS, Minor KS, Baillie LE, & Dahir AM (2008). Clarifying the linguistic signature: Measuring personality from natural speech. Journal of Personality Assessment, 90(6), 559–563. 10.1080/00223890802388459 [DOI] [PubMed] [Google Scholar]
  10. Cohen S & Wills TA (1985). Stress, social support, and the buffering hypothesis. Psychological bulletin, 98(2), 310–357. doi: 10.1037/0033-2909.98.2.310 [DOI] [PubMed] [Google Scholar]
  11. Connell RW (2005). Masculinities Berkeley: University of California Press. [Google Scholar]
  12. Cooper ML, Frone MR, Russell M, & Mudar P (1995). Drinking to regulate positive and negative emotions: A motivational model of alcohol use. Journal of Personality and Social Psychology, 5, 990–1005. [DOI] [PubMed] [Google Scholar]
  13. Cooper ML, Kuntsche E, Levitt A, Barber LL, & Wolf S (2016). Motivational models of substance use: A review of theory and research on motives for using alcohol, marijuana, and tobacco. In Sher KJ (ed.) The Oxford Handbook of Substance Use Disorders, Volume 1, pp. 375–396. [Google Scholar]
  14. Creswell JD, & Lindsay EK (2014). How does mindfulness training affect health? A mindfulness stress buffering account. Current Directions in Psychological Science, 23, 401–407. doi: 10.1177%2F0963721414547415 [Google Scholar]
  15. Dean DO, Cole V, & Bauer DJ (2014). Delineating prototypical patterns of substance initiations over time. Addiction, 110, 585–594. doi: 10.1037/a0034831 [DOI] [PubMed] [Google Scholar]
  16. Devereaux PG, Weigel DJ, Ballard-Reisch D, Leigh G, & Cahoon KL (2009). Immediate and longer term connections between support and stress in pregnant/parenting and non-pregnant/nonparenting adolescents. Child and Adolescent Social Work Journal, 26, 431–446. doi: 10.1007/s10560-009-0175-z [DOI] [Google Scholar]
  17. DeHart T, Tennen H, Armeli S, Todd M, & Mohr C (2009). A diary study of implicit self esteem, interpersonal interactions and alcohol consumption in college students. Journal of Experimental Social Psychology, 45(4), 720–730. doi: 10.1016/j.jesp.2009.04.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Duggan M, Ellison NB, Lampe C, Lenhart A, & Madden M (2015). Social media update 2014 Pew Research Center. Available at: http://www.pewinternet.org/2015/01/09/social-media-update-2014/. [Google Scholar]
  19. Fucito LM, & Juliano LM (2009). Depression moderates smoking behavior in response to a sad mood induction. Psychology of Addictive Behaviors, 23, 546–551. doi: 10.1037/a0016529 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Gavin DR, Ross HE, & Skinner HA (1989). Diagnostic validity of the Drug Abuse Screening Test in the assessment of DSM-III drug disorders. British Journal of Addiction, 84, 301–307. doi: 10.1111/j.1360-0443.1989.tb03463.x [DOI] [PubMed] [Google Scholar]
  21. Gelman A, & Hill J (2014). Data analysis using regression and multilevelhierarchical models (Vol. 1). New York, NY, USA: Cambridge University Press. [Google Scholar]
  22. Gibson C, Perley L, Bailey J, Barbour R, & Kershaw T (2015). Social network and census tract-level influences on substance use among emerging adult males: An activity spaces approach. Health & Place, 35, 28–36. doi: 10.1016/j.healthplace.2015.06.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Graham JR, Calloway A, & Roemer L (2015). The buffering effects of emotion regulation in the relationship between experiences of racism and anxiety in a Black American sample. Cognitive Therapy and Research, 39, 553–563. doi: 10.1007/s10608-015-9682-8 [DOI] [Google Scholar]
  24. Harris F, Palmer RT, Struve LE (2011). “Cool Posing” on campus: A qualitative study of masculinities and gender expression among Black men at a private research institution. The Journal of Negro Education, 80, 47–62. [Google Scholar]
  25. Hatzenbuehler ML, Nolan-Hoeksema S, & Erickson SJ (2008). Minority stress predictors of HIV risk behavior, sub- stance use, and depressive symptoms: Results from a prospective study of bereaved gay men. Health Psychology, 27(4), 455–462. doi: 10.1037/0278-6133.27.4.455 [DOI] [PubMed] [Google Scholar]
  26. Hofmann SG, Moore PM, Gutner C, & Weeks JW (2012). Linguistic correlates of social anxiety disorder. Cognition and Emotion, 26, 720–726. doi: 10.1080/02699931.2011.602048 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hser Y, Anglin MD (2010). Addiction treatment and recovery careers. In Kelly JE and White WL (eds.), Addiction Recovery Management: Theory Science and Practice (pp. 9–29). New York: Springer Science. [Google Scholar]
  28. Karriker-Jaffe KJ, Zemore SE, Mulia N, Jones-Webb RJ, Bond J, & Greenfield TK (2012). Neighborhood disadvantage and adult alcohol outcomes: Differential risk by race and gender. Journal of Studies on Alcohol and Drugs, 73, 865–873. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Kashy DA, & Kenny DA (2000). The analysis of data from dyads and groups. Handbook of research methods in social and personality psychology, 38, 451–477 [Google Scholar]
  30. Krivo LJ, Peterson RD, & Kuhl DC (2009). Segregation, racial structure, and neighborhood violent crime. American Journal of Sociology, 114, 1765–1802. [DOI] [PubMed] [Google Scholar]
  31. Lenhart A (2015). Teens, social media & technology overview 2015 Pew Research Center. Available at: http://www.pewinternet.org/2015/04/09/teens-social-media-technology-2015/ [Google Scholar]
  32. McGuire KM, Berhanu J, Davis CHF, & Harper SR (2014). In search of progressive black masculinities: Critical self-reflections on gender identity development among black undergraduate men. Men and Masculinities, 17, 253–277. doi: 10.1177%2F1097184X13514055 [Google Scholar]
  33. McNeil Smith S & Tayor J (2015). The relationship between social stress and substance use among black youths residing in South Florida. Journal of Child & Adolescent Substance Abuse, 25, 372–378. [Google Scholar]
  34. Mehl MR, & Pennebaker JW (2003). The sounds of social life: A psychometric analysis of students’ daily social environments and conversations. Journal of Personality and Social Psychololgy, 84, 857–870. doi: 10.1080/1067828X.2013.872062 [DOI] [PubMed] [Google Scholar]
  35. Nolen-Hoeksema S, & Watkins ER (2011). A heuristic for developing transdiagnostic models of psychopathology: Explaining multifinality and divergent trajectories. Perspectives on Psychological Science, 6(6), 589–609. doi: 10.1177/1745691611419672 [DOI] [PubMed] [Google Scholar]
  36. Ong AD, Bergman CS, Bisconti TL, & Wallace KA (2006). Psychological resilience, positive emotions, and successful adaptation to stress in later life. Journal of Personality and Social Psychology, 91, 730–749. doi: 10.1037/0022-3514.91.4.730 [DOI] [PubMed] [Google Scholar]
  37. Ong AD, Burrow AL, & Fuller-Rowell TE (2012). Positive emotions and the social broadening effects of Barack Obama. Cultural Diversity and Ethnic Minority Psychology, 18(4), 424. doi: 10.1037/a0029758 [DOI] [PubMed] [Google Scholar]
  38. Papa T & Bonanno G (2008). Smiling in the face of adversity: The interpersonal and intrapersonal functions of smiling. Emotion, 8, 1–12. doi: 10.1037/1528-3542.8.1.1 [DOI] [PubMed] [Google Scholar]
  39. Pasupathi M, Henry RM, & Carstensen LL (2002). Age and ethnicity differences in storytelling to young children: Emotionality, relationality and socialization. Psychology and Aging, 17(4), 610. doi: 10.1037/0882-7974.17.4.610 [DOI] [PubMed] [Google Scholar]
  40. Pennebaker J (2011). The secret life of pronouns: What our words say about us New York: Bloomsbury Press. [Google Scholar]
  41. Pennebaker J, Boyd RL, Jordan K, & Blackburn K (2015). The development and psychometric properties of LIWC2015 Austin TX: University of Texas at Austin. [Google Scholar]
  42. Pennebaker JW, Chung CK, Ireland M, Gonzales A, & Booth RJ (2007). The development and psychometric properties of LIWC2007: LIWC. net. Google Scholar
  43. Pennebaker JW, & King LA (1999). Linguistic styles: Language use as an individual difference. Journal of personality and social psychology, 77(6), 1296. doi: 10.1177/0002764293036003002 [DOI] [PubMed] [Google Scholar]
  44. Pleck JH (1995). The gender role strain paradigm. An update. In Levant RF & Pollack WS (Eds.), The new psychology of men (pp. 11–32). New York: Basic Books. [Google Scholar]
  45. Preacher KJ, Curran PJ, & Bauer DJ (2006). Computational tools for probing interactions in multiple linear regression, multilevel modeling, and latent curve analysis. Journal of educational and behavioral statistics, 31(4), 437–448. doi: 10.3102/10769986031004437 [DOI] [Google Scholar]
  46. Rodriguez AJ, Holleran SE, & Mehl MR (2010). Reading between the lines: the lay assessment of subclinical depression from written self-descriptions. Journal of Personality, 78, 575–598. doi: 10.1111/j.1467-6494.2010.00627.x [DOI] [PubMed] [Google Scholar]
  47. Rosenberg J, & Egbert N (2011). Online impression management: Personality traits and concerns for secondary goals as predictors of self-presentation tactics on Facebook. Journal of Computer-Mediated Communication, 17, 1–18. doi: 10.1111/j.1083-6101.2011.01560.x [DOI] [Google Scholar]
  48. Scherer KR (2005). Unconscious processes in emotion. In Barrett LF, Niedenthal PM, & Winkielman P (Eds.), Emotion and consciousness (pp. 312–34). New York: Guilford Press. [Google Scholar]
  49. Seidman G (2013). Self-presentation and belonging on Facebook: How personality influences social media use and motivations. Personality and Individual Differences, 54, 402–407. doi: 10.1016/j.paid.2012.10.009 [DOI] [Google Scholar]
  50. Settanni M & Marengo D (2015). Sharing feelings online: studying emotional well-being via automated text analysis of Facebook posts. Frontiers in Psychology, 6: 1045. doi: 10.3389/fpsyg.2015.01045 [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Silge J, & Robinson D (2018). Text mining with R: A tidy approach O’Reilly Media, Inc: Sebastopol, CA. [Google Scholar]
  52. Sinha R (2008). Chronic stress, drug use, and vulnerability to addiction. Annals of the New York Academy of Science, 1141, 105–130. doi: 10.1196/annals.1441.030 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Smith A (2015). U.S. Smartphone use in 2015. Pew Research Center Available at http://www.pewinternet.org/2015/04/01/us-smartphone-use-in-2015/
  54. Somerville LH, Jones RM, & Casey BJ (2010). A time of change: Behavioral and neural correlates of adolescent sensitivity to appetitive and aversive environmental cues. Brain and Cognition, 72, 124–133. doi: 10.1016/j.bandc.2009.07.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Steinberg L (2008). A social neuroscience perspective on adolescent risk-taking. Developmental Review, 28, 78–106. doi: 10.1016/j.dr.2007.08.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Stone AL, Becker LG, Huber AM, & Catalano RF (2012). Review of risk and protective factors of substance use and problem use in emerging adulthood. Addictive Behaviors, 37, 747–775. doi: 10.1016/j.addbeh.2012.02.014 [DOI] [PubMed] [Google Scholar]
  57. Substance Abuse and Mental Health Services Administration [SAMHSA]. (2015). Results from the 2014 National Survey on Drug Use and Health Retrieved from http://www.samhsa.gov/data/sites/default/files/NSDUH-FRR1-2014/NSDUH-FRR1-2014.pdf [PubMed]
  58. Sudhinaraset M, Wigglesworth C, & Takeuchi DT (2016). Social and cultural contexts of alcohol use: Influences in a social-ecological framework. Alcohol Research, 38, 35–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Tov W, Ng KL, Lin H, & Qiu L (2013). Detecting well-being via computerized content analysis of brief diary entries. Psychological Assessment, 25, 1069–1078. doi: 10.3389/fpsyg.2015.01045 [DOI] [PubMed] [Google Scholar]
  60. Vine V & Aldao A (2014). Impaired emotional clarity and psychopathology: A transdiagnostic deficit with symptom-specific pathways through emotion regulation. Journal of Social and Clinical Psychology, 33, 319–342. doi: 10.1521/jscp.2014.33.4.319 [DOI] [Google Scholar]
  61. Wahler EA & Otis MD (2014). Social stress, economic hardship, and psychological distress as predictors of sustained abstinence from substance use after treatment. Substance use & Misuse, 49, 1820–1832. doi: 10.3109/10826084.2014.935789 [DOI] [PubMed] [Google Scholar]
  62. Wang N, Kosinksi M, Stillwell DJ, & Rust J (2012). Can well-being be measured using Facebook status updates? Validation of Facebook’s gross national happiness index. Social Indices Research, 115, 483–491. doi: 10.1007%2Fs11205-012-9996-9 [Google Scholar]
  63. Wills TA, Pokhrel P Morehouse E, & Fenster B (2011). Behavioral and emotional regulation and adolescent substance use problems: A test of moderation effects in a dual-process model. Psychology of Addictive Behavior, 25, 279–292. doi: 10.1037/a0022870 [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Witbrodt J, Mulia N, Zemore SE, & Kerr WC (2014). Racial/Ethnic disparities in alcohol-related problems: Differences by gender and level of heavy drinking. Alcoholism: Clinical and Experimental Research, 38, 1662–1670. doi: 10.1111/acer.12398 [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Wu L, Zhu H, & Swartz MS (2016). Trends in cannabis use disorders among racial/ethnic populations groups in the United States. Drug and Alcohol Dependence 165, 171–190. doi: 10.1016/j.drugalcdep.2016.06.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Yudka E, Lozhkina O, & Fouts A (2007). A comprehensive review of the psychometric properties of the Drug Abuse Screening Test. Journal of Substance Abuse Treatment, 32, 189–198. doi: 10.1016/j.jsat.2006.08.002 [DOI] [PubMed] [Google Scholar]
  67. Zemore SE, Ye Y, Mulia N, Martinez P, Jones-Webb R, & Karriker-Jaffe K (2016). Poor, persecuted, young, and alone: Toward explaining the elevated risk of alcohol problems among Black and Latino men who drink. Drug and Alcohol Dependence, 163, 31–39. doi: 10.1016/j.drugalcdep.2016.03.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Zapolski TCB, Cyders MA, & Smith GT (2009). Positive urgency predicts illegal drug use and risky sexual behavior. Psychology of Addictive Behavior, 23, 348–354. doi: 10.1037/a0014684 [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Zorn, Christopher JW. “Generalized estimating equation models for correlated data: A review with applications.” American Journal of Political Science (2001): 470–490.

RESOURCES