Evaluating Within-Person Change in Implicit Measures of Alcohol Associations: Increases in Alcohol Associations Predict Increases in Drinking Risk and Vice Versa

Evaluating Within-Person Change in Implicit Measures of Alcohol Associations: Increases in... Abstract Aims Implicit measures of alcohol associations (i.e. measures designed to assess associations that are fast/reflexive/impulsive) have received substantial research attention. Alcohol associations related to the self (drinking identity), the effects of alcohol (alcohol excite) and appetitive inclinations (alcohol approach) have been found to predict drinking cross-sectionally and over time. A critical next step in this line of research and the goal of this study is to evaluate whether increases in the strength of these associations predict increases in drinking and vice versa. These hypotheses were tested in a sample of first- and second-year US university students: a sample selected because this time period is associated with initiation and escalation of drinking, peak levels of alcohol consumption and severe alcohol-related negative consequences. Short summary This study’s purpose was to evaluate whether increases in the strength of alcohol associations with the self (drinking identity), excitement (alcohol excite) and approach (alcohol approach) as assessed by implicit measures predicted subsequent increases in drinking risk and vice versa using a longitudinal, university student sample. Results were consistent with hypotheses. Methods A sample of 506 students’ (57% women) alcohol associations and alcohol consumption were assessed every 3 months over a 2-year period. Participants’ consumption was converted to risk categories based on NIAAA’s criteria: non-drinkers, low-risk drinkers and high-risk drinkers. A series of cross-lagged panel models tested whether changes in alcohol associations predicted subsequent change in drinking risk (and vice versa). Results Across all three measures of alcohol associations, increases in the strength of alcohol associations were associated with subsequent increases in drinking risk and vice versa. Conclusion Results from this study indicate bi-directional relationships between increases in alcohol associations (drinking identity, alcohol excite and alcohol approach) and subsequent increases in drinking risk. Intervention and prevention efforts may benefit from targeting these associations. INTRODUCTION Implicit measures of alcohol associations (measures designed to assess associations that are fast/reflexive/impulsive) predict drinking behaviors cross-sectionally and prospectively, even after controlling for their explicit measure counterparts (i.e. measures assessing slower/more reflective cognitions; Reich et al., 2010). Recent studies have also found that alcohol associations can predict changes in drinking behavior (Thush and Wiers, 2007; Lindgren et al., 2016a). A critical next step is to evaluate the dynamic relationship between alcohol associations and consumption by examining whether changes in associations predict changes in consumption and vice versa. These dynamic relationships have only been studied in an early adolescent sample (Colder et al., 2014). Thus, we evaluated whether changes in three well-validated implicit measures of alcohol associations were associated with changes in alcohol consumption in a sample of first- and second-year US university students. Interest in alcohol associations is motivated by dual process models of alcohol misuse (e.g. Wiers et al., 2007), which point to the importance of implicit and explicit cognitive processes in predicting drinking. Alcohol associations are commonly described as reflecting the extent to which alcohol becomes associated with other constructs in memory (see Stacy and Wiers, 2010). They are typically assessed indirectly—often by computer-based measures of reaction time in which the underlying cognitive process and content is inferred (see Nosek et al., 2011). The current study makes use of such measures. Although general theories of implicit cognition (Greenwald and Banaji, 1995; Nosek et al., 2011) suggest that alcohol associations will strengthen in response to direct experience as well as to one’s environment and should directly influence behavior, research evaluating how these associations develop over time is scant. Several studies have evaluated alcohol associations as prospective predictors of drinking or changes in drinking (e.g. Wiers et al., 2002; Thush and Wiers, 2007; Gray et al., 2011; O’Connor and Colder, 2015; Lindgren et al., 2016a) but only one has addressed the relationship between changes in alcohol associations and changes in drinking. Colder et al. (2014) evaluated alcohol valence (good/bad) associations and alcohol consumption in a sample of 10–12 year-olds for 3 years. They found that alcohol associations became more neutral over time (initially, they were negatively valenced) and that alcohol consumption increased; however, these changes were not significantly associated with one another. These findings suggest that key tenets of implicit cognition theories—namely that changes in implicit cognitive processes should be associated with changes in behavior and vice versa—do not hold during this early developmental period. The early college/university years in the USA (typically age 18–20) is an important developmental period and has several characteristics that make it a good candidate for evaluating whether there is a bi-directional relationship between changes in drinking and changes in alcohol associations. The first 2 years are a time when drinking is becoming established (a sizeable proportion of college students transition from abstinence to drinking when they begin college, see Fromme et al., 2008) and escalating (drinking is at its peak during the college years; see Naimi et al., 2003). These years and the US college environment are also associated with increased access to alcohol (likely due to the drinking age in the US being 21 years and greater independence from being away from home), and there are high rates of severe, alcohol-related negative consequences, including emergency room admissions, blackouts and sexual assaults (Merrill and Carey, 2016). Consequently, these years are likely a time when alcohol associations increase in strength and when those increases should be associated with increases in subsequent drinking behavior (Wiers et al., 2007). Further, increases in or initiation of drinking behavior should result in increases in alcohol associations because direct experience should strengthen implicit cognition processes. Key associations to assess during this developmental period are drinking identity, excitement about alcohol and alcohol approach associations. These associations predicted drinking in college students in previous studies and link to psychological theories of drinking. First, drinking identity associations—associations with the self and being a drinker—link to theoretical models that suggest that identity (or self-concept) influences alcohol use and misuse and smoking (Lindgren et al., 2016b), and cessation and treatment (Dingle et al., 2015; Frings and Albery, 2015; Tombor et al., 2015). These models stem from social psychological theories that suggest that groups can be powerful sources of identity and can motivate individuals to behave consistently with group norms (e.g. Tajfel and Turner, 1979). Second, alcohol excite associations are thought to reflect motivations for why individuals drink. They also can be thought of as implicit counterparts to enhancement drinking expectancies and motives, which are consistent, proximal predictors of drinking (see Cox and Klinger, 1988; Cooper et al., 1995). Third, alcohol approach associations link to theoretical models suggesting that substance-related cues can elicit an appetitive response to approach and consume alcohol that, with repeated and increased use, become increasingly automatic and result in continued, compulsive use despite negative consequences (see Baker et al., 2004; Berridge et al., 2009). Empirically, all three associations predict drinking but are only weakly correlated with one another (Lindgren et al., 2013a; 2016a); together, they provide a wide-ranging but non-redundant evaluation of key associations for this age group. Study overview Study data derive from a large, longitudinal study of implicit and explicit measures as predictors of drinking in a sample of first- and second-year university students (Lindgren et al., 2016a). In this paper, we focus exclusively on implicit measures of alcohol associations to test a tenet of the dual process model and implicit cognitive theory that had not been tested in a young adult sample. Specifically, we evaluated the hypothesis that changes in alcohol associations (drinking identity, alcohol excite, and alcohol approach) would be positively associated with changes in drinking and vice versa. METHODS Procedures The university’s Institutional Review Board approved all procedures. Full-time students aged 18–20, who were in their first or second year, were approached at the beginning of the academic year via e-mail to participate. Participants completed each 50-min assessment online (assessments were hosted on the Project Implicit platform via a private study url). Each assessment included informed consent, three Implicit Association Tests (IATs; Greenwald et al., 1998) to measure alcohol associations and self-report measures. There were eight assessments, occurring at 3-month intervals. Compensation was $25 for the first three assessments, $30 for the last five assessments, and bonuses of $5 upon completion of all of the first four assessments and $10 for all of the final four. A $5 incentive was added to the last assessment for those ineligible for the $10 bonus. See Lindgren et al. (2016a) for the full description of study procedures and additional detail about study participants. Participants Participants (N = 506, 57% female, age: M = 18.58, SD = 0.69) were undergraduates at a large public university in the Pacific Northwest. Fifty-two percent identified as White, 32% as Asian American, 11% as multiracial and the remaining 5% as Black or African American, American Indian or Alaskan Native, unknown or declined to answer. Eight percent of participants identified as Hispanic or Latino. Retention rates for Time 2 through Time 8 were 90%, 76%, 76%, 77%, 72%, 67% and 66%. Because both initiation and escalation of drinking were of interest, participants did not have to currently drink alcohol to be eligible for the study. Consistent with other studies of early college students (see Fromme et al., 2008), 50% of participants reported no consumption at the first assessment, dropping to 24% at the last assessment. Measures Alcohol associations Three variants of the IAT (Greenwald et al., 1998) evaluated drinking identity, alcohol excite and alcohol approach associations. The IAT measures reaction times when classifying stimuli into categories as quickly as possible. Two target and two attribute categories are placed in pairs on the screen. In the drinking identity IAT, for example, participants classify stimuli representing the target categories ‘me’ and ‘not me’ and the attribute categories ‘drinker’ and ‘non-drinker.’ During each IAT trial, a stimulus appears at the center of the screen. Participants must classify it according to the categories displayed on the left or right side of the screen using a designated key (e for left and i for right). If an error is made, participants must correct it before proceeding to the next trial (no time penalty is applied to error trials and the reaction time for the corrected response is included in IAT scores). Each IAT includes seven blocks: three practice blocks and four test blocks. Practice blocks (1, 2 and 5) include either the target or attribute categories on each side of the screen. For example, Block 1 of the drinking identity IAT might start with ‘me’ on the left and ‘not me’ on the right, and participants classify stimuli representing those categories. Test blocks (3, 4, 6 and 7) pair one target category with one attribute category on each side of the screen. For example, Blocks 3 and 4 of the drinking identity IAT might pair ‘drinker’ with ‘me’ on the left and ‘non-drinker’ with ‘not me’ on the right. Thus, items fitting either the ‘drinker’ or ‘me’ categories are classified on the left, and items fitting either ‘non-drinker’ or ‘not me’ are classified on the right. The pairings would be reversed in Blocks 6 and 7: ‘drinker’ would be paired with ‘not me’ and ‘non-drinker’ would be paired with ‘me.’ The order in which category pairings are presented is counterbalanced across participants. Participants’ reaction times classifying stimuli in Blocks 3, 4, 6 and 7 (i.e. the paired category blocks) were used to calculate IAT scores following the D-score algorithm (Greenwald et al., 2003). The D-score is a standardized difference score that reflects a participant’s average reaction time for trials in each set of paired blocks (i.e. average reaction time for Blocks 3 and 4 is subtracted from Blocks 6 and 7). Scores were calculated such that higher scores indicated stronger associations (faster reaction times) with the categories in the IAT’s name. For example, higher scores on the drinking identity IAT indicated faster reaction times when drinker and me are paired and thus a stronger association between me and drinking or a stronger drinker identity. The order in which the IATs were presented was randomized across participants and interspersed among the self-report measures to reduce fatigue. As per Nosek and colleagues’ (2007) recommendations for screening, IAT scores were excluded when 10% or more trials were faster than 300 ms or when 30% or more trials included errors. Such exclusions rarely exceeded 10% across time points. Internal consistency for the IAT was calculated by correlating two D-scores: one for Blocks 3 and 6 and one for Blocks 4 and 7 (see Greenwald et al., 2003). Consistencies typically range from 0.5 to 0.6 for these IATs (see Lindgren et al., 2013a), rs: drinking identity = 0.58, alcohol approach = 0.55, alcohol excite = 0.57. IAT stimuli are reported in Lindgren et al. (2016a). Alcohol consumption The Daily Drinking Questionnaire (Collins et al., 1985) assessed participants’ self-reported alcohol consumption on each day of a typical week in the last 3 months. US standard drink equivalencies were provided. Following NIAAA guidelines (2017), participants were classified into three risk categories for developing an alcohol use disorder. Non-drinkers consumed zero drinks per week. Low-risk men and women consumed no more than 14/7 drinks per week and no more than 4 or 3 drinks on a single day, respectively. High risk men and women consumed more than 14 or 7 drinks per week or 4 or 3 drinks on a single day, respectively. Data Analysis We used cross-lagged panel models (Newsom, 2015) to examine the relationships among the alcohol associations (IAT scores) and the drinking risk categories. The IATs were treated as continuous variables and drinking category was treated as an ordered categorical variable with three levels (1 = No Risk, 2 = Low Risk, 3 = High Risk). We used drinking category as the outcome because it is not currently possible to implement a cross-lagged model with a zero-inflated negative binomial outcome. Although cross-lagged models would be possible by treating the drinking outcome as normally distributed, doing so is problematic generally (see Atkins et al., 2013). Moreover, the distribution of the drinking in this sample has both a large number of zeros and is positively skewed. Consequently, we chose a middle-ground option where drinking is divided into categories of low and high risk. These categories were put forth by the US National Institute on Alcohol Abuse and Alcoholism (see NIAAA, 2017) as clinically, relevant distinctions. Further, one can use a latent variable formulation of categorical variables that allows them to be used in cross-lagged models (B. O. Muthén and Asparouhov, 2002). The path model for the drinking identity IAT and drinking is depicted in Fig. 1; we used identical models for the alcohol approach and excite IATs. As indicated in Fig. 1, we estimated autoregressive and cross-lagged parameters. We estimated four key relationships in all models. For example, in the drinking identity model, the parameters were (a) an autoregressive relationship between identity at Time t and identity at Time t – 1; (b) an autoregressive relationship between drinking at Time t and drinking at Time t – 1; (c) a cross-lagged relationship between identity at Time t and drinking at Time t – 1; and (d) a cross-lagged relationship between drinking at Time t and identity at Time t – 1. Beginning with Time 2, we constrained the estimates to be constant across time because we did not have any predictions about time-related patterns. Further, from Time 2 onward the parameters represent relationships controlling for previous time points whereas from Time 1 to Time 2, there is no previous time point. Consequently, we estimated unique autoregressive and cross-lagged parameters from Time 1 to Time 2. The model also included birth sex and drinking history prior to Time 1 (1 = history of alcohol consumption, 0 = no history of alcohol consumption) as predictors of the IATs and drinking risk (Birth sex was controlled for in analyses due to long-standing evidence of sex differences in consumption among US college students (see Schulenberg et al., 2017) and evidence of sex differences in the distribution of men and women among the three drinking risk categories in six (of the eight) time points. The overall pattern of results does not differ if birth sex is not controlled for. Prior history of drinking was added as a covariate to test whether initial changes IAT scores were associated with bigger changes in risk (and vice versa) amongst those who were completely new to drinking. We thank an anonymous reviewer for this suggestion.). As with the other parameters, we estimated unique values for the covariates at Time 1 and then constrained them to be equal across time beginning at Time 2 (i.e. one value for each covariate from Time 2 to Time 8). Fig. 1. View largeDownload slide Path diagram for the cross-lagged model relating drinking identity and drinking risk. Comparable models for approach and excite were also estimated. All models included birth sex and drinking history prior to Time 1 as covariates as well as covariance among the residuals at each time point. Iden = Implicit Drinking Identity; Risk = Drinking Risk Category. Fig. 1. View largeDownload slide Path diagram for the cross-lagged model relating drinking identity and drinking risk. Comparable models for approach and excite were also estimated. All models included birth sex and drinking history prior to Time 1 as covariates as well as covariance among the residuals at each time point. Iden = Implicit Drinking Identity; Risk = Drinking Risk Category. To fit this model, a latent variable formulation of drinking risk had to be used. The latent variable approach treats the observed categorical variable as a coarse measurement of an underlying continuous variable—in this instance, drinking category is a coarse measure of continuous drinking risk (Muthén and Asparouhov, 2002). Continuous drinking risk is divided into three parts, separated by two thresholds. Participants below the first threshold are in the no risk category; participants between the first and second thresholds are in the low risk category and participants above the second threshold are in the high-risk category. The cross-lagged models were estimated using Mplus (Muthén and Muthén, 1998–2017, Version 8.0) with the weighted least squares (WLSMV) estimator. RESULTS Figure 2 shows the distribution of each IAT over time and Fig. 3 shows the distribution of the drinking risk categories over time for the entire sample. See Supplemental Table 1 for descriptive statistics and zero-order correlations for study baseline variables. At the sample level, the IATs were relatively static across time. For example, median values varied slightly across time, but generally stayed just below zero. Likewise, the range of IAT scores stayed fairly consistent across time. The distribution of risk categories changed more over time. Specifically, the number of people in the no risk category decreased substantially from Time 1 to Time 8 whereas the low risk and high-risk categories fluctuated. Fig. 2. View largeDownload slide Distribution of each IAT over time. Identity refers to the drinking identity IAT, approach to the alcohol approach IAT and excite to the alcohol excite IAT. IAT values are D-scores. IAT = Implicit Association Test. Fig. 2. View largeDownload slide Distribution of each IAT over time. Identity refers to the drinking identity IAT, approach to the alcohol approach IAT and excite to the alcohol excite IAT. IAT values are D-scores. IAT = Implicit Association Test. Fig. 3. View largeDownload slide Distribution of drinking category over time. No risk, low risk, and high risk refer to NIAAA drinking categories. No risk drinkers consumed zero drinks per week. Low-risk men/women consumed no more than 14/7 drinks per week and no more than 4/3 drinks on a single day. High risk men/women consumed more than 14/7 drinks per week or 4/3 drinks on a single day. Fig. 3. View largeDownload slide Distribution of drinking category over time. No risk, low risk, and high risk refer to NIAAA drinking categories. No risk drinkers consumed zero drinks per week. Low-risk men/women consumed no more than 14/7 drinks per week and no more than 4/3 drinks on a single day. High risk men/women consumed more than 14/7 drinks per week or 4/3 drinks on a single day. Table 1 presents the parameter estimates and model fit statistics for the cross-lagged models for each IAT. Model fit was good for all three models: the comparative fit index ranged between 0.95 and 0.97, the Tucker–Lewis index ranged between 0.94 and 0.96, and the root mean square error of approximation ranged between 0.06 and 0.07. Findings from all three models suggested that changes in IAT scores predicted later changes in drinking risk and that changes in drinking risk predicted later changes in IAT scores, holding constant birth sex and history of drinking. Table 1. Results of cross-lagged panel models testing changes in implicit alcohol associations and changes in drinking risk category Identity Approach Excite Time 1  IAT1 → IAT2 βˆ1 0.45*** 0.43*** 0.44***  Risk1 → Risk2 βˆ2 1.88*** 1.90*** 1.79***  Risk1 → IAT2 βˆ3 0.06** 0.03 0.06**  IAT1 → Risk2 βˆ4 −0.42 −0.49 0.12  Birth sex → IAT1 βˆ5 −0.14** −0.11** −0.12**  Birth sex → Risk1 βˆ6 0.06 0.06 0.09  History → IAT1 βˆ7 0.25*** 0.22*** 0.2**  History → Risk1 βˆ8 1.84** 1.83*** 1.83*** Time 2–8  IATt−1 → IATt βˆ9 0.76*** 0.74*** 0.78***  Riskt−1 → Riskt βˆ10 0.9*** 0.91*** 0.92***  Riskt−1 → IATt βˆ11 0.02*** 0.01*** 0.01***  IATt−1 → Riskt βˆ12 0.35** 0.45* 0.39***  Birth sex → IATt βˆ13 −0.04*** −0.02* 0.001  Birth sex → Riskt βˆ14 0.17** 0.24* 0.14**  History → IATt βˆ15 −0.04* −0.01 0.02  History → Riskt βˆ16 0.25* 0.16** 0.24* Model fit CFI 0.97 0.95 0.96 TLI 0.96 0.94 0.96 RMSEA 0.06 (90% CI = 0.05, 0.07) 0.07 (90% CI = 0.06, 0.08) 0.06 (90% CI = 0.05, 0.07) Identity Approach Excite Time 1  IAT1 → IAT2 βˆ1 0.45*** 0.43*** 0.44***  Risk1 → Risk2 βˆ2 1.88*** 1.90*** 1.79***  Risk1 → IAT2 βˆ3 0.06** 0.03 0.06**  IAT1 → Risk2 βˆ4 −0.42 −0.49 0.12  Birth sex → IAT1 βˆ5 −0.14** −0.11** −0.12**  Birth sex → Risk1 βˆ6 0.06 0.06 0.09  History → IAT1 βˆ7 0.25*** 0.22*** 0.2**  History → Risk1 βˆ8 1.84** 1.83*** 1.83*** Time 2–8  IATt−1 → IATt βˆ9 0.76*** 0.74*** 0.78***  Riskt−1 → Riskt βˆ10 0.9*** 0.91*** 0.92***  Riskt−1 → IATt βˆ11 0.02*** 0.01*** 0.01***  IATt−1 → Riskt βˆ12 0.35** 0.45* 0.39***  Birth sex → IATt βˆ13 −0.04*** −0.02* 0.001  Birth sex → Riskt βˆ14 0.17** 0.24* 0.14**  History → IATt βˆ15 −0.04* −0.01 0.02  History → Riskt βˆ16 0.25* 0.16** 0.24* Model fit CFI 0.97 0.95 0.96 TLI 0.96 0.94 0.96 RMSEA 0.06 (90% CI = 0.05, 0.07) 0.07 (90% CI = 0.06, 0.08) 0.06 (90% CI = 0.05, 0.07) Note. *P < 0.05; **P < 0.01; ***P < 0.001; N = 501 (5 of the 506 participants did not provide a response to the item assessing lifetime drinking history); IAT, Implicit Association Test; Risk, drinking risk category (higher categories = higher risk of alcohol use disorder); History, drinking history prior to Time 1; CFI, Comparative Fit Index; TLI, Tucker–Lewis Index; RMSEA, root mean square error of approximation; CI, confidence interval. Table 1. Results of cross-lagged panel models testing changes in implicit alcohol associations and changes in drinking risk category Identity Approach Excite Time 1  IAT1 → IAT2 βˆ1 0.45*** 0.43*** 0.44***  Risk1 → Risk2 βˆ2 1.88*** 1.90*** 1.79***  Risk1 → IAT2 βˆ3 0.06** 0.03 0.06**  IAT1 → Risk2 βˆ4 −0.42 −0.49 0.12  Birth sex → IAT1 βˆ5 −0.14** −0.11** −0.12**  Birth sex → Risk1 βˆ6 0.06 0.06 0.09  History → IAT1 βˆ7 0.25*** 0.22*** 0.2**  History → Risk1 βˆ8 1.84** 1.83*** 1.83*** Time 2–8  IATt−1 → IATt βˆ9 0.76*** 0.74*** 0.78***  Riskt−1 → Riskt βˆ10 0.9*** 0.91*** 0.92***  Riskt−1 → IATt βˆ11 0.02*** 0.01*** 0.01***  IATt−1 → Riskt βˆ12 0.35** 0.45* 0.39***  Birth sex → IATt βˆ13 −0.04*** −0.02* 0.001  Birth sex → Riskt βˆ14 0.17** 0.24* 0.14**  History → IATt βˆ15 −0.04* −0.01 0.02  History → Riskt βˆ16 0.25* 0.16** 0.24* Model fit CFI 0.97 0.95 0.96 TLI 0.96 0.94 0.96 RMSEA 0.06 (90% CI = 0.05, 0.07) 0.07 (90% CI = 0.06, 0.08) 0.06 (90% CI = 0.05, 0.07) Identity Approach Excite Time 1  IAT1 → IAT2 βˆ1 0.45*** 0.43*** 0.44***  Risk1 → Risk2 βˆ2 1.88*** 1.90*** 1.79***  Risk1 → IAT2 βˆ3 0.06** 0.03 0.06**  IAT1 → Risk2 βˆ4 −0.42 −0.49 0.12  Birth sex → IAT1 βˆ5 −0.14** −0.11** −0.12**  Birth sex → Risk1 βˆ6 0.06 0.06 0.09  History → IAT1 βˆ7 0.25*** 0.22*** 0.2**  History → Risk1 βˆ8 1.84** 1.83*** 1.83*** Time 2–8  IATt−1 → IATt βˆ9 0.76*** 0.74*** 0.78***  Riskt−1 → Riskt βˆ10 0.9*** 0.91*** 0.92***  Riskt−1 → IATt βˆ11 0.02*** 0.01*** 0.01***  IATt−1 → Riskt βˆ12 0.35** 0.45* 0.39***  Birth sex → IATt βˆ13 −0.04*** −0.02* 0.001  Birth sex → Riskt βˆ14 0.17** 0.24* 0.14**  History → IATt βˆ15 −0.04* −0.01 0.02  History → Riskt βˆ16 0.25* 0.16** 0.24* Model fit CFI 0.97 0.95 0.96 TLI 0.96 0.94 0.96 RMSEA 0.06 (90% CI = 0.05, 0.07) 0.07 (90% CI = 0.06, 0.08) 0.06 (90% CI = 0.05, 0.07) Note. *P < 0.05; **P < 0.01; ***P < 0.001; N = 501 (5 of the 506 participants did not provide a response to the item assessing lifetime drinking history); IAT, Implicit Association Test; Risk, drinking risk category (higher categories = higher risk of alcohol use disorder); History, drinking history prior to Time 1; CFI, Comparative Fit Index; TLI, Tucker–Lewis Index; RMSEA, root mean square error of approximation; CI, confidence interval. Drinking identity From Times 1 to 2, drinking risk predicted increases in drinking identity ( βˆ3 = 0.06, P < 0.01), holding constant Time 1 identity ( βˆ1 = 0.45, P < 0.001), birth sex ( βˆ5 = −0.14, P < 0.01) and history of drinking ( βˆ7 = 0.25, P < 0.01). In contrast, Time 1 identity did not predict Time 2 risk ( βˆ4 = −0.42, ns), holding constant Time 1 risk ( βˆ2 = 1.88, P < 0.001), birth sex ( βˆ6 = 0.09, ns) and history of drinking ( βˆ8 = 0.25, P < 0.01). As expected, from Times 2 to 8, increases in drinking identity were associated with increased drinking risk at the following time point ( βˆ12 = 0.35, P < 0.01), holding constant previous drinking risk changes ( βˆ10 = 0.9, P < 0.001), birth sex ( βˆ14 = 0.17, P < 0.01) and history of drinking ( βˆ16 = 0.25, P < 0.05). The model also suggested that increases in drinking risk were associated with increased identity at the following time point ( βˆ11 = 0.02, P < 0.001), holding constant previous identity changes ( βˆ9 = 0.76, P < 0.001), birth sex ( βˆ13 = −0.04, P < 0.001) and history of drinking ( βˆ15 = −0.04, P < 0.05). Alcohol approach From Times 1 to 2, drinking risk did not predict changes in alcohol approach or vice versa. In contrast, from Times 2 to 8, increases in alcohol approach were associated with increased risk at the following time point ( βˆ12 = 0.45, P < 0.05), holding constant previous drinking risk changes ( βˆ10 = 0.91, P < 0.001), birth sex ( βˆ14 = 0.24, P < 0.05) and history of drinking ( βˆ16 = 0.16, P < 0.01). Likewise, increases in drinking risk were associated with increased alcohol approach at the following time point ( βˆ11 = 0.01, P < 0.001), holding constant previous alcohol approach changes ( βˆ9 = 0.74, P < 0.001), birth sex ( βˆ13 = −0.02, P < 0.05) and history of drinking ( βˆ15 = −0.01, ns). Alcohol excite From Times 1 to 2, alcohol excite did not predict changes in drinking risk. However, risk did predict increases in alcohol excite ( βˆ3 = 0.06, P < 0.01), holding constant Time 1 alcohol excite ( βˆ1 = 0.44, P < 0.001), birth sex ( βˆ5 = −0.12, P < 0.05) and history of drinking ( βˆ7 = 0.2, P < 0.01). From Times 2 to 8, increases in alcohol excite were associated with increased drinking risk at the following time point ( βˆ12 = 0.39, P < 0.001), holding constant previous drinking risk changes ( βˆ10 = 0.92, P < 0.001), birth sex ( βˆ14 = 0.14, P < 0.01) and history of drinking ( βˆ16 = 0.24, P < 0.05). Likewise, increases in drinking risk were associated with increased alcohol excite ( βˆ11 = 0.01, P < 0.001), holding constant previous alcohol excite changes ( βˆ9 = 0.78, P < 0.001), birth sex ( βˆ13 = 0.001, ns) and history of drinking ( βˆ15 = 0.02, ns). DISCUSSION The study evaluated questions of change in alcohol associations and change in drinking risk in a sample of US students in their first or second year of college—a time of increased initiation and escalation of drinking and increased drinking risk. The study’s purpose was to test a key—yet rarely tested—tenet of implicit cognition theories; that increases in the strength of alcohol associations would be associated with subsequent increases in drinking and vice versa. This is the second study we know of to test for bi-directional relationships between changes in alcohol associations and changes in drinking and the first study to find evidence of such relationships. Bi-directional relationships were observed for all three associations in this study—drinking identity, alcohol excite and alcohol approach. Notably, study results differ from those of Colder et al. (2014), which found no evidence of reciprocal effects with a sample of US children and early adolescents. A related study that evaluated a slightly older sample of Dutch adolescents and focused on a different type of implicit cognitions (Janssen et al., 2015) also found no evidence of reciprocal effects. Collectively, findings from these studies may indicate the moderating influence of age, developmental stage, the university/college environment, availability of alcohol and/or one of many other factors that vary when examining drinking from a developmental perspective. Research that expands the current model to include such factors and other hypothesized moderators of impulsive processes (e.g. self-regulatory processes, see O’Connor and Colder, 2015) will be crucial for providing a more comprehensive understanding of when and why changes in alcohol associations become predictive of changes in drinking and vice versa. Implications The current results point to the importance of measuring change processes. Earlier analyses from the larger parent study (Lindgren et al., 2016a) did not find that a static measure of alcohol approach associations predicted change in drinking behaviors at the next time point (though drinking identity and alcohol excite associations were predictive), while the present study found that change in each of the alcohol associations was predictive. Although the analyses were not directly comparable in other ways, the distinct results raise the possibility that measuring absolute level of alcohol associations may only tell part of the story of individual risk. A dynamic assessment strategy may, instead, be needed to determine when an individual is becoming more at risk. More generally, although causal claims cannot be made with these data, results support assertions that targeting alcohol associations in prevention and intervention efforts may be helpful (see Wiers et al., 2006). Some progress has been made on this front via computer-based tasks that seek to modify alcohol associations (see Wiers et al., 2011), but these tasks have yielded null results in US college samples (see Lindgren et al., 2015) and a recent meta-analysis (Cristea et al., 2016) raised concerns about the tasks’ efficacy (though there is disagreement about how to interpret the negative findings; Wiers et al., in press). Taken together, this suggests the importance of developing strategies that can target alcohol associations in college student populations. Developing interventions that seek to strengthen the influence of slower, reflective processes and override the alcohol associations may also be helpful. Limitations and future directions Several limitations should be noted. First, the variables representing drinking risk, while based on NIAAA guidelines, were categorical and so are relatively coarse measures. Second, when using categorical variables in cross-lagged models, we needed to treat them as latent variables, meaning we could not make inferences about the probability of being in a given category. Third, the study sample, while relatively large, focuses on a single sample of undergraduates from a single university, and there was attrition over time. We also note that other alcohol associations (e.g. alcohol and valence [good/bad]) have been shown to predict drinking (Wiers et al., 2002) and could be evaluated in future studies. Finally, we note the lower split-half reliabilities of the IATs used in the study. While they are similar to the reliabilities observed when using these IATs in other studies (see Lindgren et al., 2013a) as well as IATs used in other studies (see Greenwald et al., 2003), IATs and implicit measures in general tend to have lower internal consistencies than explicit measures (see Bar-Anan and Nosek, 2014). This is a general challenge for the field. Despite this limitation, there are many examples of the IAT’s predictive validity generally (Bar-Anan and Nosek, 2014) and in alcohol research specifically (Reich et al., 2010; Lindgren et al., 2013b). CONCLUSION Despite these limitations, this study is an important step toward evaluating changes in alcohol associations and drinking. It is the first study to find a bi-directional relationship between increases in the strength of alcohol associations and increases in drinking risk in a sample of students in the early college years, and findings are consistent with theories of implicit cognition. They provide additional evidence that alcohol excite, alcohol approach and drinking identity are important factors to consider in cognitive models of alcohol use and misuse. SUPPLEMENTARY MATERIAL Supplementary data are available at Alcohol And Alcoholism online. ACKNOWLEDGEMENTS The authors affirm there are no conflicts of interest for this work. B.A.T has a significant financial interest in Project Implicit, Inc., which provided services for hosting data collection for this project FUNDING This research was supported by the National Institute on Alcohol Abuse and Alcoholism [Grant nos. R01AA021763 (PI: Lindgren) and R01AA024732 (PI: Lindgren)]. NIAAA had no role in the study design, collection, analysis or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication. CONFLICT OF INTEREST STATEMENT None declared. REFERENCES Atkins DC , Baldwin SA , Zheng C , et al. . ( 2013 ) A tutorial on count regression and zero-altered count models for longitudinal substance use data . Psychol Addict Behav 27 : 166 . https://doi.org/10.1037/a0029508.supp . Google Scholar CrossRef Search ADS PubMed Baker TB , Piper ME , McCarthy DE , et al. . ( 2004 ) Addiction motivation reformulated: an affective processing model of negative reinforcement . Psychol Rev 111 : 33 – 51 . https://doi.org/10.1037/0033-295×.111.1.33 . Google Scholar CrossRef Search ADS PubMed Bar-Anan Y , Nosek BA . ( 2014 ) A comparative investigation of seven indirect measures of social cognition . Behav Res Methods 46 : 668 – 88 . https://doi.org/10.3758/s13428-013-0410-6 . Google Scholar CrossRef Search ADS PubMed Berridge KC , Robinson TE , Aldridge JW . ( 2009 ) Dissecting components of reward: ‘liking’, ‘wanting’, and learning . Curr Opin Pharmacol 9 : 65 – 73 . https://doi.org/10.1016/j.coph.2008.12.014 . Google Scholar CrossRef Search ADS PubMed Colder CR , O’Connor RM , Read JP , et al. . ( 2014 ) Growth trajectories of alcohol information processing and associations with escalation of drinking in early adolescence . Psychol Addict Behav 28 : 659 – 70 . https://doi.org/10.1037/a0035271 . Google Scholar CrossRef Search ADS PubMed Collins RL , Parks GA , Marlatt GA . ( 1985 ) Social determinants of alcohol consumption: the effects of social interaction and model status on the self-administration of alcohol . J Consult Clin Psychol 53 : 189 – 200 . https://doi.org/10.1037/0022-006×.53.2.189 . Google Scholar CrossRef Search ADS PubMed Cooper ML , Frone MR , Russell M , et al. . ( 1995 ) Drinking to regulate positive and negative emotions: a motivational model of alcohol use . J Pers Soc Psychol 69 : 990 – 1005 . https://doi.org/10.1037/0022-3514.69.5.990 . Google Scholar CrossRef Search ADS PubMed Cox WM , Klinger E . ( 1988 ) A motivational model of alcohol use . J Abnorm Psychol 97 : 168 – 80 . https://doi.org/10.1037/0021-843×.97.2.168 . Google Scholar CrossRef Search ADS PubMed Cristea IA , Kok RN , Cuijpers P . ( 2016 ) The effectiveness of cognitive bias modification interventions for substance addictions: a meta-analysis . PLOS ONE 11 : e0162226 . https://doi.org/10.1371/journal.pone.0162226 . Google Scholar CrossRef Search ADS PubMed Dingle GA , Cruwys T , Frings D . ( 2015 ) Social identities as pathways into and out of addiction . Front Psychol 6 : 1795 . https://doi.org/10.3389/fpsyg.2015.01795 . Google Scholar CrossRef Search ADS PubMed Frings D , Albery IP . ( 2015 ) The social identity model of cessation maintenance: formulation and initial evidence . Addict Behav 44 : 35 – 42 . https://doi.org/10.1016/j.addbeh.2014.10.023 . Google Scholar CrossRef Search ADS PubMed Fromme K , Corbin WR , Kruse MI . ( 2008 ) Behavioral risks during the transition from high school to college . Dev Psychol 44 : 1497 – 1504 . Google Scholar CrossRef Search ADS PubMed Gray HM , LaPlante DA , Bannon BL , et al. . ( 2011 ) Development and validation of the Alcohol Identity Implicit Associations Test (AI-IAT) . Addict Behav 36 : 919 – 26 . doi:10.1016/j.addbeh.2011.05.003 . Google Scholar CrossRef Search ADS PubMed Greenwald AG , Banaji MR . ( 1995 ) Implicit social cognition: attitudes, self-esteem, and stereotypes . Psychol Rev 102 : 4 – 27 . http://dx.doi.org/10.1037/0033-295×.102.1.4 . Google Scholar CrossRef Search ADS PubMed Greenwald AG , McGhee DE , Schwartz JK . ( 1998 ) Measuring individual differences in implicit cognition: the implicit association test . J Pers Soc Psychol 74 : 1464 – 80 . http://dx.doi.org/10.1037/0022-3514.74.6.1464 . Google Scholar CrossRef Search ADS PubMed Greenwald AG , Nosek BA , Banaji MR . ( 2003 ) Understanding and using the Implicit Association Test: I. An improved scoring algorithm . J Pers Soc Psychol 85 : 197 – 216 . http://dx.doi.org/10.1037/0022-3514.85.2.197 . Google Scholar CrossRef Search ADS PubMed Janssen T , Larsen H , Vollebergh WAM , et al. . ( 2015 ) Longitudinal relations between cognitive bias and adolescent alcohol use . Addict Behav 44 : 51 – 7 . https://doi.org/10.1016/j.addbeh.2014.11.018 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Foster DW , Westgate EC , et al. . ( 2013 b) Implicit drinking identity: Drinker+me associations predict college student drinking consistently . Addict Behav 38 : 2163 – 6 . https://doi.org/10.1016/j.addbeh.2013.01.026 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Neighbors C , Gasser ML , et al. . ( 2016 b) A review of implicit and explicit substance self-concept as a predictor of alcohol and tobacco use and misuse . Am J Drug Alcohol Abuse 43 : 237 – 46 . https://doi.org/10.1080/00952990.2016.1229324 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Neighbors C , Teachman BA , et al. . ( 2013 a) I drink therefore I am: validating alcohol-related implicit association tests . Psychol Addict Behav 27 : 1 – 13 . https://doi.org/10.1037/a0027640 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Neighbors C , Teachman BA , et al. . ( 2016 a) Implicit alcohol associations, especially drinking identity, predict drinking over time . Health Psychol 35 : 908 – 18 . https://doi.org/10.1037/hea0000396 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Wiers RW , Teachman BA , et al. . ( 2015 ) Attempted training of alcohol approach and drinking identity associations in US undergraduate drinkers: null results from two studies . PLOS ONE 10 : e0134642 . https://doi.org/10.1371/journal.pone.0134642 . Google Scholar CrossRef Search ADS PubMed Merrill JE , Carey KB . ( 2016 ) Drinking over the lifespan . Alcohol Res 38 : 103 – 14 . Retrieved from http://www.arcr.niaaa.nih.gov/arcr/arcr381/article12.htm. Google Scholar PubMed Muthén B , Asparouhov T . ( 2002 ) Latent variable analysis with categorical outcomes: Multiple-group and growth modeling in Mplus. Mplus Web Notes: No. 4. www.statmodel.com Muthén LK , Muthén BO ( 1998 –2017) Mplus User’s Guide , 8th edn. Los Angeles, CA : Muthén &Muthén . Naimi TS , Brewer RD , Mokdad A , et al. . ( 2003 ) Binge drinking among US adults . J Am Med Assoc 289 : 70 – 5 . https//doi.org/10.1016/j.amepre.2009.09.039 . Google Scholar CrossRef Search ADS National Institute on Alcohol Abuse and Alcoholism . ( 2017 ) Drinking Levels Defined. Retrieved from NIAAA website: https://www.niaaa.nih.gov/alcohol-health/overview-alcohol-consumption/moderate-binge-drinking Newsom JT . ( 2015 ) Longitudinal Structural Equation Modeling: A Comprehensive Introduction . New York : Routledge . Nosek BA , Greenwald AG , Banaji MR . ( 2007 ) The Implicit Association Test at age 7: a methodological and conceptual review. In Bargh JA (ed) . Automatic Processes in Social Thinking and Behavior . New York : Psychology Press , 265 – 92 . Nosek BA , Hawkins CB , Frazier RS . ( 2011 ) Implicit social cognition: from measures to mechanisms . Trends Cogn Sci 15 : 152 – 9 . https://doi.org/10.1016/j.tics.2011.01.005 . Google Scholar CrossRef Search ADS PubMed O’Connor RM , Colder CR . ( 2015 ) The prospective joint effects of self-regulation and impulsive processes on early adolescence alcohol use . J Stud Alcohol Drugs 76 : 884 – 94 . https://doi.org/10.15288/jsad.2015.76.884 . Google Scholar CrossRef Search ADS PubMed Reich RR , Below MC , Goldman MS . ( 2010 ) Explicit and implicit measures of expectancy and related alcohol cognitions: a meta-analytic comparison . Psychol Addict Behav 24 : 13 – 25 . https://doi.org/10.1037/a0016556 . Google Scholar CrossRef Search ADS PubMed Schulenberg JE , Johnston LD , O’Malley PM , et al. . ( 2017 ). Monitoring the Future national survey results on drug use, 1975–2016: Volume II, College students and adults ages 19–55 . Ann Arbor: Institute for Social Research, The University of Michigan. http://monitoringthefuture.org/pubs.html#monographs . Stacy AW , Wiers RW . ( 2010 ) Implicit cognition and addiction: a tool for explaining paradoxical behavior . Annu Rev Clin Psychol 6 : 551 – 75 . https://doi.org/10.1146/annurev.clinpsy.121208.131444 . Google Scholar CrossRef Search ADS PubMed Tajfel H , Turner JC . ( 1979 ) An integrative theory of intergroup conflict. In Austin WG , Worchel S (eds) . The Social Psychology of Intergroup Relations . Monterey, CA : Brooks/Cole Publishing Company , 33 – 47 . Thush C , Wiers RW . ( 2007 ) Explicit and implicit alcohol-related cognitions and the prediction of future drinking in adolescents . Addict Behav 32 : 1367 – 83 . https://doi.org/10.1016/j.addbeh.2006.09.011 . Google Scholar CrossRef Search ADS PubMed Tombor I , Shahab L , Brown J , et al. . ( 2015 ) Does non-smoker identity following quitting predict long-term abstinence? Evidence from a population survey in England . Addict Behav 49 : 99 – 103 . http://dx.doi.org/10.1016/j.addbeh.2015.01.026 . Google Scholar CrossRef Search ADS Wiers RW , Bartholow BD , van den Wildenberg E , et al. . ( 2007 ) Automatic and controlled processes and the development of addictive behaviors in adolescents: a review and a model . Pharmacol, Biochem Behav 86 : 263 – 83 . https://doi.org/10.1016/j.pbb.2006.09.021 . Google Scholar CrossRef Search ADS Wiers RW , Boffo M , Field M . (in press) What’s in a trial? On the importance of distinguishing between experimental lab-studies and randomized controlled trials; the case of cognitive bias modification and alcohol use disorders . J Stud Alcohol Drugs . Wiers RW , Cox WM , Field M , et al. . ( 2006 ) The search for new ways to change implicit alcohol-related cognitions in heavy drinkers . Alcohol Clin Exp Res 30 : 320 – 31 . https://doi.org/10.1111/j.1530-0277.2006.00037.x . Google Scholar CrossRef Search ADS PubMed Wiers RW , Eberl C , Rinck M , et al. . ( 2011 ) Retraining automatic action tendencies changes alcoholic patients’ approach bias for alcohol and improves treatment outcome . Psychol Sci 22 : 490 – 7 . https://doi.org/10.1177/0956797611400615 . Google Scholar CrossRef Search ADS PubMed Wiers RW , van Woerden N , Smulders FTY , et al. . ( 2002 ) Implicit and explicit alcohol-related cognitions in heavy and light drinkers . J Abnorm Psychol 111 : 648 – 58 . https://doi.org/10.1037/0021-843×.111.4.648 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Medical Council on Alcohol and Oxford University Press. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Alcohol and Alcoholism Oxford University Press

Evaluating Within-Person Change in Implicit Measures of Alcohol Associations: Increases in Alcohol Associations Predict Increases in Drinking Risk and Vice Versa

Loading next page...
 
/lp/ou_press/evaluating-within-person-change-in-implicit-measures-of-alcohol-tpZfhzhMxj
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Medical Council on Alcohol and Oxford University Press. All rights reserved.
ISSN
0735-0414
eISSN
1464-3502
D.O.I.
10.1093/alcalc/agy012
Publisher site
See Article on Publisher Site

Abstract

Abstract Aims Implicit measures of alcohol associations (i.e. measures designed to assess associations that are fast/reflexive/impulsive) have received substantial research attention. Alcohol associations related to the self (drinking identity), the effects of alcohol (alcohol excite) and appetitive inclinations (alcohol approach) have been found to predict drinking cross-sectionally and over time. A critical next step in this line of research and the goal of this study is to evaluate whether increases in the strength of these associations predict increases in drinking and vice versa. These hypotheses were tested in a sample of first- and second-year US university students: a sample selected because this time period is associated with initiation and escalation of drinking, peak levels of alcohol consumption and severe alcohol-related negative consequences. Short summary This study’s purpose was to evaluate whether increases in the strength of alcohol associations with the self (drinking identity), excitement (alcohol excite) and approach (alcohol approach) as assessed by implicit measures predicted subsequent increases in drinking risk and vice versa using a longitudinal, university student sample. Results were consistent with hypotheses. Methods A sample of 506 students’ (57% women) alcohol associations and alcohol consumption were assessed every 3 months over a 2-year period. Participants’ consumption was converted to risk categories based on NIAAA’s criteria: non-drinkers, low-risk drinkers and high-risk drinkers. A series of cross-lagged panel models tested whether changes in alcohol associations predicted subsequent change in drinking risk (and vice versa). Results Across all three measures of alcohol associations, increases in the strength of alcohol associations were associated with subsequent increases in drinking risk and vice versa. Conclusion Results from this study indicate bi-directional relationships between increases in alcohol associations (drinking identity, alcohol excite and alcohol approach) and subsequent increases in drinking risk. Intervention and prevention efforts may benefit from targeting these associations. INTRODUCTION Implicit measures of alcohol associations (measures designed to assess associations that are fast/reflexive/impulsive) predict drinking behaviors cross-sectionally and prospectively, even after controlling for their explicit measure counterparts (i.e. measures assessing slower/more reflective cognitions; Reich et al., 2010). Recent studies have also found that alcohol associations can predict changes in drinking behavior (Thush and Wiers, 2007; Lindgren et al., 2016a). A critical next step is to evaluate the dynamic relationship between alcohol associations and consumption by examining whether changes in associations predict changes in consumption and vice versa. These dynamic relationships have only been studied in an early adolescent sample (Colder et al., 2014). Thus, we evaluated whether changes in three well-validated implicit measures of alcohol associations were associated with changes in alcohol consumption in a sample of first- and second-year US university students. Interest in alcohol associations is motivated by dual process models of alcohol misuse (e.g. Wiers et al., 2007), which point to the importance of implicit and explicit cognitive processes in predicting drinking. Alcohol associations are commonly described as reflecting the extent to which alcohol becomes associated with other constructs in memory (see Stacy and Wiers, 2010). They are typically assessed indirectly—often by computer-based measures of reaction time in which the underlying cognitive process and content is inferred (see Nosek et al., 2011). The current study makes use of such measures. Although general theories of implicit cognition (Greenwald and Banaji, 1995; Nosek et al., 2011) suggest that alcohol associations will strengthen in response to direct experience as well as to one’s environment and should directly influence behavior, research evaluating how these associations develop over time is scant. Several studies have evaluated alcohol associations as prospective predictors of drinking or changes in drinking (e.g. Wiers et al., 2002; Thush and Wiers, 2007; Gray et al., 2011; O’Connor and Colder, 2015; Lindgren et al., 2016a) but only one has addressed the relationship between changes in alcohol associations and changes in drinking. Colder et al. (2014) evaluated alcohol valence (good/bad) associations and alcohol consumption in a sample of 10–12 year-olds for 3 years. They found that alcohol associations became more neutral over time (initially, they were negatively valenced) and that alcohol consumption increased; however, these changes were not significantly associated with one another. These findings suggest that key tenets of implicit cognition theories—namely that changes in implicit cognitive processes should be associated with changes in behavior and vice versa—do not hold during this early developmental period. The early college/university years in the USA (typically age 18–20) is an important developmental period and has several characteristics that make it a good candidate for evaluating whether there is a bi-directional relationship between changes in drinking and changes in alcohol associations. The first 2 years are a time when drinking is becoming established (a sizeable proportion of college students transition from abstinence to drinking when they begin college, see Fromme et al., 2008) and escalating (drinking is at its peak during the college years; see Naimi et al., 2003). These years and the US college environment are also associated with increased access to alcohol (likely due to the drinking age in the US being 21 years and greater independence from being away from home), and there are high rates of severe, alcohol-related negative consequences, including emergency room admissions, blackouts and sexual assaults (Merrill and Carey, 2016). Consequently, these years are likely a time when alcohol associations increase in strength and when those increases should be associated with increases in subsequent drinking behavior (Wiers et al., 2007). Further, increases in or initiation of drinking behavior should result in increases in alcohol associations because direct experience should strengthen implicit cognition processes. Key associations to assess during this developmental period are drinking identity, excitement about alcohol and alcohol approach associations. These associations predicted drinking in college students in previous studies and link to psychological theories of drinking. First, drinking identity associations—associations with the self and being a drinker—link to theoretical models that suggest that identity (or self-concept) influences alcohol use and misuse and smoking (Lindgren et al., 2016b), and cessation and treatment (Dingle et al., 2015; Frings and Albery, 2015; Tombor et al., 2015). These models stem from social psychological theories that suggest that groups can be powerful sources of identity and can motivate individuals to behave consistently with group norms (e.g. Tajfel and Turner, 1979). Second, alcohol excite associations are thought to reflect motivations for why individuals drink. They also can be thought of as implicit counterparts to enhancement drinking expectancies and motives, which are consistent, proximal predictors of drinking (see Cox and Klinger, 1988; Cooper et al., 1995). Third, alcohol approach associations link to theoretical models suggesting that substance-related cues can elicit an appetitive response to approach and consume alcohol that, with repeated and increased use, become increasingly automatic and result in continued, compulsive use despite negative consequences (see Baker et al., 2004; Berridge et al., 2009). Empirically, all three associations predict drinking but are only weakly correlated with one another (Lindgren et al., 2013a; 2016a); together, they provide a wide-ranging but non-redundant evaluation of key associations for this age group. Study overview Study data derive from a large, longitudinal study of implicit and explicit measures as predictors of drinking in a sample of first- and second-year university students (Lindgren et al., 2016a). In this paper, we focus exclusively on implicit measures of alcohol associations to test a tenet of the dual process model and implicit cognitive theory that had not been tested in a young adult sample. Specifically, we evaluated the hypothesis that changes in alcohol associations (drinking identity, alcohol excite, and alcohol approach) would be positively associated with changes in drinking and vice versa. METHODS Procedures The university’s Institutional Review Board approved all procedures. Full-time students aged 18–20, who were in their first or second year, were approached at the beginning of the academic year via e-mail to participate. Participants completed each 50-min assessment online (assessments were hosted on the Project Implicit platform via a private study url). Each assessment included informed consent, three Implicit Association Tests (IATs; Greenwald et al., 1998) to measure alcohol associations and self-report measures. There were eight assessments, occurring at 3-month intervals. Compensation was $25 for the first three assessments, $30 for the last five assessments, and bonuses of $5 upon completion of all of the first four assessments and $10 for all of the final four. A $5 incentive was added to the last assessment for those ineligible for the $10 bonus. See Lindgren et al. (2016a) for the full description of study procedures and additional detail about study participants. Participants Participants (N = 506, 57% female, age: M = 18.58, SD = 0.69) were undergraduates at a large public university in the Pacific Northwest. Fifty-two percent identified as White, 32% as Asian American, 11% as multiracial and the remaining 5% as Black or African American, American Indian or Alaskan Native, unknown or declined to answer. Eight percent of participants identified as Hispanic or Latino. Retention rates for Time 2 through Time 8 were 90%, 76%, 76%, 77%, 72%, 67% and 66%. Because both initiation and escalation of drinking were of interest, participants did not have to currently drink alcohol to be eligible for the study. Consistent with other studies of early college students (see Fromme et al., 2008), 50% of participants reported no consumption at the first assessment, dropping to 24% at the last assessment. Measures Alcohol associations Three variants of the IAT (Greenwald et al., 1998) evaluated drinking identity, alcohol excite and alcohol approach associations. The IAT measures reaction times when classifying stimuli into categories as quickly as possible. Two target and two attribute categories are placed in pairs on the screen. In the drinking identity IAT, for example, participants classify stimuli representing the target categories ‘me’ and ‘not me’ and the attribute categories ‘drinker’ and ‘non-drinker.’ During each IAT trial, a stimulus appears at the center of the screen. Participants must classify it according to the categories displayed on the left or right side of the screen using a designated key (e for left and i for right). If an error is made, participants must correct it before proceeding to the next trial (no time penalty is applied to error trials and the reaction time for the corrected response is included in IAT scores). Each IAT includes seven blocks: three practice blocks and four test blocks. Practice blocks (1, 2 and 5) include either the target or attribute categories on each side of the screen. For example, Block 1 of the drinking identity IAT might start with ‘me’ on the left and ‘not me’ on the right, and participants classify stimuli representing those categories. Test blocks (3, 4, 6 and 7) pair one target category with one attribute category on each side of the screen. For example, Blocks 3 and 4 of the drinking identity IAT might pair ‘drinker’ with ‘me’ on the left and ‘non-drinker’ with ‘not me’ on the right. Thus, items fitting either the ‘drinker’ or ‘me’ categories are classified on the left, and items fitting either ‘non-drinker’ or ‘not me’ are classified on the right. The pairings would be reversed in Blocks 6 and 7: ‘drinker’ would be paired with ‘not me’ and ‘non-drinker’ would be paired with ‘me.’ The order in which category pairings are presented is counterbalanced across participants. Participants’ reaction times classifying stimuli in Blocks 3, 4, 6 and 7 (i.e. the paired category blocks) were used to calculate IAT scores following the D-score algorithm (Greenwald et al., 2003). The D-score is a standardized difference score that reflects a participant’s average reaction time for trials in each set of paired blocks (i.e. average reaction time for Blocks 3 and 4 is subtracted from Blocks 6 and 7). Scores were calculated such that higher scores indicated stronger associations (faster reaction times) with the categories in the IAT’s name. For example, higher scores on the drinking identity IAT indicated faster reaction times when drinker and me are paired and thus a stronger association between me and drinking or a stronger drinker identity. The order in which the IATs were presented was randomized across participants and interspersed among the self-report measures to reduce fatigue. As per Nosek and colleagues’ (2007) recommendations for screening, IAT scores were excluded when 10% or more trials were faster than 300 ms or when 30% or more trials included errors. Such exclusions rarely exceeded 10% across time points. Internal consistency for the IAT was calculated by correlating two D-scores: one for Blocks 3 and 6 and one for Blocks 4 and 7 (see Greenwald et al., 2003). Consistencies typically range from 0.5 to 0.6 for these IATs (see Lindgren et al., 2013a), rs: drinking identity = 0.58, alcohol approach = 0.55, alcohol excite = 0.57. IAT stimuli are reported in Lindgren et al. (2016a). Alcohol consumption The Daily Drinking Questionnaire (Collins et al., 1985) assessed participants’ self-reported alcohol consumption on each day of a typical week in the last 3 months. US standard drink equivalencies were provided. Following NIAAA guidelines (2017), participants were classified into three risk categories for developing an alcohol use disorder. Non-drinkers consumed zero drinks per week. Low-risk men and women consumed no more than 14/7 drinks per week and no more than 4 or 3 drinks on a single day, respectively. High risk men and women consumed more than 14 or 7 drinks per week or 4 or 3 drinks on a single day, respectively. Data Analysis We used cross-lagged panel models (Newsom, 2015) to examine the relationships among the alcohol associations (IAT scores) and the drinking risk categories. The IATs were treated as continuous variables and drinking category was treated as an ordered categorical variable with three levels (1 = No Risk, 2 = Low Risk, 3 = High Risk). We used drinking category as the outcome because it is not currently possible to implement a cross-lagged model with a zero-inflated negative binomial outcome. Although cross-lagged models would be possible by treating the drinking outcome as normally distributed, doing so is problematic generally (see Atkins et al., 2013). Moreover, the distribution of the drinking in this sample has both a large number of zeros and is positively skewed. Consequently, we chose a middle-ground option where drinking is divided into categories of low and high risk. These categories were put forth by the US National Institute on Alcohol Abuse and Alcoholism (see NIAAA, 2017) as clinically, relevant distinctions. Further, one can use a latent variable formulation of categorical variables that allows them to be used in cross-lagged models (B. O. Muthén and Asparouhov, 2002). The path model for the drinking identity IAT and drinking is depicted in Fig. 1; we used identical models for the alcohol approach and excite IATs. As indicated in Fig. 1, we estimated autoregressive and cross-lagged parameters. We estimated four key relationships in all models. For example, in the drinking identity model, the parameters were (a) an autoregressive relationship between identity at Time t and identity at Time t – 1; (b) an autoregressive relationship between drinking at Time t and drinking at Time t – 1; (c) a cross-lagged relationship between identity at Time t and drinking at Time t – 1; and (d) a cross-lagged relationship between drinking at Time t and identity at Time t – 1. Beginning with Time 2, we constrained the estimates to be constant across time because we did not have any predictions about time-related patterns. Further, from Time 2 onward the parameters represent relationships controlling for previous time points whereas from Time 1 to Time 2, there is no previous time point. Consequently, we estimated unique autoregressive and cross-lagged parameters from Time 1 to Time 2. The model also included birth sex and drinking history prior to Time 1 (1 = history of alcohol consumption, 0 = no history of alcohol consumption) as predictors of the IATs and drinking risk (Birth sex was controlled for in analyses due to long-standing evidence of sex differences in consumption among US college students (see Schulenberg et al., 2017) and evidence of sex differences in the distribution of men and women among the three drinking risk categories in six (of the eight) time points. The overall pattern of results does not differ if birth sex is not controlled for. Prior history of drinking was added as a covariate to test whether initial changes IAT scores were associated with bigger changes in risk (and vice versa) amongst those who were completely new to drinking. We thank an anonymous reviewer for this suggestion.). As with the other parameters, we estimated unique values for the covariates at Time 1 and then constrained them to be equal across time beginning at Time 2 (i.e. one value for each covariate from Time 2 to Time 8). Fig. 1. View largeDownload slide Path diagram for the cross-lagged model relating drinking identity and drinking risk. Comparable models for approach and excite were also estimated. All models included birth sex and drinking history prior to Time 1 as covariates as well as covariance among the residuals at each time point. Iden = Implicit Drinking Identity; Risk = Drinking Risk Category. Fig. 1. View largeDownload slide Path diagram for the cross-lagged model relating drinking identity and drinking risk. Comparable models for approach and excite were also estimated. All models included birth sex and drinking history prior to Time 1 as covariates as well as covariance among the residuals at each time point. Iden = Implicit Drinking Identity; Risk = Drinking Risk Category. To fit this model, a latent variable formulation of drinking risk had to be used. The latent variable approach treats the observed categorical variable as a coarse measurement of an underlying continuous variable—in this instance, drinking category is a coarse measure of continuous drinking risk (Muthén and Asparouhov, 2002). Continuous drinking risk is divided into three parts, separated by two thresholds. Participants below the first threshold are in the no risk category; participants between the first and second thresholds are in the low risk category and participants above the second threshold are in the high-risk category. The cross-lagged models were estimated using Mplus (Muthén and Muthén, 1998–2017, Version 8.0) with the weighted least squares (WLSMV) estimator. RESULTS Figure 2 shows the distribution of each IAT over time and Fig. 3 shows the distribution of the drinking risk categories over time for the entire sample. See Supplemental Table 1 for descriptive statistics and zero-order correlations for study baseline variables. At the sample level, the IATs were relatively static across time. For example, median values varied slightly across time, but generally stayed just below zero. Likewise, the range of IAT scores stayed fairly consistent across time. The distribution of risk categories changed more over time. Specifically, the number of people in the no risk category decreased substantially from Time 1 to Time 8 whereas the low risk and high-risk categories fluctuated. Fig. 2. View largeDownload slide Distribution of each IAT over time. Identity refers to the drinking identity IAT, approach to the alcohol approach IAT and excite to the alcohol excite IAT. IAT values are D-scores. IAT = Implicit Association Test. Fig. 2. View largeDownload slide Distribution of each IAT over time. Identity refers to the drinking identity IAT, approach to the alcohol approach IAT and excite to the alcohol excite IAT. IAT values are D-scores. IAT = Implicit Association Test. Fig. 3. View largeDownload slide Distribution of drinking category over time. No risk, low risk, and high risk refer to NIAAA drinking categories. No risk drinkers consumed zero drinks per week. Low-risk men/women consumed no more than 14/7 drinks per week and no more than 4/3 drinks on a single day. High risk men/women consumed more than 14/7 drinks per week or 4/3 drinks on a single day. Fig. 3. View largeDownload slide Distribution of drinking category over time. No risk, low risk, and high risk refer to NIAAA drinking categories. No risk drinkers consumed zero drinks per week. Low-risk men/women consumed no more than 14/7 drinks per week and no more than 4/3 drinks on a single day. High risk men/women consumed more than 14/7 drinks per week or 4/3 drinks on a single day. Table 1 presents the parameter estimates and model fit statistics for the cross-lagged models for each IAT. Model fit was good for all three models: the comparative fit index ranged between 0.95 and 0.97, the Tucker–Lewis index ranged between 0.94 and 0.96, and the root mean square error of approximation ranged between 0.06 and 0.07. Findings from all three models suggested that changes in IAT scores predicted later changes in drinking risk and that changes in drinking risk predicted later changes in IAT scores, holding constant birth sex and history of drinking. Table 1. Results of cross-lagged panel models testing changes in implicit alcohol associations and changes in drinking risk category Identity Approach Excite Time 1  IAT1 → IAT2 βˆ1 0.45*** 0.43*** 0.44***  Risk1 → Risk2 βˆ2 1.88*** 1.90*** 1.79***  Risk1 → IAT2 βˆ3 0.06** 0.03 0.06**  IAT1 → Risk2 βˆ4 −0.42 −0.49 0.12  Birth sex → IAT1 βˆ5 −0.14** −0.11** −0.12**  Birth sex → Risk1 βˆ6 0.06 0.06 0.09  History → IAT1 βˆ7 0.25*** 0.22*** 0.2**  History → Risk1 βˆ8 1.84** 1.83*** 1.83*** Time 2–8  IATt−1 → IATt βˆ9 0.76*** 0.74*** 0.78***  Riskt−1 → Riskt βˆ10 0.9*** 0.91*** 0.92***  Riskt−1 → IATt βˆ11 0.02*** 0.01*** 0.01***  IATt−1 → Riskt βˆ12 0.35** 0.45* 0.39***  Birth sex → IATt βˆ13 −0.04*** −0.02* 0.001  Birth sex → Riskt βˆ14 0.17** 0.24* 0.14**  History → IATt βˆ15 −0.04* −0.01 0.02  History → Riskt βˆ16 0.25* 0.16** 0.24* Model fit CFI 0.97 0.95 0.96 TLI 0.96 0.94 0.96 RMSEA 0.06 (90% CI = 0.05, 0.07) 0.07 (90% CI = 0.06, 0.08) 0.06 (90% CI = 0.05, 0.07) Identity Approach Excite Time 1  IAT1 → IAT2 βˆ1 0.45*** 0.43*** 0.44***  Risk1 → Risk2 βˆ2 1.88*** 1.90*** 1.79***  Risk1 → IAT2 βˆ3 0.06** 0.03 0.06**  IAT1 → Risk2 βˆ4 −0.42 −0.49 0.12  Birth sex → IAT1 βˆ5 −0.14** −0.11** −0.12**  Birth sex → Risk1 βˆ6 0.06 0.06 0.09  History → IAT1 βˆ7 0.25*** 0.22*** 0.2**  History → Risk1 βˆ8 1.84** 1.83*** 1.83*** Time 2–8  IATt−1 → IATt βˆ9 0.76*** 0.74*** 0.78***  Riskt−1 → Riskt βˆ10 0.9*** 0.91*** 0.92***  Riskt−1 → IATt βˆ11 0.02*** 0.01*** 0.01***  IATt−1 → Riskt βˆ12 0.35** 0.45* 0.39***  Birth sex → IATt βˆ13 −0.04*** −0.02* 0.001  Birth sex → Riskt βˆ14 0.17** 0.24* 0.14**  History → IATt βˆ15 −0.04* −0.01 0.02  History → Riskt βˆ16 0.25* 0.16** 0.24* Model fit CFI 0.97 0.95 0.96 TLI 0.96 0.94 0.96 RMSEA 0.06 (90% CI = 0.05, 0.07) 0.07 (90% CI = 0.06, 0.08) 0.06 (90% CI = 0.05, 0.07) Note. *P < 0.05; **P < 0.01; ***P < 0.001; N = 501 (5 of the 506 participants did not provide a response to the item assessing lifetime drinking history); IAT, Implicit Association Test; Risk, drinking risk category (higher categories = higher risk of alcohol use disorder); History, drinking history prior to Time 1; CFI, Comparative Fit Index; TLI, Tucker–Lewis Index; RMSEA, root mean square error of approximation; CI, confidence interval. Table 1. Results of cross-lagged panel models testing changes in implicit alcohol associations and changes in drinking risk category Identity Approach Excite Time 1  IAT1 → IAT2 βˆ1 0.45*** 0.43*** 0.44***  Risk1 → Risk2 βˆ2 1.88*** 1.90*** 1.79***  Risk1 → IAT2 βˆ3 0.06** 0.03 0.06**  IAT1 → Risk2 βˆ4 −0.42 −0.49 0.12  Birth sex → IAT1 βˆ5 −0.14** −0.11** −0.12**  Birth sex → Risk1 βˆ6 0.06 0.06 0.09  History → IAT1 βˆ7 0.25*** 0.22*** 0.2**  History → Risk1 βˆ8 1.84** 1.83*** 1.83*** Time 2–8  IATt−1 → IATt βˆ9 0.76*** 0.74*** 0.78***  Riskt−1 → Riskt βˆ10 0.9*** 0.91*** 0.92***  Riskt−1 → IATt βˆ11 0.02*** 0.01*** 0.01***  IATt−1 → Riskt βˆ12 0.35** 0.45* 0.39***  Birth sex → IATt βˆ13 −0.04*** −0.02* 0.001  Birth sex → Riskt βˆ14 0.17** 0.24* 0.14**  History → IATt βˆ15 −0.04* −0.01 0.02  History → Riskt βˆ16 0.25* 0.16** 0.24* Model fit CFI 0.97 0.95 0.96 TLI 0.96 0.94 0.96 RMSEA 0.06 (90% CI = 0.05, 0.07) 0.07 (90% CI = 0.06, 0.08) 0.06 (90% CI = 0.05, 0.07) Identity Approach Excite Time 1  IAT1 → IAT2 βˆ1 0.45*** 0.43*** 0.44***  Risk1 → Risk2 βˆ2 1.88*** 1.90*** 1.79***  Risk1 → IAT2 βˆ3 0.06** 0.03 0.06**  IAT1 → Risk2 βˆ4 −0.42 −0.49 0.12  Birth sex → IAT1 βˆ5 −0.14** −0.11** −0.12**  Birth sex → Risk1 βˆ6 0.06 0.06 0.09  History → IAT1 βˆ7 0.25*** 0.22*** 0.2**  History → Risk1 βˆ8 1.84** 1.83*** 1.83*** Time 2–8  IATt−1 → IATt βˆ9 0.76*** 0.74*** 0.78***  Riskt−1 → Riskt βˆ10 0.9*** 0.91*** 0.92***  Riskt−1 → IATt βˆ11 0.02*** 0.01*** 0.01***  IATt−1 → Riskt βˆ12 0.35** 0.45* 0.39***  Birth sex → IATt βˆ13 −0.04*** −0.02* 0.001  Birth sex → Riskt βˆ14 0.17** 0.24* 0.14**  History → IATt βˆ15 −0.04* −0.01 0.02  History → Riskt βˆ16 0.25* 0.16** 0.24* Model fit CFI 0.97 0.95 0.96 TLI 0.96 0.94 0.96 RMSEA 0.06 (90% CI = 0.05, 0.07) 0.07 (90% CI = 0.06, 0.08) 0.06 (90% CI = 0.05, 0.07) Note. *P < 0.05; **P < 0.01; ***P < 0.001; N = 501 (5 of the 506 participants did not provide a response to the item assessing lifetime drinking history); IAT, Implicit Association Test; Risk, drinking risk category (higher categories = higher risk of alcohol use disorder); History, drinking history prior to Time 1; CFI, Comparative Fit Index; TLI, Tucker–Lewis Index; RMSEA, root mean square error of approximation; CI, confidence interval. Drinking identity From Times 1 to 2, drinking risk predicted increases in drinking identity ( βˆ3 = 0.06, P < 0.01), holding constant Time 1 identity ( βˆ1 = 0.45, P < 0.001), birth sex ( βˆ5 = −0.14, P < 0.01) and history of drinking ( βˆ7 = 0.25, P < 0.01). In contrast, Time 1 identity did not predict Time 2 risk ( βˆ4 = −0.42, ns), holding constant Time 1 risk ( βˆ2 = 1.88, P < 0.001), birth sex ( βˆ6 = 0.09, ns) and history of drinking ( βˆ8 = 0.25, P < 0.01). As expected, from Times 2 to 8, increases in drinking identity were associated with increased drinking risk at the following time point ( βˆ12 = 0.35, P < 0.01), holding constant previous drinking risk changes ( βˆ10 = 0.9, P < 0.001), birth sex ( βˆ14 = 0.17, P < 0.01) and history of drinking ( βˆ16 = 0.25, P < 0.05). The model also suggested that increases in drinking risk were associated with increased identity at the following time point ( βˆ11 = 0.02, P < 0.001), holding constant previous identity changes ( βˆ9 = 0.76, P < 0.001), birth sex ( βˆ13 = −0.04, P < 0.001) and history of drinking ( βˆ15 = −0.04, P < 0.05). Alcohol approach From Times 1 to 2, drinking risk did not predict changes in alcohol approach or vice versa. In contrast, from Times 2 to 8, increases in alcohol approach were associated with increased risk at the following time point ( βˆ12 = 0.45, P < 0.05), holding constant previous drinking risk changes ( βˆ10 = 0.91, P < 0.001), birth sex ( βˆ14 = 0.24, P < 0.05) and history of drinking ( βˆ16 = 0.16, P < 0.01). Likewise, increases in drinking risk were associated with increased alcohol approach at the following time point ( βˆ11 = 0.01, P < 0.001), holding constant previous alcohol approach changes ( βˆ9 = 0.74, P < 0.001), birth sex ( βˆ13 = −0.02, P < 0.05) and history of drinking ( βˆ15 = −0.01, ns). Alcohol excite From Times 1 to 2, alcohol excite did not predict changes in drinking risk. However, risk did predict increases in alcohol excite ( βˆ3 = 0.06, P < 0.01), holding constant Time 1 alcohol excite ( βˆ1 = 0.44, P < 0.001), birth sex ( βˆ5 = −0.12, P < 0.05) and history of drinking ( βˆ7 = 0.2, P < 0.01). From Times 2 to 8, increases in alcohol excite were associated with increased drinking risk at the following time point ( βˆ12 = 0.39, P < 0.001), holding constant previous drinking risk changes ( βˆ10 = 0.92, P < 0.001), birth sex ( βˆ14 = 0.14, P < 0.01) and history of drinking ( βˆ16 = 0.24, P < 0.05). Likewise, increases in drinking risk were associated with increased alcohol excite ( βˆ11 = 0.01, P < 0.001), holding constant previous alcohol excite changes ( βˆ9 = 0.78, P < 0.001), birth sex ( βˆ13 = 0.001, ns) and history of drinking ( βˆ15 = 0.02, ns). DISCUSSION The study evaluated questions of change in alcohol associations and change in drinking risk in a sample of US students in their first or second year of college—a time of increased initiation and escalation of drinking and increased drinking risk. The study’s purpose was to test a key—yet rarely tested—tenet of implicit cognition theories; that increases in the strength of alcohol associations would be associated with subsequent increases in drinking and vice versa. This is the second study we know of to test for bi-directional relationships between changes in alcohol associations and changes in drinking and the first study to find evidence of such relationships. Bi-directional relationships were observed for all three associations in this study—drinking identity, alcohol excite and alcohol approach. Notably, study results differ from those of Colder et al. (2014), which found no evidence of reciprocal effects with a sample of US children and early adolescents. A related study that evaluated a slightly older sample of Dutch adolescents and focused on a different type of implicit cognitions (Janssen et al., 2015) also found no evidence of reciprocal effects. Collectively, findings from these studies may indicate the moderating influence of age, developmental stage, the university/college environment, availability of alcohol and/or one of many other factors that vary when examining drinking from a developmental perspective. Research that expands the current model to include such factors and other hypothesized moderators of impulsive processes (e.g. self-regulatory processes, see O’Connor and Colder, 2015) will be crucial for providing a more comprehensive understanding of when and why changes in alcohol associations become predictive of changes in drinking and vice versa. Implications The current results point to the importance of measuring change processes. Earlier analyses from the larger parent study (Lindgren et al., 2016a) did not find that a static measure of alcohol approach associations predicted change in drinking behaviors at the next time point (though drinking identity and alcohol excite associations were predictive), while the present study found that change in each of the alcohol associations was predictive. Although the analyses were not directly comparable in other ways, the distinct results raise the possibility that measuring absolute level of alcohol associations may only tell part of the story of individual risk. A dynamic assessment strategy may, instead, be needed to determine when an individual is becoming more at risk. More generally, although causal claims cannot be made with these data, results support assertions that targeting alcohol associations in prevention and intervention efforts may be helpful (see Wiers et al., 2006). Some progress has been made on this front via computer-based tasks that seek to modify alcohol associations (see Wiers et al., 2011), but these tasks have yielded null results in US college samples (see Lindgren et al., 2015) and a recent meta-analysis (Cristea et al., 2016) raised concerns about the tasks’ efficacy (though there is disagreement about how to interpret the negative findings; Wiers et al., in press). Taken together, this suggests the importance of developing strategies that can target alcohol associations in college student populations. Developing interventions that seek to strengthen the influence of slower, reflective processes and override the alcohol associations may also be helpful. Limitations and future directions Several limitations should be noted. First, the variables representing drinking risk, while based on NIAAA guidelines, were categorical and so are relatively coarse measures. Second, when using categorical variables in cross-lagged models, we needed to treat them as latent variables, meaning we could not make inferences about the probability of being in a given category. Third, the study sample, while relatively large, focuses on a single sample of undergraduates from a single university, and there was attrition over time. We also note that other alcohol associations (e.g. alcohol and valence [good/bad]) have been shown to predict drinking (Wiers et al., 2002) and could be evaluated in future studies. Finally, we note the lower split-half reliabilities of the IATs used in the study. While they are similar to the reliabilities observed when using these IATs in other studies (see Lindgren et al., 2013a) as well as IATs used in other studies (see Greenwald et al., 2003), IATs and implicit measures in general tend to have lower internal consistencies than explicit measures (see Bar-Anan and Nosek, 2014). This is a general challenge for the field. Despite this limitation, there are many examples of the IAT’s predictive validity generally (Bar-Anan and Nosek, 2014) and in alcohol research specifically (Reich et al., 2010; Lindgren et al., 2013b). CONCLUSION Despite these limitations, this study is an important step toward evaluating changes in alcohol associations and drinking. It is the first study to find a bi-directional relationship between increases in the strength of alcohol associations and increases in drinking risk in a sample of students in the early college years, and findings are consistent with theories of implicit cognition. They provide additional evidence that alcohol excite, alcohol approach and drinking identity are important factors to consider in cognitive models of alcohol use and misuse. SUPPLEMENTARY MATERIAL Supplementary data are available at Alcohol And Alcoholism online. ACKNOWLEDGEMENTS The authors affirm there are no conflicts of interest for this work. B.A.T has a significant financial interest in Project Implicit, Inc., which provided services for hosting data collection for this project FUNDING This research was supported by the National Institute on Alcohol Abuse and Alcoholism [Grant nos. R01AA021763 (PI: Lindgren) and R01AA024732 (PI: Lindgren)]. NIAAA had no role in the study design, collection, analysis or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication. CONFLICT OF INTEREST STATEMENT None declared. REFERENCES Atkins DC , Baldwin SA , Zheng C , et al. . ( 2013 ) A tutorial on count regression and zero-altered count models for longitudinal substance use data . Psychol Addict Behav 27 : 166 . https://doi.org/10.1037/a0029508.supp . Google Scholar CrossRef Search ADS PubMed Baker TB , Piper ME , McCarthy DE , et al. . ( 2004 ) Addiction motivation reformulated: an affective processing model of negative reinforcement . Psychol Rev 111 : 33 – 51 . https://doi.org/10.1037/0033-295×.111.1.33 . Google Scholar CrossRef Search ADS PubMed Bar-Anan Y , Nosek BA . ( 2014 ) A comparative investigation of seven indirect measures of social cognition . Behav Res Methods 46 : 668 – 88 . https://doi.org/10.3758/s13428-013-0410-6 . Google Scholar CrossRef Search ADS PubMed Berridge KC , Robinson TE , Aldridge JW . ( 2009 ) Dissecting components of reward: ‘liking’, ‘wanting’, and learning . Curr Opin Pharmacol 9 : 65 – 73 . https://doi.org/10.1016/j.coph.2008.12.014 . Google Scholar CrossRef Search ADS PubMed Colder CR , O’Connor RM , Read JP , et al. . ( 2014 ) Growth trajectories of alcohol information processing and associations with escalation of drinking in early adolescence . Psychol Addict Behav 28 : 659 – 70 . https://doi.org/10.1037/a0035271 . Google Scholar CrossRef Search ADS PubMed Collins RL , Parks GA , Marlatt GA . ( 1985 ) Social determinants of alcohol consumption: the effects of social interaction and model status on the self-administration of alcohol . J Consult Clin Psychol 53 : 189 – 200 . https://doi.org/10.1037/0022-006×.53.2.189 . Google Scholar CrossRef Search ADS PubMed Cooper ML , Frone MR , Russell M , et al. . ( 1995 ) Drinking to regulate positive and negative emotions: a motivational model of alcohol use . J Pers Soc Psychol 69 : 990 – 1005 . https://doi.org/10.1037/0022-3514.69.5.990 . Google Scholar CrossRef Search ADS PubMed Cox WM , Klinger E . ( 1988 ) A motivational model of alcohol use . J Abnorm Psychol 97 : 168 – 80 . https://doi.org/10.1037/0021-843×.97.2.168 . Google Scholar CrossRef Search ADS PubMed Cristea IA , Kok RN , Cuijpers P . ( 2016 ) The effectiveness of cognitive bias modification interventions for substance addictions: a meta-analysis . PLOS ONE 11 : e0162226 . https://doi.org/10.1371/journal.pone.0162226 . Google Scholar CrossRef Search ADS PubMed Dingle GA , Cruwys T , Frings D . ( 2015 ) Social identities as pathways into and out of addiction . Front Psychol 6 : 1795 . https://doi.org/10.3389/fpsyg.2015.01795 . Google Scholar CrossRef Search ADS PubMed Frings D , Albery IP . ( 2015 ) The social identity model of cessation maintenance: formulation and initial evidence . Addict Behav 44 : 35 – 42 . https://doi.org/10.1016/j.addbeh.2014.10.023 . Google Scholar CrossRef Search ADS PubMed Fromme K , Corbin WR , Kruse MI . ( 2008 ) Behavioral risks during the transition from high school to college . Dev Psychol 44 : 1497 – 1504 . Google Scholar CrossRef Search ADS PubMed Gray HM , LaPlante DA , Bannon BL , et al. . ( 2011 ) Development and validation of the Alcohol Identity Implicit Associations Test (AI-IAT) . Addict Behav 36 : 919 – 26 . doi:10.1016/j.addbeh.2011.05.003 . Google Scholar CrossRef Search ADS PubMed Greenwald AG , Banaji MR . ( 1995 ) Implicit social cognition: attitudes, self-esteem, and stereotypes . Psychol Rev 102 : 4 – 27 . http://dx.doi.org/10.1037/0033-295×.102.1.4 . Google Scholar CrossRef Search ADS PubMed Greenwald AG , McGhee DE , Schwartz JK . ( 1998 ) Measuring individual differences in implicit cognition: the implicit association test . J Pers Soc Psychol 74 : 1464 – 80 . http://dx.doi.org/10.1037/0022-3514.74.6.1464 . Google Scholar CrossRef Search ADS PubMed Greenwald AG , Nosek BA , Banaji MR . ( 2003 ) Understanding and using the Implicit Association Test: I. An improved scoring algorithm . J Pers Soc Psychol 85 : 197 – 216 . http://dx.doi.org/10.1037/0022-3514.85.2.197 . Google Scholar CrossRef Search ADS PubMed Janssen T , Larsen H , Vollebergh WAM , et al. . ( 2015 ) Longitudinal relations between cognitive bias and adolescent alcohol use . Addict Behav 44 : 51 – 7 . https://doi.org/10.1016/j.addbeh.2014.11.018 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Foster DW , Westgate EC , et al. . ( 2013 b) Implicit drinking identity: Drinker+me associations predict college student drinking consistently . Addict Behav 38 : 2163 – 6 . https://doi.org/10.1016/j.addbeh.2013.01.026 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Neighbors C , Gasser ML , et al. . ( 2016 b) A review of implicit and explicit substance self-concept as a predictor of alcohol and tobacco use and misuse . Am J Drug Alcohol Abuse 43 : 237 – 46 . https://doi.org/10.1080/00952990.2016.1229324 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Neighbors C , Teachman BA , et al. . ( 2013 a) I drink therefore I am: validating alcohol-related implicit association tests . Psychol Addict Behav 27 : 1 – 13 . https://doi.org/10.1037/a0027640 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Neighbors C , Teachman BA , et al. . ( 2016 a) Implicit alcohol associations, especially drinking identity, predict drinking over time . Health Psychol 35 : 908 – 18 . https://doi.org/10.1037/hea0000396 . Google Scholar CrossRef Search ADS PubMed Lindgren KP , Wiers RW , Teachman BA , et al. . ( 2015 ) Attempted training of alcohol approach and drinking identity associations in US undergraduate drinkers: null results from two studies . PLOS ONE 10 : e0134642 . https://doi.org/10.1371/journal.pone.0134642 . Google Scholar CrossRef Search ADS PubMed Merrill JE , Carey KB . ( 2016 ) Drinking over the lifespan . Alcohol Res 38 : 103 – 14 . Retrieved from http://www.arcr.niaaa.nih.gov/arcr/arcr381/article12.htm. Google Scholar PubMed Muthén B , Asparouhov T . ( 2002 ) Latent variable analysis with categorical outcomes: Multiple-group and growth modeling in Mplus. Mplus Web Notes: No. 4. www.statmodel.com Muthén LK , Muthén BO ( 1998 –2017) Mplus User’s Guide , 8th edn. Los Angeles, CA : Muthén &Muthén . Naimi TS , Brewer RD , Mokdad A , et al. . ( 2003 ) Binge drinking among US adults . J Am Med Assoc 289 : 70 – 5 . https//doi.org/10.1016/j.amepre.2009.09.039 . Google Scholar CrossRef Search ADS National Institute on Alcohol Abuse and Alcoholism . ( 2017 ) Drinking Levels Defined. Retrieved from NIAAA website: https://www.niaaa.nih.gov/alcohol-health/overview-alcohol-consumption/moderate-binge-drinking Newsom JT . ( 2015 ) Longitudinal Structural Equation Modeling: A Comprehensive Introduction . New York : Routledge . Nosek BA , Greenwald AG , Banaji MR . ( 2007 ) The Implicit Association Test at age 7: a methodological and conceptual review. In Bargh JA (ed) . Automatic Processes in Social Thinking and Behavior . New York : Psychology Press , 265 – 92 . Nosek BA , Hawkins CB , Frazier RS . ( 2011 ) Implicit social cognition: from measures to mechanisms . Trends Cogn Sci 15 : 152 – 9 . https://doi.org/10.1016/j.tics.2011.01.005 . Google Scholar CrossRef Search ADS PubMed O’Connor RM , Colder CR . ( 2015 ) The prospective joint effects of self-regulation and impulsive processes on early adolescence alcohol use . J Stud Alcohol Drugs 76 : 884 – 94 . https://doi.org/10.15288/jsad.2015.76.884 . Google Scholar CrossRef Search ADS PubMed Reich RR , Below MC , Goldman MS . ( 2010 ) Explicit and implicit measures of expectancy and related alcohol cognitions: a meta-analytic comparison . Psychol Addict Behav 24 : 13 – 25 . https://doi.org/10.1037/a0016556 . Google Scholar CrossRef Search ADS PubMed Schulenberg JE , Johnston LD , O’Malley PM , et al. . ( 2017 ). Monitoring the Future national survey results on drug use, 1975–2016: Volume II, College students and adults ages 19–55 . Ann Arbor: Institute for Social Research, The University of Michigan. http://monitoringthefuture.org/pubs.html#monographs . Stacy AW , Wiers RW . ( 2010 ) Implicit cognition and addiction: a tool for explaining paradoxical behavior . Annu Rev Clin Psychol 6 : 551 – 75 . https://doi.org/10.1146/annurev.clinpsy.121208.131444 . Google Scholar CrossRef Search ADS PubMed Tajfel H , Turner JC . ( 1979 ) An integrative theory of intergroup conflict. In Austin WG , Worchel S (eds) . The Social Psychology of Intergroup Relations . Monterey, CA : Brooks/Cole Publishing Company , 33 – 47 . Thush C , Wiers RW . ( 2007 ) Explicit and implicit alcohol-related cognitions and the prediction of future drinking in adolescents . Addict Behav 32 : 1367 – 83 . https://doi.org/10.1016/j.addbeh.2006.09.011 . Google Scholar CrossRef Search ADS PubMed Tombor I , Shahab L , Brown J , et al. . ( 2015 ) Does non-smoker identity following quitting predict long-term abstinence? Evidence from a population survey in England . Addict Behav 49 : 99 – 103 . http://dx.doi.org/10.1016/j.addbeh.2015.01.026 . Google Scholar CrossRef Search ADS Wiers RW , Bartholow BD , van den Wildenberg E , et al. . ( 2007 ) Automatic and controlled processes and the development of addictive behaviors in adolescents: a review and a model . Pharmacol, Biochem Behav 86 : 263 – 83 . https://doi.org/10.1016/j.pbb.2006.09.021 . Google Scholar CrossRef Search ADS Wiers RW , Boffo M , Field M . (in press) What’s in a trial? On the importance of distinguishing between experimental lab-studies and randomized controlled trials; the case of cognitive bias modification and alcohol use disorders . J Stud Alcohol Drugs . Wiers RW , Cox WM , Field M , et al. . ( 2006 ) The search for new ways to change implicit alcohol-related cognitions in heavy drinkers . Alcohol Clin Exp Res 30 : 320 – 31 . https://doi.org/10.1111/j.1530-0277.2006.00037.x . Google Scholar CrossRef Search ADS PubMed Wiers RW , Eberl C , Rinck M , et al. . ( 2011 ) Retraining automatic action tendencies changes alcoholic patients’ approach bias for alcohol and improves treatment outcome . Psychol Sci 22 : 490 – 7 . https://doi.org/10.1177/0956797611400615 . Google Scholar CrossRef Search ADS PubMed Wiers RW , van Woerden N , Smulders FTY , et al. . ( 2002 ) Implicit and explicit alcohol-related cognitions in heavy and light drinkers . J Abnorm Psychol 111 : 648 – 58 . https://doi.org/10.1037/0021-843×.111.4.648 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Medical Council on Alcohol and Oxford University Press. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Alcohol and AlcoholismOxford University Press

Published: Mar 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off