RISK TAKING AS AN OUTCOME OF AN ALCOHOL HARM REDUCTION TRIAL

Alcohol related risk taking behaviors are often assessed within measures of alcohol related problems and consequences. Although some research has found evidence that brief intervention on certain risk taking behaviors is effective, the larger scope of risk taking behaviors is often ignored. The present study aims to fill the gap of risk taking measures by (a) identifying a subscale from a scale assessing alcohol-related risk and consequences that captures risk taking behavior outside of alcohol consumption, (b) confirming that this risk taking scale holds across samples and (c) that the risk taking measure will show change over time, and (d) evaluating whether the risk taking scale is moderated by a brief intervention treatment. Results indicated that there a risk taking scale could be identified over a larger scale of risks and consequences. Confirmatory factor analysis confirmed the factor structure of the scale. And finally, growth curve modeling provided evidence for measuring risk taking over time. In the current study, no effect of treatment was found on risk taking, though due to study design, this was not unexpected.

iii Acknowledgement I would like to thank my major professor Dr. Robert G. Laforge for all of his help in the development of this Master's thesis. I would also like to thank my committee members, Dr. Nicole H. Weiss and Dr. Natallia Katenka, for their continued support through this process.
iv    Brief intervention techniques for alcohol use have been studied extensively for nearly four decades and are considered to be effective in identifying and helping to reduce problematic drinking behaviors in nondependent adults (O'Donnell et al., 2013).
A number of studies have investigated the use of brief intervention in emergency departments in an effort to zero in on risk taking behaviors (Blow et al., 2006;Blow et al., 2011;Cunningham et al, 2015;Houry, Hankin, Daugherty, Smith, & Kaslow, 2011;Longabaugh et al., 2001;Mello, Longabagh, Baird, Nirenberg, & Woolard, 2008;Sommers et al, 2013). Sommers and colleagues (2013) found that delivering brief interventions for more than one risky behavior in an emergency department significantly reduced risky driving for 9 months and hazardous drinking for 6 months (Sommers et al., 2013). Although these studies often have significant findings for the efficacy of brief interventions, these results are moderated by the attribution of the injury to alcohol by the patient (Walton et al., 2008).
Risk taking behaviors are goal-directed and may result in more than one outcome, one of which is often undesirable and/or dangerous (Furby & Beyth-Maron, 1992).
Understanding why and how people engage in risky behaviors surrounding alcohol use and misuse has been a focus of much research over the years. Risk taking behaviors are often attributed to impulsivity, which is a potential underlying mechanism influencing altered decision making (Krause et al., 2017) and the development of substance use disorders (Jupp & Dalley, 2014;Littlefield & Sher, 2010). Alcohol myopia theory (Steele & Josephs, 1990), which posits that short-sighted information processing that is part of alcohol intoxication, has been widely attributed to risk taking behaviors (Franzen, et al., 2018;Massa et al., 2018;Norris, Davis, George, Martell, & Heiman, 2002). According to alcohol myopia theory, the pharmacological effects of alcohol taxes a person's cognitive resources, narrowing one's limited attention onto the most salient cues in the environment, with more peripheral information being largely ignored (Steele & Josephs, 1990). In this model, risk taking behavior is more likely to occur when risk taking cues are present (e.g., aggression in the context of domestic conflict).
Another possibility as to why people engage in risky behaviors when consuming alcohol is more motivational than cognitive. Tyszka, Macko, and Stańczak (2015) examined ambiguity aversion in an effort to understand why people become more risk prone when they consume alcohol. Ambiguity aversion refers to the preference for situations with known risks over situations with unknown risks. The researchers found that along with becoming less risk averse, people under the influence of alcohol would become less ambiguity averse. Tyszka and colleagues also attribute these results to socially and culturally valued patterns of conduct. Moderate risk is often valued in Western culture and people will shift toward risky decisions to gain approval from peers (Tyszka et al., 2015). Despite the fact that alcohol-related risk taking behaviors are of great interest to researchers, these behaviors are often assessed with measures including alcohol-related consequences, which confounds a cause (i.e., alcohol-related risk taking behaviors) with an effect (i.e., alcohol-related consequences; Miller, Tonigan, & Longabaugh, 1995;Saunders, Aasland, Babor, de la Fuente, & Grant, 1993;. Differentiating risk taking behaviors from consequences and alcohol consumption may allow for more nuanced analyses of alcohol harm interventions. Risk taking behaviors are often targeted because of the public health problems that they pose (e.g., drinking and driving; sexual risk taking; aggression). When brief intervention studies target risk taking behaviors, they are often recruited from emergency departments where potential participants are experiencing problems and consequences related to risky drinking (Blow et al., 2006;Blow et al., 2011;Cunningham et al., 2010;Mello et al., 2008;Nilsen et al., 2008;Sommers et al., 2013;Suffoletto et al., 2012).
Since the goal of brief interventions is often to target risky drinking behaviors in an effort to prevent alcohol use disorders (Zoorob, Snell, Kihlberg, & Senturias, 2014), it stands to reason that identifying and intervening on general risk taking behaviors prior to consequences could benefit public health. Organization. They were randomly assigned to an experimental treatment (n = 430), an assessment matched control condition (n = 438), or a minimally assessed condition (n = 461) using urn randomization to ensure baseline group equivalence on prognostic indicators, including gender and high-risk drinking behaviors and alcohol related problems (Laforge et al., 2003;Stout et al., 1994).
Participants were proactively recruited from the membership of a Managed Care Organization and data were collected by telephone survey on up to six occasions over a two-year period: baseline, 3-, 6-, 12-, 18-and 24-months post-baseline. The treatment group received brief tailored multidimensional motivational feedback reports following the baseline, 3-and 6-month assessments. As part of a larger battery, information was collected on demographics (baseline only), and repeated measures of drinking behaviors and related cognitive measures, such as situational temptations to drink, reactance to alcohol harm reduction messaging, stage of readiness to change high risk drinking, decisional balance, several measures of processes of change (e.g., consciousness raising, dramatic relief, counter-conditioning, stimulus control), measures of health care utilization (e.g., "During the last 12 months, how many times have you been to a hospital emergency room about your own health""), and 25 items that assess alcohol related risk taking and negative or harmful consequences, including the 15-item short inventory of problems (SIP-2R; Miller, Tonigan, & Longabaugh, 1995).

Present Study Design
The proposed study was conducted in three phases. Phase I involved the development of a psychometrically sound measure of alcohol-related risk taking behavior, other than alcohol consumption, from available measures in the PARR data set.
Baseline data from participants randomized into the minimally assessed group (n = 461) was used for this analysis. Phase II involved the confirmation of an alcohol-related risk taking measure using confirmatory factor analysis (CFA) to develop a risk taking variable to be used in Phase III. A random split half sample (n = 430) of the combined treatment and control group was used for CFA analysis. Phase III evaluated the efficacy of the brief individualized feedback intervention on change in alcohol-related risk taking behavior.
Phase III analyses uses mixed model regression to examine whether longitudinal change in risk taking is moderated by exposure to the intervention. Then, each time point for the remaining random split half sample (n = 438) was used for Phase III analyses. See Figure   1 for flow chart of sample.

Participants
The Note. PARR min = minimally assessed group used in Phase I only.

Measures used in all phases
Construct Validity Measures Additional measures included in the larger study were evaluated as part of validity efforts. These included smoking, the four dropped items from the original 25-item alcohol-related risk and consequences scale (e.g., "I have had an accident while drinking or intoxicated;" "While drinking I have gotten into sexual situations that I later regretted") as well as questions about drinking behaviors (i.e., "Over the past three months, how many drinks containing alcohol did you have on a typical day when you are drinking?;" "During the past 30 days, what is the highest number of drinks that you had on any one occasion?" "How often during the last year have you found that you were not able to stop drinking once you had started?"), frequency of binge drinking in the past month (five or more drinks in a row for men, four or more drinks in a row for women) and history of drinking (i.e., age of first drink, age first got drunk).

Analyses used in all phases
Construct validity Construct validity of the risk taking measure was evaluated in each of the participant samples studied in the three phases of this thesis by examining the statistical association of the risk taking summary score in each sample with other measures of risk taking behavior. Convergent validity was determined by comparing the risk-taking measure to risky behaviors such as smoking, binge drinking frequency, and exceeding NIAAA recommended drinks per week. Discriminant validity was completed by comparing the risk-taking measure with low risk behaviors such as low risk drinking and low volume drinking. Convergent validity was evaluated using Pearson's correlation for continuous items. Point biserial, a special case of Pearson's correlation, was used for dichotomous variables (i.e., "I have had an accident while drinking or intoxicated;" "While drinking, I have gotten into sexual situations I later regretted").
Chapter 3 Phase I: Development of a Risk taking Measure

Description
Phase I of the study sought to develop a psychometrically sound measure of alcohol-related risk taking behavior from a measure of alcohol-related problems. Many scales evaluating consequences of alcohol use include measures of alcohol-related risk taking behaviors as indicators of alcohol related adverse consequences. The goal of Phase I was to explore the factor structure of the measure of alcohol-related problems used in the PARR study to identify whether alcohol related risk taking variables load on a single factor, distinct from other the other variables that assess other dimensions of alcohol related adverse consequences.

Phase I Participants
Phase I of the study consisted of a subset of the larger PARR study who were randomized into a minimally assessed group and completed baseline only. The minimally assessed group comprised of 461 participants with an average age of 42.91 (SD = 12.71).
The sample was 52.1% female, 89.2% White, and 73.5% employed at least part time.

Alcohol-Related Risk and Consequence Measure
The PARR study repeatedly assessed 25 items adapted from several alcohol consequences scales that assess different dimensions of alcohol related risk taking and adverse consequences and are commonly mentioned in alcohol literature on adult drinking (Allen & Columbus, 1997;Miller, 1996). The first 15 items of the PARR alcohol risks and consequences measure comes from a revised version of the short inventory of problems (SIP-R2). A subset of the longer 50-item Drinking Inventory of Consequences (DrInC ), the SIP-R2 is a selfreport measure of recent alcohol-related consequences that measures five domains of problems non-dependent alcohol drinkers might have experienced in the recent past: physical, inter-personal, intra-personal, impulse control, and social responsibility (Miller et al., 1995). The SIP-2R takes three items from each of the established five-factors of the DrInC, forming a shorter inventory of alcohol related problems while, theoretically, maintaining the relationship between the observed and unobserved variables. Participants are asked to indicate how often each of the listed items has occurred in the past three months ("never," "once or a few times," "once or twice a week," "daily or almost daily"; scored 0-3). Ten additional items of risk taking and consequences using the same response format was used to complete the 25-item measure.
The following four items from the consequences scale were found to have nonnormally distributed (dichotomous) responses and were dropped from the model: I have had money problems because of my drinking; I have had an accident while drinking or intoxicated; I have been stopped or arrested for driving under the influence; While drinking, I have gotten into sexual situations that I later regretted.

Phase I Data Analytic Strategy
Phase I of the study used exploratory factor analysis (EFA) to define the factor structure of the 25 alcohol-related risk and consequences measure. Four of the items were dropped from the analyses due to dichotomous responses resulting in the analysis of 21 alcohol-related risk and consequences measure. Factor analysis was justified using Bartlett's test of sphericity and the Kaiser-Meyer-Oklin (KMO) measure of sampling adequacy (Bartlett, 1950;Kaiser, 1970). The Bartlett's Test of Sphericity should be significant (p < .05) and the KMO index, which ranges from 0 to 1, should be at least 0.50 considered suitable for factor analysis (Field, 2009 , Steger, 2006). Parallel analysis has been indicated by many studies as a consistently accurate model of factor extraction (Henson & Roberts, 2006;Thompson & Daniel, 1996;Thompson, 2004). Parallel analysis is a Monte Carlo method comparing observed eigenvalues extracted from the correlation matrix to be analyzed with those obtained from uncorrelated normal variables (Horn, 1965). For the present study, the number of factors to retain was based on parallel analysis, a visual examination of the scree plot, and considerations regarding the meaning and interpretability of the factor model.
Principal axis factoring (PAF), an extraction method that has no distributional assumptions (Fabrigar, et al., 1999), was conducted to define underlying latent factors for the alcohol-related risk and consequences measure. PAF is less likely to inflate factor loadings or underestimate factor correlations than other methods (Fabrigar et al., 1999) as it recognizes measurement error (Baglin, 2014). Although an orthogonal risk taking factor was anticipated, oblimin, an oblique rotation factor pattern solution was chosen for PAF due to risk taking behaviors being correlated with alcohol-related problems (Cyders, Flory, Rainer, & Smith, 2009;Magid, MacLean, & Colder, 2007). If the items that load on the risk factor are actually orthogonal to the other scale items, they will all be found to load on the same factor, regardless of whether oblique or some other type of factor rotation is used. The goal of the PAF was to identify a risk taking subscale distinct from the consequences items. Items were considered part of the risk taking scale if they loaded onto the same latent factor at |.30| or higher. Items that loaded |.30| or higher on two or more factors were defined as complex.
The underlying structure of the hypothesized model was tested to evaluate the degree of model fit and assess whether the fit could be improved as a function of testing alternative models. Four models were tested: a five-factor model based on the existing structure of the SIP-2R as well as three-, four-, and six-factor models to determine if the established five-factor model held with the present data and additional items.
Once the final risk taking measure was identified, tests for preliminary construct validity were completed. Convergent validity was determined using Pearson's correlation for continuous items. Point biserial, a special case of Pearson's correlation, was used for dichotomous variables (i.e., "I have had an accident while drinking or intoxicated;" "While drinking, I have gotten into sexual situations I later regretted"). For items with ordinal responses (i.e., "Have you smoked cigarettes in the past year?") Spearman's rho was calculated to determine preliminary validity.

Phase I Results
A positive correlation was observed among 25 items of alcohol consequences in a correlation matrix. Bartlett's test of sphericity, which tests that the correlation matrix has an identity matrix, was found statistically significant (χ 2 (24) = 5318.8, p < 0.001). A statistically significant Bartlett's test provides a minimum standard to proceed for factor analysis. The KMO measure of sampling adequacy was found to be 0.84, which is acceptable to justify factor analysis (Kaiser, 1974).
Four models were run with the 21 items from the PARR alcohol-related risk and consequences scale distributed across three-, four-, five-, and six-factors. Of the tested models, the four-factor had the best fit, explaining 41% of the variance. All items in the final model loaded at or above |.34| (see Table 2). The final four factor structure included interpersonal conflicts, intrapersonal conflicts, life consequences, and risk taking subscales.
As shown in Table 2

0.68
My family has been hurt by my drinking.

0.87
A friendship or close relationship has been damaged by my drinking.

0.71
My drinking has damaged my social life, popularity, or reputation.

0.53
I have experienced 'conflicts' at home due to my drinking 0.42 My physical health has been harmed by my drinking

0.47
My physical appearance has been harmed by my drinking.

0.52
I have spent too much or lost a lot of money because of my drinking

0.49
Construct validity of the Phase 1 risk taking scale All correlation coefficients were in the expected direction and showed shared variance (see Table 3 for means and standard deviations of validity items). As expected, the risk taking scale was positively correlated with risky behaviors such as smoking (rs = .231, p < .001), typical numbers of alcohol drinks per day (r = .342, p < .001), frequency of binge drinking in past month (r = .455, p <.001), highest number of drinks per drinking occasion (r = .482, p < .001), and inability to stop drinking once started (r = .343, p <.001). The risk taking scale was also positively correlated with dropped items from the alcohol-related risks and consequences: I have had an accident while drinking or intoxicated (rpb = .238, p <.001) and While drinking I have gotten into sexual situations I regretted later (rpb = .401, p <.001).
Additionally, the risk taking scale was negatively correlated with age of first drink (r = -.102, p = .042) and age first got drunk (r = -.145, p = .004), indicating that alcohol-related risk taking is correlated with drinking and getting drunk at an earlier age.

Phase I Discussion
Factor extraction using PAF and a priori evaluation of the factor structure identified a four-factor model. The proposed factor structure assumed by the SIP-2R (Miller et al., 1995) did not hold in this sample. However, analyses did identify a risk taking factor among the 25-item risk and consequences measure in the PARR study.
These results replicated in an orthogonal model. For Phases II and III, the combined treatment and assessment matched control groups who had been randomly assigned to group at baseline (n = 868), were now randomly split into two half-samples; with one-half ample used in each phase. The split half sample used for the Phase II CFA consisted of 430 participants, with an average age of 43.33 (SD = 13.14). The sample was 51.6% female, primarily white (89.9%), and employed at least part time (85.5%).

Phase II Measures
Risk Taking Measure Phase II used the seven-item risk taking scale identified by the EFA performed in Phase I of the study. These items asked "During the last 3 months, about how often has this happened to you?" ("never," "once or a few times," "once or twice a week," "daily or almost daily").

Phase II Results
The results of the final CFA model are shown in Figure 2 and boat or other motor vehicle"). These two items assess risk taking related to alcohol use and motor vehicles, and it appears that the correlated errors are due to method error related to similar item content. Hence, the error terms for items 16 and 18 were allowed to covary in the final CFA model. See Table 4 for item loadings. See Figure 2 for the final CFA model with standardized results.

Construct Validity of the Risk taking scale in the Phase II sample population
All correlation coefficients were in the expected direction and showed shared variance (see Table 5 for means and standard deviations of validity items). As expected, the risk taking scale was positively correlated with risky behaviors such as typical numbers of alcohol drinks per day (r = .356, p < .001), binge drinking in past month (r = .384, p <.001), highest number of drinks per occasion (r = .452, p < .001), and inability to stop drinking once started (r = .239, p <.001). The risk taking scale was also positively correlated with dropped items from the alcohol-related risks and consequences: I have had an accident while drinking or intoxicated (rpb = .304, p <.001) and While drinking I have gotten into sexual situations I regretted later (rpb = .398, p <.001). Additionally, the risk taking scale was negatively correlated with age of first drink (r = -.102, p = .042), indicating that alcohol-related risk taking is correlated with drinking and getting drunk at an earlier age. The risk taking scale was not correlated with smoking (rs = .094, p = .052) and age first got drunk (r = -.066, p = .214) in this sample. non-smoking, and 86.0% employed at least part time.

Risk Taking A risk taking measure was developed in Phases I and confirmed in
Phase II of the present study. The risk taking measure consisted of seven items of alcohol-related risk taking behaviors. Participants were primed with "During the last 3 months, about how often has this happened to you?" ("never," "once or a few times," "once or twice a week," "daily or almost daily"). Treatment Treatment refers to the experimental group into which the participants were randomly assigned. Treatment is an indicator variable representing experimental condition coded to indicate the brief feedback intervention condition (Treatment = 1) and the assessment matched "Control" condition (Treatment = 0). The assessment matched control group received the same survey assessments as the Treatment group at each assessment wave, but did not receive the brief tailored motivational feedback reports following the baseline, 3-and 6-month assessments. The random intercept estimates individual variation from the group level fixed effect estimate of the intercept. The random slope(s) estimate(s) represent the amount of individual variation from the group level estimate(s) of the fixed slope(s).

The Unconditional Model Analysis
The purpose of the unconditional model analyses was to find the best functional form of time for the growth model for risk taking. The unconditional growth model analysis assesses whether there is systematic variation in the change in risk taking, and is used to identify which of the different possible parameterizations of the functional form of time best fits the underlying trend in the data. To identify the optimal functional form of time for risk taking, a series of models with successively complex parameters for time was compared and the most appropriate candidate model (or models) is determined based on a comparison of fit statistics analogous to the approach described in Phase II. For nested growth curve models, goodness of fit was compared using Liklihood ratio test. For non-nested growth curve models, the Akaike Information Criterion (AIC; Akaike, 1973) and Bayesian Information Criterion (BIC; Schwarz, 1978) were used, with smaller values indicating a better fitting model. The initial shape of the underlying pattern of growth was explored with the nominal time model, and then by comparing a series of successively higher order linear polynomial growth models. The end result of the unconditional model analyses is to identify an optimal expression of time for use in the conditional model analyses.

Conditional Model Analyses
The results of the unconditional analyses are then used as the initial parameterization of time for the growth model. The conditional growth curve model considers repeated measures of an outcome behavior (i.e., risk taking) as a function of time and other individual and group level measures (Duncan & Duncan, 1995;Singer and Willet, 2003). In this study, the next step was to determine if the change in risk taking behaviors is moderated by treatment group.
The hypothesis being tested in the conditional analyses proposes that the risk taking over time was differentially reduced more in the treatment group compared to the control group. This model tests whether there are significantly different patterns of growth between the two experimental conditions since all participants were randomly assigned to experimental condition, the conditional model included terms for the predictors; Treatment group, Time and Treatment group by Time. The moderation hypothesis is tested by evaluating the statistical significance of the Treatment by Time interaction term in the GLMM model.

Cross Sample Validation
See Table 6     The results of the best fitting nominal time model are shown in Table 8 for illustrative purposes. These results reveal a significant decreasing linear trend in risk taking over the two-year follow-up period.

Conditional Model
The conditional results of the best fitting negative binomial GLMM nominal time model, are shown in Table 9. The Type III test of the Time by Group interaction assessing whether change in risk taking was moderated by the Treatment shows was not significant as indicated by the LL ratio test  2 (6) = -2.42, p = 0.877, for the comparison of best fitting unconditional nominal Time model Group and Group by Time interaction terms added. As was evident in the unconditional models, these results also indicated a decreasing secular trend in risk taking over most of the study period. All correlation coefficients were in the expected direction and showed shared variance (see Table 10 for means and standard deviations of validity items). As expected, the risk taking scale was positively correlated with risky behaviors such as smoking (r = .141, p = .003), typical numbers of alcohol drinks per day (r = .294, p < .001), binge drinking in past month (r = .252, p <.001), highest number of drinks per occasion (r = .394, p < .001), and inability to stop drinking once started (r = .225, p <.001). The risk taking scale was also positively correlated with dropped items from the alcohol-related risks and consequences: I have had an accident while drinking or intoxicated (rpb = .328, p <.001) and While drinking I have gotten into sexual situations I regretted later (rpb = .429, p <.001). The risk taking scale was not correlated with age of first drink (r = -.087, p < 1.000), age first drunk (r = -.092, p = .081), or typical numbers of alcohol drinks per day (r = .028, p = .556) in this sample. The goal of the present study was to identify a risk taking measure from a 25-item scale of alcohol-related risk and consequences used in the PARR study. In Phase I, exploratory analysis indicated the existence of a risk taking factor within a measure of alcohol-related risk and consequences scale. In Phase II, the factor structure of the risk taking measure was confirmed using CFA. In Phase III, analysis of the risk taking variable indicated change over time in risk taking, although this change was not moderated by treatment in this study.
Identifying a risk taking scale is an important preliminary step in targeting alcohol-related risk taking behaviors. Traditionally, measures that assess alcohol-related risk taking behaviors are assessed with alcohol-related consequences as a single alcoholrelated problems scale (Miller et al., 1995;Saunders et al, 1993;White & Labouvie, 1998). Previous studies have identified the need for targeting risk taking behaviors within alcohol brief intervention studies (Blow et al., 2006;Blow et al., 2011;Cunningham et al., 2010;Longabaugh et al., 2011;Mello et al., 2008;Sommers et al., 2013).
Although researchers have targeted specific risk taking behaviors in alcohol brief interventions, consequences such as motor vehicle collisions (Schermer et al., 2006;Sommers et al., 2013) and other emergency department visits (Suffoletto et al., 2003) often serve as the recruitment for such intervention studies. Sommers and colleagues (2013) found that targeting two risky behaviors (i.e., risky driving and hazardous drinking) in an emergency department significantly reduced both risky behaviors for an extended period of time (9-and 6-months respectively; Sommers et al., 2013). By using a general risk taking measure, researchers might be able to identify non-dependent adult drinkers who may benefit from interventions aimed at the reduction of risk taking behaviors.
The present study used PAF to identify factors within a 25-item risk and consequences scale. This exploratory factor analysis identified four factors. A single factor made up of risk taking variables was identified and the remaining factors consisted of various consequences. With the identification of a risk taking subscale, analyses proceeded to Phase II where we confirmed the factor structure of the subscale. CFA confirmed the existence of a risk taking scale. From there, we moved to demonstrating that risk taking through this scale could be modeled over time. The final analysis sought to determine if the change in risk taking over time was moderated by treatment condition (i.e., intervention and assessment matched control).
Unconditional negative binomial GLMM model analysis with random intercept and random slope using linear time provided evidence that the risk taking measure can be modeled over time. Significant temporal decreases in risk taking over time from the baseline assessment was seen at all timepoints in the unconditional model. A conditional model added the effect of study intervention to the model to determine if treatment moderated risk taking over time. In this sample, no effect of treatment was found. These results are not unexpected as the multidimensional feedback treatment for the study was designed to intervene on alcohol related problems. The failure to find treatment differences may be due to other factors that affect both treatment conditions, including repeated assessment of variables of interest. Although no effect of treatment was found in this study, evidence of change in risk taking over time provides clinical implications for treatment. The present study demonstrated that scores on the risk taking measure decreased over time, which implies that intervention could target risk taking behaviors.

Limitations
The present study used a measure specific to the PARR study and cannot be generalized to all alcohol-related risk and consequences scales. The sample of nondependent adult alcohol drinkers was predominantly White and cannot be generalized across racial and ethnic groups. Gender was measured as a binary option of "Male" and "Female" thus cannot be generalized to non-binary genders. The treatment was aimed at problems and not risk taking, so it is unknown if a randomized brief intervention study would show change in broad alcohol-related risk taking behaviors.

Future Directions
Future studies should identify a risk taking subscale in established alcohol-related risk and consequences scales such as the DrInC. By identifying a scale directly measuring alcohol-related risk taking behaviors, studies can evaluate general risk taking behavior as a risk factor for alcohol use. Since brief interventions have been effective in targeting certain risky behaviors, a brief intervention study targeting overall risk taking behaviors could add to the literature. Future studies should also assess factors related to risk taking behaviors that can be targeted clinically such.