Structural, concurrent, and predictive validity of the Substance Use Risk Profile Scale in early adolescence

Structural, concurrent, and predictive validity of the Substance Use Risk Profile Scale in early adolescence

Addictive Behaviors 36 (2011) 37–46 Contents lists available at ScienceDirect Addictive Behaviors Structural, concurrent, and predictive validity o...

203KB Sizes 0 Downloads 44 Views

Addictive Behaviors 36 (2011) 37–46

Contents lists available at ScienceDirect

Addictive Behaviors

Structural, concurrent, and predictive validity of the Substance Use Risk Profile Scale in early adolescence Marvin Krank a,⁎, Sherry H. Stewart b,c, Roisin O'Connor d, Patricia B. Woicik e, Anne-Marie Wall f, Patricia J. Conrod g a

Department of Psychology, University of British Columbia Okanagan, Kelowna, British Columbia, Canada Department of Psychiatry, Dalhousie University, Halifax, Nova Scotia, Canada Department of Psychology, Dalhousie University, Halifax, Nova Scotia, Canada d Department of Psychology, Concordia University, Montreal, Quebec, Canada e Brookhaven National Laboratory, Upton, New York, United States f Department of Psychology, York University, Toronto, Ontario, Canada g Department of Psychological Medicine, Institute of Psychiatry, King's College London, London, United Kingdom b c

a r t i c l e

i n f o

Keywords: Personality Adolescence Substance use Substance problems Development Longitudinal Prospective prediction

a b s t r a c t A brief personality risk profile (23 items), the Substance Use Risk Profile Scale was tested for concurrent and predictive validity for substance use in 1139 adolescents (grades 8–10) from a mid-sized city in western Canada. The SURPS was administered in two waves of a longitudinal study separated by 12 months (2003–04). As expected, four subscales were supported by confirmatory factor and metric invariance analysis. In regression analysis, three subscales, hopelessness, impulsivity, and sensation seeking, were positively related to current and future use; while one, anxiety sensitivity, was negatively related. Findings suggest clinical utility for screening adolescents at risk for substance use. © 2010 Elsevier Ltd. All rights reserved.

1. Introduction Early adolescence (ages 12–16 years) is a significant developmental period of risk for transitions to substance use and abuse (Flight, 2007; Schulenberg, Bryant, & O'Malley, 2004; Schulenberg, O'Malley, Bachman, Wadsworth, & Johnston, 1996; Schulenberg et al., 2005). Many young people first try not only drugs, especially alcohol, but also tobacco, marijuana, and other illicit drugs, at this age. Although at the beginning of middle school few young people are using drugs, the next few years see a rapid escalation in use of alcohol, tobacco, marijuana, and hallucinogens. Moreover, patterns of problem use may begin at these ages and earlier initiation of substance use is a risk factor for future problem use. For example, with alcohol, binge drinking is common in secondary school (D'Amico et al., 2001) and earlier age of first drink predicts later problem of alcohol use (Grant et al., 2006; McGue, lacono, Legrand, Malone, & Elkins, 2001). The rapid growth of experimentation, regular use, and problem use in adolescence demonstrates the importance of understanding the risk and protective factors leading to escalation of use and abuse among youth. The present study examines the role of individual

⁎ Corresponding author. Department of Psychology, The University of British Columbia Okanagan, 3333 University Way, Kelowna, BC, Canada V1V 1V7. Tel.: +1 250 470 9352. E-mail address: [email protected] (M. Krank). 0306-4603/$ – see front matter © 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.addbeh.2010.08.010

differences in personality-based reinforcement susceptibility. The scales used in this study have already been shown to predict susceptibility to substance abuse in older adolescents and in at risk populations (Woicik, Conrod, Stewart, & Pihl, 2009). In the present study, these effects are tested using a cohort sequential longitudinal design with a sample of early adolescents (ages 13–15 years at the first wave). This longitudinal design allows for examination of not only the concurrent validity of the personality factors in predicting substance use and misuse, but also the prospective predictive validity of the personality measures for development of use and misuse over time (Bates & Labouvie, 1995; Curran, White, & Hansell, 1997; Elkins, King, McGue, & Iacono, 2006; Scheier, Botvin, & Baker, 1997). 2. Individual differences in personality and susceptibility to substance abuse Theories of drug abuse vulnerability propose that certain personality traits reflect individual differences in susceptibility to drugreinforcement (Conrod, Pihl, Stewart, & Dongier, 2000; Cooper, Frone, Russell, & Mudar, 1995; Pihl & Peterson, 1995; Verdejo-García, PérezGarcía, & Bechara, 2006). This theoretical perspective is supported by research demonstrating that personality factors differentiate substance abusers based on clinical profile (Cloninger, 1987a,b), treatment response (Morgenstern, Kahler, & Epstein, 1998), different motivations for substance use (Comeau, Stewart, & Loba, 2001; Cooper et al., 1995), and patterns of subjective, behavioral, and

38

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46

neurophysiological responses to the acute effects of substance abuse (Brunelle et al., 2004; Conrod, Pihl, & Vassileva, 1998; Leyton et al., 2002; MacDonald, Baker, Stewart, & Skinner, 2000; Perkins, Gerlach, Broge, Grobe, & Wilson, 2000; Stephens & Curtin, 1995). The Substance Use Risk Profile Scale (SURPS; Woicik et al., 2009) is a brief personality scale that was designed to measure non-overlapping variance along four dimensions of personality that have recently been shown to have particular relevance to substance abuse vulnerability: Anxiety sensitivity (AS), hopelessness (H), impulsivity (IMP), and sensation seeking (SS). Traits related to anxiety disorder vulnerability, such as AS, have been shown to be related to a variety of substance-related behaviors in adults (including alcohol use, anxiolytic/sedative use, and tobacco use), particularly for the purpose of reducing/managing anxiety [see reviews by Stewart, Samoluk, and MacDonald (1999) and Stewart and Kushner (2001)]. Personality traits related to depression vulnerability, such as low self-esteem, introversion, and H, have also been shown to be associated with risk for alcohol dependence (Caspi, Moffitt, Newman, & Silva, 1998) as well as susceptibility to opiate use/misuse (Conrod et al., 2000), and use of alcohol specifically for the management of depression symptoms (Woicik et al., 2009). IMP has been shown to be associated with a wide variety of psychopathology, including antisocial tendencies, poly-substance use, stimulant use/misuse, and drinking problem (Conrod et al., 2000; Finn, Mazas, Justus, & Steinmetz, 2002; Jackson & Sher, 2003). In contrast, SS appears to be more specifically related to quantity and frequency of alcohol consumption in adult and older adolescent samples (Conrod, Stewart, & Comeau, 2006; Conrod et al., 2000; Cooper et al., 1995) likely because SS would be expected to be associated with illegal drug use only when it occurs in combination with high IMP. Until the development of the SURPS, there was no instrument that contained subscales that distinctly and independently assess these four personality risk factors of AS, H, IMP, and SS. Instead, researchers wishing to assess all of these traits would have had to administer at least three different scales. This could be problematic for several reasons. First, currently available scales were not originally designed to measure these constructs independent of one another. Second, use of several scales can be time-consuming (Sher, Wood, Crews, & Vandiver, 1995) which might be particularly problematic in studies with younger adolescents where increased administration time could increase participant dropout due to boredom or fatigue. It might be argued that there is little need to develop a new personality inventory since existing well-validated measures, such as the Tridimensional Personality Questionnaire (TPQ) (Cloninger, 1987b) or the NEO Five Factor Inventory (NEO-FFI) (Ruiz, Pincus, & Dickinson, 2003) which assess broad dimensions of personality, should capture variance related to AS, H, SS, and IMP. However, it appears that the personality constructs of greatest interest to substance use and misuse involve specific, rather than general, trait vulnerability factors (Cloninger, 1987b). For these reasons, a brief single measure tapping all four specific personality vulnerability factors was developed (i.e., the SURPS) (Woicik et al., 2009). The SURPS scales were developed based on a study from a community-recruited sample of adult substance users who were administered a large battery of personality and symptom inventories, including the NEO Five Factor Inventory [NEO-FFI; Costa & McCrae, 1992]; the Sensation Seeking Scale [SSS; Zuckerrnan, 1979]; the trait subscale from the State-Trait Anxiety Inventory [STAI-T; Spielberger, Lushene, Vagg, & Jacobs, 1983]; the Anxiety Sensitivity Index [ASI; Peterson & Reiss, 1992]; the Cognitions Checklist [CCL; Beck, Brown, Steer, Eidelson, & Riskind, 1987]; the Beck Depression Inventory [BDI; Beck, Ward, Mendelson, Mock, & Erbaugh, 1961]; the Self-Esteem Scale [SES; Rosenberg, 1989]; the Posttraumatic Stress Symptom Scale-Self-report [PSS-SR; Foa, Riggs, Dancu, & Rothbaum, 1993]; the Beck Hopelessness Scale [BHS; Beck, Weissman, Lester, & Trexler, 1974]; and the Impulsiveness and Venturesomeness Scale [I.7;

Eysenck & Eysenck, 1978]. Although the content of the selected items were derived conceptually from these established scales, all items were rewritten from the original versions and were put onto a single, consistent Likert-type scale. A factor analysis on these data revealed that four factors reliably emerged: AS, H, IMP, and SS (Conrod et al., 2000). Internal consistency analysis identified those items which were redundant within a subscale or which significantly detracted from the overall subscale alpha and confirmatory factor analyses on two randomly selected sub-samples indicated a stable factor structure with good scale internal consistency for a refined 23-item scale with four subscales (Woicik et al., 2009). Relationships between the subscales and theoretically relevant personality measures indicated that the four constructs assessed by the new 23-item scale represented valid dimensions of the personality constructs on which they were originally based: AS (Peterson & Reiss, 1992), H (Beck et al., 1974), IMP (Eysenck & Eysenck, 1978), and SS (Eysenck & Eysenck, 1978; Zuckerrnan, 1979). Moreover, existing but longer measures of each personality variable did not add additional information above the SURPS scales for predicting substance use and misuse, showing that crucial information is not lost when assessing each of the four traits with the briefer SURPS. The SURPS subscales explain additional variance above the domain scores from the NEO-FFI in predicting alcohol problems, providing evidence of incremental validity. Additional studies suggest that the SURPS subscales correlate with specific patterns of substance use in an undergraduate sample: Finally, twomonth test–retest analysis of the SURPS indicated that each subscale taps personality traits that are relatively stable over time (intra-class correlations [ICCs] ranging from 0.68 [for AS] to 0.88 [for SS]; Woicik et al., 2009). 3. Longitudinal study of SURPS in early adolescence This study will be the first to assess the predictive validity (i.e. prospective association) of the four personality factors measured by the SURPS on the major transitions to substance use in a general early adolescent sample. The present study assessed predictive validity of the SURPS in a longitudinal design. The present study sought to demonstrate both the concurrent and predictive validity of the scale with respect to predicting the onset and escalation of substance use during early adolescence. In addition, although Woicik et al. (2009) established the validity of the SURPS in predicting various alcoholrelated criterion variables in adolescents, and established the validity of the SURPS in predicting the use of other drugs in undergraduates, the validity of the SURPS in predicting other drug use in adolescents remains unknown. Thus, the present study sought to determine whether the SURPS was a useful measure in predicting not only the use of alcohol but also the use of other drugs in adolescence. 4. Method 4.1. Participants The participants in this study were from a three year longitudinal study on risk behavior, the Project on Adolescent Trajectories and Health (PATH) (Krank & Wall, 2006; Krank, Wall, Stewart, Wiers, & Goldman, 2005). Briefly, students were recruited from all grade seven to nine classes in a large school district in western Canada, with a population of about 100,000. After obtaining parental informed consent and student assent, 1315 students completed the survey in the first year. Twelve students were omitted from further analysis because they failed to complete more than 50% of the survey leaving a total first-year sample of 1303.1 The data for the present study included 1139 and 1008 participants in the second and third waves of 1

No participants were excluded in Waves 2 and 3 for this reason.

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46

the study, respectively. All sampling methods and procedures described here were reviewed and approved by the Research Ethics Board of Okanagan University College (now the University of British Columbia, Okanagan Campus) under the Canadian Research Ethics Guidelines of the Tri-councils. At waves two and three, the sample had slightly more females than males (i.e., 55.0%/56.4%). Ethnicity was not measured in the questionnaire, but national census information indicates that the population is predominantly white (about 95%) with the largest minority being aboriginal (3.7%). The majority of participants lived with either their primary parents (70.5%) or a parent and step parent (11.8%), with 14.6% living with a single parent; the rest were in other types of living arrangements. Most parents had graduated from high school (32.0%), received a college diploma (29.4%), or graduated with a University degree (28.7%). Family income status was rated by student self-report as 7.9% below average, 50.4% average and 41.7% above average. Of the 1303 students who completed the first wave, 1139 (87.4%) completed surveys including the SURPS in the second wave 12 months later (Wave 2) and 953 (73.1%) students completed surveys including the SURPS 24 months after the initial survey (Wave 3). An analysis of differences between completers and dropouts showed that dropouts were more likely to be older, be in higher grades, have aboriginal status, have a step father, or report lower income status. Students with both a mother and a father in the household were less likely to drop out than students from single parent families or families with a step parent. In addition, dropouts from year two to year three had higher negative thinking scores (M = 2.02) than completers (M = 1.87, p b 0.01), but did not differ on any other of the subscales of the SURPS. 4.2. Procedure Participants completed the survey in groups of varying sizes ranging from 20 to 70 students.2 Participants were identified each year by an identification code which allowed the data to be linked across survey years but preserved the participants' anonymity. Each survey administration followed the same procedure: identification codes were confirmed and general instructions were read, followed by a series of sections with questions grouped by similarity (see Measures section). Sections were timed and the surveys were completed within 1 h. Surveys were coded using Remark software with the data initially saved in SPSS format for statistical analysis. SPSS version 15.0 was used to run all analyses, other than where indicated. All ambiguous objective marks were reviewed and entered as missing if not resolvable by visual inspection. 4.3. Measures The assessment used in each year was a paper-and-pencil questionnaire that included survey items, word and phrase associations, and a picture drawing task. The survey items included both closed- and open-ended responses. Each questionnaire included the following sections: alcohol, drugs, leisure activities, sex, dating partners, family and friends, measures of perceived safety, violence exposure and neglect, personality scales, and several outcome measures including problem behaviors and health outcomes. The primary items of interest for this report include the SURPS and measures of substance use/misuse. 4.3.1. Substance use Substance use was assessed at each wave using standardized recency questions (i.e., ‘When was the last time you used [each drug]’) 2 Occasionally when participants missed a scheduled session for their school or for a school with few participants, they were tested individually or in smaller groups.

39

with the response options: never, more than a year ago, in the past year, in the past month, and in the past week. The five substance use items assessed at each wave were: alcohol,3 tobacco, marijuana, hallucinogens, and stimulants. In addition, participants were asked to indicate when they had last been ‘drunk’ (self-defined). Each recency score was converted to standard estimates of ever used, past year use, past month use, and past week use. Contingency analysis between these measures and the recency values indicate that past year use accounted for over 90% of the information in the recency measure. The data analyzed and reported here thus utilized past year use, dichotomously scored as yes/no for each of the seven substances. To gain an overall measure of past year involvement in substance use, we also analyzed a substance use index of past year use (referred to as Index of Use) which was the sum of past year use for all drugs analyzed (i.e., possible range of 0–5). Percent past year use and the mean Index of Use with SDs are shown in Table 1 as a function of grade. 4.3.2. Alcohol and drug use problems Alcohol and drug use problems were measured in waves two and three using the CRAFFT (Knight, Sherritt, Harris, Gates, & Chang, 2003). The CRAFFT measure includes six yes or no items designed to indicate problems experienced with alcohol and other drugs in the past year previously validated on youth samples (Kelly, Donovan, Chung, Cook, & Delbridge, 2004; Knight et al., 2003; Levy et al., 2004). The mean and SD CRAFFT scores are shown in Table 1 as a function of grade. 4.3.3. Substance Use Risk Profile Scale (SURPS) The SURPS (Woicik et al., 2009) is a 23 item questionnaire with four subscales: AS, H, IMP, and SS. This questionnaire was developed for use with adults and adolescents. The specific items are shown in Table 2, organized by subscale. 5. Analyses 5.1. Results 5.1.1. Factor analyses The goal was to validate the hypothesized four-factor structure of the SURPS. This was done by first using a confirmatory factor analysis (CFA) to test and refine the model with Wave 2 data. Next, the stability of the final (refined) model over time was confirmed. Specifically, the longitudinal measurement invariance of the final four-factor model across Waves 2 and 3 was assessed using a series of CFAs. The procedures for testing longitudinal measurement invariance recommended by Brown (2006) and Vandenberg and Lance (2000) (see Schmitt and Kuljanin (2008) for recent review) were followed. Mplus version 5.1 (Muthén & Muthén, 1998–2007) was used to conduct all CFAs. A preliminary examination of the data at both Waves 2 and 3 revealed multivariate non-normality (Mardia's (1970) coefficient for skewness and kurtosis: ps b 0.001). Accordingly, the maximum likelihood estimator with robust standard errors (MLR), which is robust to multivariate non-normality, was used to conduct the factor analyses (Muthén & Muthén, 1998–2007). Several fit indices were used to evaluate the CFAs, including the Comparative Fit Index (CFI, Bentler, 1990), Tucker–Lewis Index (TLI, Tucker & Lewis, 1973), Root Mean Square Error of Approximation (RMSEA, Steiger, 1990), and the Standardized Root Mean Square Residual (SRMR, Joreskog, 1993). The Hu and Bentler (1999) criteria were used to 3 In order to ensure that alcohol use represents consumption rather than sipping or tasting, which usually occurs prior to actual drinking, alcohol use was defined as having a standard drink of alcohol (i.e., a bottle of beer, a glass of wine, a shot of liquor, etc.).

40

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46

Table 1 Percentage of students using substances within the past year as a function of grade. In addition, average CRAFFT scores and Index of Drug Use are shown. Only the data collected in the last two waves of the study (grades 8–11) were analyzed in this study. Grade at time of test Seven

Index of Drug Use CRAFFT scorea Alcohol Drunk Marijuana Cigarettes Hallucinogens Stimulants a

Eight

Nine

Ten

Eleven

Mean

(SD)

Mean

(SD)

Mean

(SD)

Mean

(SD)

Mean

(SD)

0.40 – 0.26 0.09 0.07 0.05 0.01 0.00

(0.77) – (0.44) (0.29) (0.26) (0.23) (0.11) (0.07)

0.90 0.88 0.48 0.25 0.19 0.12 0.06 0.03

(1.28) (1.38) (0.50) (0.43) (0.39) (0.32) (0.25) (0.17)

1.36 1.45 0.65 0.44 0.33 0.18 0.11 0.04

(1.41) (1.76) (0.48) (0.50) (0.47) (0.39) (0.31) (0.19)

1.60 1.83 0.77 0.59 0.42 0.19 0.13 0.05

(1.40) (1.78) (0.42) (0.49) (0.49) (0.39) (0.33) (0.21)

1.81 2.04 0.82 0.63 0.45 0.22 0.15 0.07

(1.47) (1.83) (0.39) (0.48) (0.50) (0.41) (0.36) (0.25)

Not measured in the first year of the study.

indicate excellent fit (CFI and TLI ≥ 0.95, RMSEA ≤ 0.06, SRMR ≤ 0.08), while the more traditional criteria (CFI and TLI ≥ 0.90, RMSEA and SRMR ≤ 0.10; Weston, Gore, Chan, & Catalano, 2008) were used to indicate adequate fit. Interpretation of the factor loadings for all models was based on the guidelines set forth by Comrey and Lee (1992). They suggest that loadings greater than 0.71 should be considered excellent, 0.63 very good, 0.55 good, 0.45 fair, and 0.32 poor. The initial model tested with Wave 2 data was based on the results of Woicik et al. (2009) exploratory factory analyses. Woicik et al. found item 16 (“I am interested in experience for its own sake even if

Table 2 Confirmatory factor analysis: Four-factor structure of SURPS items (Wave 2 and Wave 3). Wave 2 (N = 1139)

Hopelessness 1. I am content (R) 4. I am happy (R) 7. I have faith that my future holds great promise (R) 13. I feel proud of my accomplishments (R) 17. I feel that I'm a failure 20. I feel pleasant (R) 23. I am very enthusiastic about my future (R) Scale Cronbach's alpha Impulsivity 2. I often don't think things through before I speak 5. I often involve myself in situations that I later regret being involved in 11. I usually act without stopping to think 15. Generally, I am an impulsive person Scale Cronbach's alpha Sensation seeking 3. I would like to skydive 6. I enjoy new and exciting experiences even if they are unusual 9. I like doing things that frighten me a little 12. I would like to learn how to drive a motorcycle Scale Cronbach's alpha Anxiety sensitivity 8. It's frightening to feel dizzy or faint 10. It frightens me when I feel my heart beat change 14. I get scared when I'm too nervous 18. I get scared when I experience unusual body sensations 21. It scares me when I'm unable to focus on a task Scale Cronbach's alpha

Wave 3 (N = 953)

SL

R2

SL

R2

0.627 0.706 0.656 0.731 0.614 0.708 0.719 0.861

0.393 0.498 0.430 0.534 0.377 0.501 0.517

0.719 0.733 0.802 0.833 0.750 0.723 0.781 0.917

0.518 0.537 0.644 0.694 0.563 0.523 0.610

0.659 0.434 0.622 0.387 0.617 0.380 0.549 0.302 0.913 0.834 0.809 0.654 0.476 0.226 0.428 0.183 0.752 0.693

0.596 0.355 0.615 0.378 0.524 0.274 0.689 0.475 0.633 0.401 0.610 0.372 0.475 0.226 0.609 0.371 0.637 0.722

0.565 0.578 0.590 0.542

0.320 0.334 0.348 0.294

0.617 0.632 0.408 0.451

0.380 0.399 0.167 0.204

0.574 0.329 0.456 0.208 0.702 0.632

Note. SL = Standardized factor loading. All factor loadings and correlations were significant at p b 0.05. The factor analyses used maximum likelihood estimator with robust standard errors (MLR).

it is illegal”), which was intended to load on the SS factor, to be a problematic item (Studies 1 and 2), the removal of which resulted in improved model fit (Study 2). As well, Woicik et al. found evidence supporting the correlation of errors between items 1 and 4, 4 and 20, and 7 and 23. The correlated errors were attributable to the similar wording of the items (see Table 2). Accordingly, our first CFA was performed with 22 items, where a four-factor latent structure was hypothesized (H: 7 indicators; IMP: 5 indicators; SS: 5 indicators; AS: 5 indicators), item 16 was excluded, and the error covariances between items 1 and 4, 4 and 20, and 7 and 23 were estimated. Factor covariances were estimated; however, these were expected to be weak, given the hypothesized orthogonality of the constructs.4 The fit indices of the initial model supported generally adequate fit to the data (χ2 (200, N = 1139) = 714.799; CFI = 0.903 and TLI = 0.888; RMSEA = 0.048; SRMR = 0.057). However, two items “I feel I have to be manipulative to get what I want” (an intended IMP item; item 22), and “I would enjoy hiking long distances in wild and uninhabited territory” (an intended SS item; item 19) were problematic. Defining salient loadings as those items showing a standardized factor loading of ≥0.45 (i.e., “fair” or better), these two items did not have salient loadings on the respective factors (loadings ≤ 0.377). In order to improve the model, these items were trimmed, and all other items were retained on the hypothesized factor. The refined, 20-item four-factor model, as presented in Table 2, resulted in adequate fit to the data (χ2 (161, N = 1139) = 572.738; CFI = 0.918 and TLI = 0.903; RMSEA = 0.047; SRMR = 0.052). In further support of this model, the standardized factor loadings ranged from 0.475 to 0.913 (ps b 0.001), and the variance accounted for in the observed measures by the latent factors ranged from R2 = 0.226 to 0.834. All but two (AS with IMP, p = 0.184 and AS with H, p = 0.077) of the factor covariances were statistically significant (ps b 0.05). However, the correlations between these scales, while statistically significant, were small (rs = −0.104 to 0.250). This suggests weak associations between the factors, as was intended in scale development. The alpha reliability coefficients (Cronbach's α) for the final subscales ranged from 0.637 (SS) to 0.861 (H). While Cronbach's α ≥ 0.700 is the typically recommended standard, with scales

4 A CFA was also conducted on Wave 2 data which included item 16 as an indicator of the SS factor and which constrained all error covariances to zero. The overall model fit was less than adequate (χ2 (224, N = 1139) = 1145.97; CFI = 0.839 and TLI = 0.818; RMSEA = 0.060; SRMR = 0.070), and further examination of item 16 revealed that it should be cross-loaded on H and IMP, thus supporting this as a problematic item. Statistical support also confirmed that the error covariances between items 1 and 4, 4 and 20, and 7 and 23 should be estimated. Specifically, the MIs were comparatively large, ranging from 56.753 to 114.826, and the Satorra–Bentler scaled Chi-square difference test (Muthén & Muthén, 1998–2007) supported a statistically significant improvement in the model fit (relative to the 23-item model) with the addition of the estimated error covariances (χ2diff (21) = 215.318, p b 0.001).

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46

containing less than 10 items, a Cronbach's α ≥ 0.600 is an acceptable indicator of good internal consistency (Loewenthal, 1996). The 20-item four-factor model, with three estimated error covariances was retained as the final model provided the best fit to the data at Wave 2. Next, we tested this model for measurement invariance across the two waves of data. Based on the recommendations of Brown (2006), the first step was to test the adequacy of model fit to the Wave 3 data. The fit indices (χ2 (161, N = 953) = 486.392; CFI = 0.933 and TLI = 0.921; RMSEA = 0.046; SRMR = 0.065) and standardized factor loadings (0.428–0.833; ps b 0.001) suggested fit to the data that approached excellent. In further support of this model, the variance accounted for in the observed measures by the latent factors ranged from R2 = 0.167 to 0.694. All but one (AS with H, p = 0.309) of the factor covariances were statistically significant (ps b 0.05). However, the correlations of the final scales were generally small (rs = −0.116 to −0.364), with the exception of the correlation between H and SS, which was larger than expected (r = −0.509) given the proposed orthogonality of the constructs. Lastly, the Cronbach's αs of the subscales supported good to excellent internal consistency. Support for the model at the second test time permitted further assessment of measurement invariance. Longitudinal measurement invariance was assessed using a single sample where the data from Waves 2 and 3 were included in a single input matrix. This is in contrast to using a multi-group approach. While a potential disadvantage of the single sample approach is its susceptibility to poor model fit due to the large sample (i.e., ‘overpowered’), it has a strong advantage over the multi-group approach as it takes into account the complete data structure. It permits lagged relationships of indicators and factors across time, while also testing within-time covariances (see Brown (2006) and Vandenberg and Lance (2000) for further discussion). Following the more recent recommendations of Brown, we implemented the step-up approach to test stability of the factor structure across time. First, we tested for configural invariance (equal form; identical factor structure), followed by the increasing stringent tests of metric invariance (equality of factor loadings), scalar invariance (equality of indicator intercept), and strict factorial invariance (equality of indicator residuals). Of interest was whether the increasingly constrained models resulted in a statistically significant decrement in model fit, which would suggest non-invariance of the given set of parameters across waves of data. The Satorra–Bentler scaled Chi-square difference test (Muthén & Muthén, 1998–2007) was used. When an overall decrement in model fit was found, modification indices (MIs) were used to permit exploration of partial invariance by iteratively allowing parameters to vary across Waves 2 and 3. A summary of the Χ2diff test and model fit indices is presented in Table 3. The fit indices of the baseline model, which was a test of configural invariance, supported adequate to excellent fit to the data. This suggests equal form or identical factor structure for the data at Wave 2 and Wave 3. The test for metric invariance resulted in a statistically significant decrement in model fit. MIs and an examination of the factor loadings supported the iterative free estimation of items 6, 14, 7 and 23 across time, at which point the model was no longer significantly worse compared to the unconstrained configural model. Thus, 16 of the 20 items had invariant factor loadings across time. Examination of the factor loadings revealed that items 6, 7, and 23 had higher loadings at Wave 3 (vs. Wave 2), while item 14 had a higher loading at Wave 2. The subsequent test of scalar invariance produced a statistically significant decrement in model fit. Iteratively allowing the additional indicator intercepts of items 4 and 10 to vary over time, as suggested by the MIs, resulted in support for partial scalar invariance. The indicator intercept for item 4 was higher at Wave 2 (vs. Wave 3) while the intercept for item 10 was higher at Wave 3. The final test of strict factorial invariance also produced a statistically significant decrement in model fit. Iteratively allowing the additional indicator residuals of items 11, 13, 20 and 5 to vary over time, as suggested by the MIs, resulted in support for partial strict

41

Table 3 Test of longitudinal measurement invariance of the four-factor structure of SURPS items across Wave 2 and Wave 3. Model

Χ2diff (df)

CFI

1. Configural invariance 2. Metric invariance (Model 2 vs. 1) 2a. Partial metric invariancea (Model 2a vs. 1) 3. Scaler invariancea (Model 3 vs. 2a) 3a. Partial scaler invarianceb (Model 3a vs. 2a) 4. Strict factorial invarianceb (Model 4 vs. 3a) 4a. Partial strict factorial invariancec (Model 4a vs. 3a)

– 44.79 (16) p b 0.001 21.88 (12) p = 0.05 ns 31.36 (12) p b 0.01 16.95 (10) p N 0.05 ns 55.59 (14) p b 0.01 11.57 (10) p N 0.05 ns

0.922 0.911 0.033 0.918 0.909 0.034

0.049 0.051

0.920 0.911 0.033

0.050

0.919 0.911 0.033

0.051

0.920 0.912 0.033

0.050

0.915 0.909 0.034

0.051

0.920 0.913 0.033

0.051

TLI

RMSEA SRMR

Note. aFactor loadings/indicator intercepts for items 6, 7, 14, and 23 free to vary across Wave 2 and 3; bindicator intercepts/indicator residual variance for additional items 4 and 10 free to vary across Wave 2 and 3; cindicator residual variance for additional items 5, 11, 13, and 20 free to vary across Waves 2 and 3. Satorra–Bentler scaled Chisquare difference test (Muthén & Muthén, 1998–2007) was used to assess significant decrement in model fit.

factorial invariance. The amount of variance accounted for in items 5 and 11 by the IMP latent factor was larger at Wave 2 (vs. Wave 3), while the amount of variance accounted for in items 13 and 20 by the H latent factor was larger at Wave 3. In sum, partial longitudinal measurement invariance was supported, suggesting that the form of the model generalizes across time, but some of the indicator parameters may vary. The non-invariant results, however, may in part be attributable to overly stringent Χ2 difference testing. As previously noted, our large sample size may have erroneously increased our power to detect even small decrements in model fit. Thus, our test of invariance may be interpreted as somewhat conservative. 5.1.1.1. Scoring of SURPS. Overall, the results of the CFA and metric invariance supported the general factor structure of the SURPS as previously-tested in adult and older adolescent samples, other than the three identified problematic items. Accordingly, we used the subscale scoring recommended by the scale authors (Woicik et al., 2009), with the exception that the three items that did not load on the expected factors were excluded. The subscale scores maintained the range of the individual items (1–5) by calculating the average score for each item on the subscale. 5.1.2. Test–retest reliability Test–retest reliabilities were assessed by calculating intra-class correlation coefficients (ICC) for each subscale across the two test times. The ICC values were significant but modest, ranging from lowest to highest as follows: H, 0.426; AS, 0.502; SS, 0.575; and IMP, 0.646. Although these values are low for “personality” measures, they are within the range for test–retest comparisons across 12 months and for the Big Five personality scales over this age range (Hampson & Goldberg, 2006; Laidra, Allik, Harro, Merenäkk, & Harro, 2006; Pullmann, Raudsepp, & Allik, 2006). As expected, the subscales tapping ‘internalizing’ traits, H and AS, were less stable than those tapping ‘externalizing’ traits, IMP and SS. 5.1.3. Validity in predicting substance use/misuse The variables used in this study to confirm the concurrent and predictive validity of the SURPS were self reports of past year substance use activities in the second and third waves of the PATH study including: alcohol, drunk, marijuana, cigarettes, hallucinogens, stimulants, and index of substance use. In addition, we included the

42

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46

CRAFFT from these waves as indicators of alcohol and drug problems. Given the established influences of gender and developmental phase on adolescent substance use (Finn, 2006; Rohrbach, Sussman, Dent, & Sun, 2005), gender and cohort grade were used as control variables in the present study. 5.1.3.1. Concurrent validity. We assessed concurrent validity of the SURPS subscales in terms of relations with substance use/misuse with generalized linear model (GzLM) analyses for Index of Use and CRAFFT scores (Neal & Simons, 2007) and hierarchical logistic regressions for each past year use measure. In each initial GzLM model, we entered gender and grade followed by a model including all SURPS subscale scores to determine if the SURPS predicted concurrent substance use/misuse over and above demographics. The analyses were replicated for both waves (W2 and W3). The Index of Use GzLM analysis provides an omnibus test for level of substance use. A Poisson distribution was assumed for the GzLM analyses. Table 4 shows the Chi-square for change from the first model to the second at the far left of the table followed on the right by the individual odds ratios for each SURPS subscale score. Odds ratios significantly greater than one indicate increased risk for substance use and odds ratios significantly less than one indicate reduced risk for substance use. The analyses reveal that the addition of the block of SURPS subscales improved the regression significantly for all substance use/misuse measures (all ps b 0.001). The pattern of results was generally consistent across the two waves of the study; in general Wave 2 relations were stronger than those at Wave 3. The pattern of univariate effects indicated that each SURPS subscale had independent predictive value for certain types of substance use/misuse. Increased scores on H, IMP, and SS were strongly and consistently related with higher CRAFFT and Index of Use scores (all p b 0.001). In addition, higher scores on the H, IMP, and SS SURPS subscales were significantly and consistently associated with an increased likelihood of past year alcohol use, drunkenness, marijuana use, tobacco use, and hallucinogen use and stimulant use. Higher AS scores were associated with lower scores on most substance use/misuse measures, although the significance levels were generally lower than for IMP, SS, and H. AS showed no significant relation to tobacco or hallucinogen use at either testing time. AS was significantly associated with lower CRAFFT and Index of Use scores at only Wave 2. 5.1.3.2. Predictive validity. The SURPS subscale scores from Wave 2 were next used to predict Wave 3 substance use/misuse including all past year use measures, the Index of Use, and CRAFFT. The predictive value of the SURPS subscale scores was evaluated using GzLM models (CRAFFT and Index of Use, Poisson distribution) or logistic regression (drunk and past year use measures). The first analysis included

gender and grade, the second analysis included the Wave 2 score on the dependent measure, and the third analysis added the four SURPS subscale scores. This procedure tests the prediction of differences between Wave 2 use and Wave 3 use. Table 5 indicates, using asterisks, which model change scores and SURPS subscale values are significant with the inclusion of the previous year's use/misuse and demographics in the model. Model change statistics were significant, p b 0.001. After controlling Wave 2 use/misuse levels, SS remained a significant predictor for drunkenness, marijuana use, tobacco use, and hallucinogen use as well as for CRAFFT scores and the Index of Use measure. IMP remained a significant predictor for past year alcohol use, tobacco use, hallucinogen use, CRAFFT scores, and the Index of Use measure. H remained a significant independent predictor in the model marijuana use, tobacco use, hallucinogen use, CRAFFT scores, and the Index of Use measure. AS was not a significant independent predictor — either positive or negative. 6. Discussion The present study examined the utility of a four-factor scale of personality that is specifically to substance use in a longitudinal study of the development of substance use in young adolescents. The results support the hypothesis that this brief 23-item scale provides a relatively stable assessment of four factors: hopelessness (H), anxiety sensitivity (AS), impulsivity (IMP), and sensation seeking (SS). Moreover, scores on these factors are strong independent correlates of alcohol and drug use and problems among these youth showing concurrent validity. The longitudinal design allowed us to demonstrate also the predictive validity of the SURPS. In particular, elevated scores on H, IMP, and SS predicted initiation and escalation of substance use over the next 12 months. The stability of the assessment and its subscales was supported by replication of the four-factor structure proposed by Woicik et al. (2009) using confirmatory factor analysis in two separate waves. Given the Woicik et al. (2009) results suggesting that the item referring to “illegal” activities was problematic, it was not included in the proposed model as an SS item. Thus, a 22-item scale was tested. The results across two data collection waves separated by one year supported the proposed factor structure with the exception two items. One item that proved problematic in the SURPS contained language that might be difficult for young teens. Specifically, the term “manipulation” may be a difficult concept for young teens to comprehend, which may explain why this intended IMP item did not strongly load on the intended factor. It is unclear why the “hiking in wild and uninhabited territory” item did not load on SS, but this may be related to the greater likelihood of hiking as a normative activity in this sample from interior BC. This observation suggests the

Table 4 Concurrent prediction by SURPS subscales at both Wave 2 (W2) and Wave 3 (W3) after accounting for effects of gender and grade according to the generalized linear model (Poisson) for the CRAFFT and Index of Use variables and logistic regression for substance use in the past year. Model Chi-square change is shown for the addition of the SURPS subscale scores after including age and gender. The values under the particular scales show the significance and magnitude of the independent contribution of that subscale at each wave. Odds ratios significantly above 1 indicate an increased risk of substance use. Odds ratios significantly below 1 indicate a decreased risk of substance use. Model

Index of Use CRAFFT Alcohol Drunk Marijuana Tobacco Hallucinogens Stimulants

SURPS subscale scores (odds ratio)

Chi-square change

Hopelessness

Anxiety sensitivity

Impulsivity

W2

W3

W2

W3

W2

W3

W2

W3

W2

W3

319.3 395.3 152.0 163.2 158.8 118.9 86.2 34.9

189.6 220.9 88.3 96.5 102.7 77.4 77.8 20.3

1.292 1.364 1.594 1.640 1.763 2.242 2.070 1.481

1.232 1.263 1.519 1.402 1.602 1.664 1.564 1.428

0.906 0.949 0.779 0.785 0.830 0.873 0.811 0.705

0.877 0.933 0.665 0.719 0.806 0.858 0.773 0.642

1.365 1.425 1.647 1.955 1.943 1.859 1.863 1.872

1.419 1.320 1.793 1.726 1.639 1.639 1.586 1.356

1.232 1.408 1.931 1.688 1.887 1.725 1.835 1.811

1.278 1.253 1.434 1.415 1.572 1.727 2.207 1.586

Notes: p b 0.001 in bold, p b 0.01 underlined, p b 0.05 in italics.

Sensation seeking

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46 Table 5 Prospective prediction of substance use/misuse at Wave 3 (W3) by SURPS subscales at Wave 2 (W2) after accounting for effects of gender and grade, and previous year's substance use/misuse. Model change refers to the Chi-square for the addition of the SURPS subscale scores after including age and gender and past year use. The values under the particular subscales show the significance and magnitude of the independent contribution of that subscale at each wave. Log likelihoods significantly above 1 indicate an increased risk of substance use. Log likelihoods significantly below 1 indicate a decreased risk of substance use.

Index of Use CRAFFT Alcohol Drunk Marijuana Tobacco Hallucinogens Stimulants

Model change

SURPS subscale scores (odds ratio)

Chi-square

Hopelessness Anxiety Impulsivity Sensation sensitivity seeking

49.2*** 69.4*** 23.6*** 35.9*** 34.8*** 36.7*** 52.0*** 8.7

1.097** 1.189*** 1.343* 1.216 1.424** 1.426* 1.681** 1.160

0.958 0.967 0.837 0.895 0.873 0.965 0.891 0.707

1.145*** 1.132* 1.494** 1.221 1.241 1.518** 1.928*** 1.480

1.149** 1.117* 1.191 1.748*** 1.688*** 1.878*** 1.985*** 1.137

Notes: Significance levels for model Chi-square and subscale log likelihood changes with demographics and previous year substance use controlled: ***p b 0.001, **p b 0.01, *p b 0.05.

potential for regional or cultural bias in this item. With the exception of these three items, the structural validity of the SURPS scale was supported in this younger adolescent sample. In addition, the design of the study permitted a test of measurement invariance across these two waves (Brown, 2006; Schmitt & Kuljanin, 2008; Vandenberg & Lance, 2000). The analysis supported configural invariance, partial metric invariance, partial scalar invariance, and partial strict factorial invariance. The lack of full support for full metric, scalar, and strict factorial invariance should not be surprising. First, given the sample size, the tests may have been overpowered, thus measuring differences that are significant, but not meaningful. Moreover, the two tests spanned a developmental period with significant social, cognitive, and neurological changes. For example, future orientation increases during adolescence (Steinberg et al., 2009) and thus viewing the future in a negative light, such as giving a low endorsement to items such as “I am very enthusiastic about my future” (item 23) and “I have faith that the future hold great promise” (item 7), may be more relevant to older adolescents' hopelessness. Similarly, older adolescents have more opportunities to achieve “accomplishments” and may attach greater pride to these accomplishments, thus the absence of this pride (item 13) would involve more hopelessness in older than younger adolescents. These examples illustrate plausible ways that social or cognitive changes inherent in adolescent developmental would contribute to individual item variance over time. Nevertheless, the relatively strong metric invariance analysis supports the utility of the SURPS for adolescents. The moderate levels of test–retest reliability did follow the expected pattern, being stronger for internalizing compared to externalizing factors (Woicik et al., 2009). These correlations are within the range for test–retest comparisons for other traits across 12 months and for the Big Five personality scales over this age range (Hampson & Goldberg, 2006; Laidra et al., 2006; Pullmann et al., 2006). Although lower than sometimes reported, the 12 month lapse between tests is considerably longer than usual test–retest intervals. In addition, these “personality traits” may be less stable in young adolescents who are in a significant period of developmental change including major changes in life circumstances and neurological development. It is also possible that substance use may contribute to changes in personality in this age range. Although this is a natural hypothesis in the context of studying the relationship between personality and substance use, it was not possible to test this

43

interesting possibility in this study in the absence of three waves of data to test reciprocal causation. The validity of the four subscales for the SURPS in predicting substance use and misuse was strongly supported in both concurrent and predictive validity tests. In the case of every substance-related dependent measure across both the concurrent and predictive validity analyses, the block of four SURPS measures added to the prediction of substance use/misuse above and beyond the prediction available using age and gender information. In addition, in the predictive validity tests, the block of four SURPS measures added to the prediction of future substance use/misuse above and beyond both demographics and past year use/misuse. Moreover, use of individual substances, poly-substance use, and problem use were independently predicted by each of the SURPS subscales. Of the SURPS personality variables, H, SS and IMP were the most consistent and the strongest predictors of the substance use/misuse criterion variables in both the concurrent and predictive validity analyses. Higher levels of IMP were associated with greater use and misuse for many of the substance-related dependent measures in both concurrent and predictive validity tests. IMP scores were also a strong predictor of the Index of Use measure, tapping use of multiple different drugs in the last year. This pattern is consistent with previous work suggesting IMP as a factor associated with polysubstance use and poly-substance problems (Conrod et al., 2000). SS also proved an independent predictor of many forms of substance use and misuse. SS was a strong and consistent predictor of past year use of alcohol and past year drunkenness. This finding is consistent with previous findings that SS substance abusers are particularly at risk of heavy drinking and alcohol problems (Conrod et al., 2000). Importantly, in the predictive validity analyses, SS proved an important independent predictor over demographics, past year use, and other personality factors (including IMP) in the prospective prediction of past year drunkenness, marijuana use, tobacco use, and hallucinogen use. SS also proved related to substance use problems (CRAFFT scores) and Index of Use (trying multiple substances) in the most stringent predictive validity analyses where past year use/ misuse was controlled. These findings underscore the importance of separate assessment of IMP and SS in determining adolescent's susceptibility to particular substance use problems with SS adolescents at greatest risk for initiation and escalation of heavy drinking (Conrod et al., 2006; Magid & Colder, 2007; Whiteside & Lynam, 2003; Woicik et al., 2009). Consistent with theoretical prediction, in the concurrent validity analyses, H proved a significant independent predictor of past year alcohol use and drunkenness. In terms of predicting future use of specific drugs in predictive validity tests, after controlling past year use, H predicted use of marijuana, tobacco, and hallucinogens. Attesting to the importance of this personality variable in the development of substance use/misuse in teens, in prospective validity tests, H was a predictor of both CRAFFT scores and Index of Use scores independent of SS and IMP. Previous research has shown AS to be positively related to alcohol use and alcohol/drug problems and negatively related to use of marijuana and stimulants (see review by Stewart et al. (1999)). The present findings indicated only a significant negative relationship between AS and many of the substance use measures. This pattern of results suggests that individuals scoring higher on this scale avoid substance use in early adolescence. Stewart et al. (1999) suggested that high AS individuals might avoid the use of drugs like marijuana and stimulants due to their fear of arousal symptoms (stimulants) and of cognitive symptoms like derealization (marijuana) — both of which were observed in the present study in the concurrent validity analyses. Unexpectedly, higher AS levels were also associated with decreased use of alcohol. The unexpected negative relationship between AS and drunkenness may be due to those with AS social concerns avoiding excessive drunkenness for fear of appearing foolish

44

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46

to others or doing things that might lead to embarrassment (Morris, Stewart, & Ham, 2005). Very unexpectedly, higher levels of AS were associated with fewer substance-related problems on the CRAFFT. However, this unexpected effect is consistent with the finding that high AS individuals in this sample of young adolescents had lower levels of drug and alcohol use. AS did not prove to be a significant predictor in the predictive validity analyses, at least after past year substance use was taken into account. Taken together, this pattern of findings suggests that AS, as assessed with the SURPS, is not a risk factor for substance misuse in early adolescence in contrast to the documented relations of AS with substance use and misuse in adults (Stewart & Kushner, 2001). Several explanations are possible. First, the AS scale on the SURPS may not adequately capture the aspects of AS most associated with substance misuse in adolescents. Second, AS may become a more important risk factor for substance misuse at later stages of development. Finally, AS might relate, in this age range, to some other aspects of substance use behavior not assessed in the present study, which might in turn set up high AS adolescents for future substance-related problems. For example, the AS subscale of the SURPS might be related to coping or conformity motives for use of certain drugs like alcohol (Comeau et al., 2001) which might in turn increase risk for the future development of problems with those substances independent of actual substance use levels. Limitations Several potential study limitations should be acknowledged. First, the present study did not include additional measures of the constructs being tapped by the SURPS scales (e.g., the Childhood Anxiety Sensitivity Index (Silverman, Fleisig, Rabian, & Peterson, 1991) in the case of the AS SURPS scale) due to limitations in the time available for students to complete the measures in the PATH study. Thus, it remains for future research to examine the construct validity of the SURPS in the early adolescent age range. Second, the study did not include measures of substance use motives that might relate to the various SURPS personality variables in this age range. Third, although retention rates were quite high in the present study, there were dropouts between the second and third waves of testing. Those dropping out of the study had higher rates of substance use and substance problems consistent with documented evidence on the relationship of substance use/misuse to school dropout in this age range (Oetting & Beauvais, 1990). In addition, completers had lower levels of negative thinking. Given the positive relationship between negative thinking and substance use, this would suggest that the present findings are actually conservative estimates of the validity of the SURPS in predicting substance use/misuse over one year. This limitation in measuring the predictive validity with these high risk students, however, does not detract from the prediction of substance use transition in the general population. Finally, the present study did not include other measures of relevant personality constructs (e.g. Big Five) to test the incremental validity of the SURPS. While Woicik et al. (2009) have demonstrated the incremental validity of the SURPS relative to the NEO-FFI (Costa & McCrae, 1992) domains, the incremental validity relative to the more specific facets assessed by the NEO-PI-R (Costa & McCrae, 1992) remains to be established. Conclusions Despite these caveats, the present findings do provide strong support for the structural validity, internal consistency, moderate oneyear stability, and good concurrent and predictive validity of the SURPS in the early adolescent age range. These findings suggest promise for this brief measure in future research on adolescent substance use behavior. The results also suggest that the SURPS should serve as a very useful tool for selecting adolescents for preventive interventions targeted toward their underlying personality risk for substance abuse.

Such personality-matched approaches have already shown promise as early interventions in reducing substance use behavior and problems among older adolescents (Conrod et al., 2006; Mushquash, Comeau, & Stewart, 2007). Validation of the personality risk model with the SURPS in the present study with younger adolescents suggests that this approach might also have true preventive utility if applied to adolescents as young as 12 years of age, prior to the onset of substance use (Conrod, Castellanos, & Mackie, 2008; Conrod, Castellanos-Ryan, & Strang, 2010). Role of Funding Sources This research was supported by a grant from the Social Sciences and Humanities Research Council of Canada and Canadian Institutes of Health Research awarded to the first and last authors. Development of the Substance Use Risk Profile Scale (SURPS) was supported through grants to the second, third and/or fourth authors from The State University of New York at Stony Brook Research Foundation, the National Health Research Development Program (NHRDP), Health Canada, and the Alcoholic Beverage Medical Research Foundation (ABMRF). The second author was supported through a Killam Research Professorship from the Dalhousie University Faculty of Science at the time of the study. These funding agencies had no further role in study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the paper for publication.

Contributors Author Krank designed the longitudinal study, collected and processed the data, and conducted the regression analyses. Authors Krank and Stewart wrote the first draft and coordinated the final drafts. Author Conrod wrote portions of the Introduction and Discussion. Authors Conrod and Woicik developed the SURPS instrument used in this study. Author O'Connor undertook the confirmatory factor and metric invariance analyses, wrote the Results section for these, and contributed to the discussion of these results. The last author, Anne-Marie Wall, is now deceased. She was involved in the design and data collection for the present study. All authors, with the exception of Wall, reviewed, edited, and have approved the final manuscript.

Conflict of Interest All authors declare that they have no conflicts of interest.

Acknowledgements The authors gratefully acknowledge Daniel Lai, Peter Molloy, and Christine Wekerle, who were collaborators on the PATH study, and Tricia Johnson and Aarin Frigon, who served as project coordinators for the study. The authors would like to thank Abby Goldstein, Jana Atkins, Jessica van Exan, Jonathan Brown, Tara Schoenfeld, Rob Callaway, Tabatha Freimuth, and Adrienne Girling for their research assistance.

References Bates, M. E., & Labouvie, E. W. (1995). Personality environment constellations and alcohol use: A process-oriented study of intraindividual change during adolescence. Psychology of Addictive Behaviors, 9(1), 23−35. Beck, A. T., Brown, G., Steer, R. A., Eidelson, J. I., & Riskind, J. H. (1987). Differentiating anxiety and depression: A test of the cognitive content-specificity hypothesis. Journal of Abnormal Psychology, 96(3), 179−183. Beck, A. T., Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventory for measuring depression. Archives of General Psychiatry, 4, 561−571. Beck, A. T., Weissman, A., Lester, D., & Trexler, L. (1974). The measurement of pessimism: The hopelessness scale. Journal of Consulting and Clinical Psychology, 42(6), 861−865. Bentler, P. M. (1990). Comparative fit indexes in structural models. Psychological Bulletin, 107(2), 238−246. Brown, T. A. (2006). Confirmatory factor analysis for applied research. New York, NY: The Guilford Press. Brunelle, C., Assaad, J., Barrett, S. P., Ávila, C., Conrod, P. J., Tremblay, R. E., et al. (2004). Heightened heart rate response to alcohol intoxication is associated with a rewardseeking personality profile. Alcoholism: Clinical and Experimental Research, 28(3), 394−401. Caspi, A., Moffitt, T. E., Newman, D. L., & Silva, P. A. (1998). In M. E. Hertzig, & E. A. Farber (Eds.), Behavioral observations at age 3 years predict adult psychiatric disorders: Longitudinal evidence from a birth cohort. Philadelphia, PA, US: Brunner/Mazel. Cloninger, C. R. (1987a). Neurogenetic adaptive mechanisms in alcoholism. Science, 236 (4800), 410−416. Cloninger, C. R. (1987b). A systematic method for clinical description and classification of personality variants: A proposal. Archives of General Psychiatry, 44(6), 573−588. Comeau, N., Stewart, S. H., & Loba, P. (2001). The relations of trait anxiety, anxiety sensitivity and sensation seeking to adolescents' motivations for alcohol, cigarette and marijuana use. Addictive Behaviors, 26(6), 803−825. Comrey, A. L., & Lee, H. B. (1992). A first course in factor analysis, 2nd ed. Hillsdale, NJ England: Lawrence Erlbaum Associates, Inc..

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46 Conrod, P. J., Castellanos, N., & Mackie, C. (2008). Personality-targeted interventions delay the growth of adolescent drinking and binge drinking. Journal of Child Psychology & Psychiatry, 49(2), 181−190. Conrod, P. J., Castellanos-Ryan, N., & Strang, J. (2010). Brief, personality-targeted coping skills interventions and survival as a non-drug user over a 2-year period during adolescence. Archives of General Psychiatry, 67(1), 85−93. Conrod, P. J., Pihl, R. O., Stewart, S. H., & Dongier, M. (2000). Validation of a system of classifying female substance abusers on the basis of personality and motivational risk factors for substance abuse. Psychology of Addictive Behaviors, 14(3), 243−256. Conrod, P. J., Pihl, R. O., & Vassileva, J. (1998). Differential sensitivity to alcohol reinforcement in groups of men at risk for distinct alcoholism subtypes. Alcoholism: Clinical & Experimental Research, 22(3), 585−597. Conrod, P. J., Stewart, S. H., & Comeau, N. M. (2006). Efficacy of cognitive-behavioral interventions targeting personality risk factors for youth alcohol misuse. Journal of Clinical Child and Adolescent Psychology, 35(4), 550. Cooper, M. L., Frone, M. R., Russell, M., & Mudar, P. (1995). Drinking to regulate positive and negative emotions: A motivational model of alcohol use. Journal of Personality and Social Psychology, 69(5), 990−1005. Costa, P. T., & McCrae, R. R. (1992). Revised NEO personality inventory (NEO-PI-R) and NEO five-factor inventory (NEO-FFI) professional manual. Odessa, FL: Psychological Assessment Resources. Curran, G. M., White, H. R., & Hansell, S. (1997). Predicting problem drinking: A test of an interactive social learning model. Alcoholism: Clinical and Experimental Research, 21(8), 1379−1390. D'Amico, E. J., Metrik, J., McCarthy, D. M., Frissell, K. C., Appelbaum, M., & Brown, S. A. (2001). Progression into and out of binge drinking among high school students. Psychology of Addictive Behaviors, 15(4), 341−349. Elkins, I. J., King, S. M., McGue, M., & Iacono, W. G. (2006). Personality traits and the development of nicotine, alcohol, and illicit drug disorders: Prospective links from adolescence to young adulthood. Journal of Abnormal Psychology, 115(1), 26−39. Eysenck, S. B., & Eysenck, H. J. (1978). Impulsiveness and venturesomeness: Their position in a dimensional system of personality description. Psychological Reports, 43(3), 1247−1255. Finn, K. V. (2006). Patterns of alcohol and marijuana use at school. Journal of Research on Adolescence, 16(1), 69−77. Finn, P. R., Mazas, C. A., Justus, A. N., & Steinmetz, J. (2002). Early-onset alcoholism with conduct disorder: Go/no go learning deficits, working memory capacity, and personality. Alcoholism: Clinical and Experimental Research, 26(2), 186−206. Flight, J. (2007). Canadian addiction survey: A national survey of Canadians' use of alcohol and other drugs: Substance use by youth. Ottawa: Health Canada. Foa, E. B., Riggs, D. S., Dancu, C. V., & Rothbaum, B. O. (1993). Reliability and validity of a brief instrument for assessing post-traumatic stress disorder. Journal of Traumatic Stress, 6(4), 459−473. Grant, J. D., Scherrer, J. F., Lynskey, M. T., Lyons, M. J., Eisen, S. A., Tsuang, M. T., et al. (2006). Adolescent alcohol use is a risk factor for adult alcohol and drug dependence: Evidence from a twin design. Psychological Medicine, 36(1), 109−118. Hampson, S. E., & Goldberg, L. R. (2006). A first large cohort study of personality trait stability over the 40 years between elementary school and midlife. Journal of Personality and Social Psychology, 91(4), 763−779. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6(1), 1−55. Jackson, K. M., & Sher, K. J. (2003). Alcohol use disorders and psychological distress: A prospective state-trait analysis. Journal of Abnormal Psychology, 112(4), 599−613. Joreskog, K. G. (1993). Testing structural equation models. In K. A. Bollen, & J. S. Long (Eds.), Testing structural equation models (pp. 294−316). Newbury Park, CA: Sage Publications. Kelly, T. M., Donovan, J. E., Chung, T., Cook, R. L., & Delbridge, T. R. (2004). Alcohol use disorders among emergency department-treated older adolescents: A new brief screen (RUFT-cut) using the AUDIT, CAGE, CRAFFT, and RAPS-QF. Alcoholism: Clinical and Experimental Research, 28(5), 746−753. Knight, J. R., Sherritt, L., Harris, S. K., Gates, E. C., & Chang, G. (2003). Validity of brief alcohol screening tests among adolescents: A comparison of the AUDIT, POSIT, CAGE, and CRAFFT. Alcoholism: Clinical and Experimental Research, 27(1), 67−73. Krank, M. D., & Wall, A. (2006). Context and retrieval effects on implicit cognition for substance use. In R. W. Wiers, & A. W. Stacy (Eds.), Handbook of implicit cognition and addiction (pp. 281−292). Thousand Oaks, CA: Sage Publications, Inc.. Krank, M. D., Wall, A., Stewart, S. H., Wiers, R. W., & Goldman, M. S. (2005). Context effects on alcohol cognitions. Alcoholism: Clinical and Experimental Research, 29(2), 196−206. Laidra, K., Allik, J., Harro, M., Merenäkk, L., & Harro, J. (2006). Agreement among adolescents, parents, and teachers on adolescent personality. Assessment, 13(2), 187−196. Levy, S., Sherritt, L., Harris, S. K., Gates, E. C., Holder, D. W., Kulig, J. W., et al. (2004). Test–retest reliability of adolescents' self-report of substance use. Alcoholism: Clinical and Experimental Research, 28(8), 1236−1241. Leyton, M., Boileau, I., Benkelfat, C., Diksic, M., Baker, G., & Dagher, A. (2002). Amphetamine-induced increases in extracellular dopamine, drug wanting and novelty seeking: A PET/[¹¹C]raclopride study in healthy men. Neuropsychopharmacology, 27(6), 1027−1035. Loewenthal, K. M. (1996). An introduction to psychological tests and scales. London UK: UCL Press Limited. MacDonald, A. B., Baker, J. M., Stewart, S. H., & Skinner, M. (2000). Effects of alcohol on the response to hyperventilation of participants high and low in anxiety sensitivity. Alcoholism: Clinical and Experimental Research, 24(11), 1656−1665.

45

Magid, V., & Colder, C. R. (2007). The UPPS impulsive behavior scale: Factor structure and associations with college drinking. Personality and Individual Differences, 43(7), 1927−1937. Mardia, K. V. (1970). Measures of multivariate skewness and kurtosis with applications. Biometrica, 57, 519−530. McGue, M., lacono, W. G., Legrand, L. N., Malone, S., & Elkins, I. (2001). Origins and consequences of age at first drink: I. Associations with substance-use disorders, disinhibitory behavior and psychopathology, and P3 amplitude. Alcoholism: Clinical and Experimental Research, 25(8), 1156−1165. Morgenstern, J., Kahler, C. W., & Epstein, E. (1998). Do treatment process factors mediate the relationship between type A–type B and outcome in 12-step oriented substance abuse treatment? Addiction, 93(12), 1765−1776. Morris, E. P., Stewart, S. H., & Ham, L. S. (2005). The relationship between social anxiety disorder and alcohol use disorders: A critical review. Clinical Psychology Review, 25(6), 734−760. Mushquash, C., Comeau, N. M., & Stewart, S. H. (2007). An alcohol abuse early intervention approach with Mi'kmaq adolescents. The First Peoples Child and Family Review, 3, 17−26. Muthén, L. K., & Muthén, B. O. (1998–2007). MPlus user's guide, 5th ed. Los Angeles, CA: Muthén & Muthén. Neal, D. J., & Simons, J. S. (2007). Inference in regression models of heavily skewed alcohol use data: A comparison of ordinary least squares, generalized linear models, and bootstrap resampling. Psychology of Addictive Behaviors, 21, 441−452. Oetting, E. R., & Beauvais, F. (1990). Adolescent drug use: Findings of national and local surveys. Journal of Consulting and Clinical Psychology, 58(4), 385−394. Perkins, K. A., Gerlach, D., Broge, M., Grobe, J. E., & Wilson, A. (2000). Greater sensitivity to subjective effects of nicotine in nonsmokers high in sensation seeking. Experimental and Clinical Psychopharmacology, 8(4), 462−471. Peterson, R. A., & Reiss, S. (1992). The anxiety sensitivity index manual, 2nded. Worthington, OH: International Diagnostic Services. Pihl, R. O., & Peterson, J. B. (1995). Alcoholism: The role of different motivational systems. Journal of Psychiatry & Neuroscience, 20(5), 372−396. Pullmann, H., Raudsepp, L., & Allik, J. (2006). Stability and change in adolescents' personality: A longitudinal study. European Journal of Personality, 20(6), 447−459. Rohrbach, L. A., Sussman, S., Dent, C. W., & Sun, P. (2005). Tobacco, alcohol, and other drug use among high-risk young people: A five-year longitudinal study from adolescence to emerging adulthood. Journal of Drug Issues, 35(2), 333−355. Rosenberg, M. (1989). Society and the adolescent self-image, rev. ed. Middletown, CT, England: Wesleyan University Press. Ruiz, M. A., Pincus, A. L., & Dickinson, K. A. (2003). NEO PI-R predictors of alcohol use and alcohol-related problems. Journal of Personality Assessment, 81(3), 226−236. Scheier, L. M., Botvin, G. J., & Baker, E. (1997). Risk and protective factors as predictors of adolescent alcohol involvement and transitions in alcohol use: A prospective analysis. Journal of Studies on Alcohol, 58(6), 652−667. Schmitt, N., & Kuljanin, G. (2008). Measurement invariance: Review of practice and implications. Human Resource Management Review, 18, 210−222. Schulenberg, J. E., Bryant, A. L., & O'Malley, P. M. (2004). Taking hold of some kind of life: How developmental tasks relate to trajectories of well-being during the transition to adulthood. Development and Psychopathology, 16(4), 1119−1140. Schulenberg, J. E., Merline, A. C., Johnston, L. D., O'Malley, P. M., Bachman, J. G., & Laetz, V. B. (2005). Trajectories of marijuana use during the transition to adulthood: The big picture based on national panel data. Journal of Drug Issues, 35(2), 255−280. Schulenberg, J. E., O'Malley, P. M., Bachman, J. G., Wadsworth, K. N., & Johnston, L. D. (1996). Getting drunk and growing up: Trajectories of frequent binge drinking during the transition to young adulthood. Journal of Studies on Alcohol, 57(3), 289−304. Sher, K. J., Wood, M. D., Crews, T. M., & Vandiver, P. A. (1995). The tridimensional personality questionnaire: Reliability and validity studies and derivation of a short form. Psychological Assessment, 7(2), 195−208. Silverman, W. K., Fleisig, W., Rabian, B., & Peterson, R. A. (1991). Child anxiety sensitivity index. Journal of Clinical Child Psychology, 20(2), 162−168. Spielberger, C. D., Lushene, R., Vagg, P. R., & Jacobs, G. A. (1983). Manual for the statetrait anxiety inventory (form Y). Palo Alto, CA: Consulting Psychologists Press. Steiger, J. H. (1990). Structural model evaluation and modification: An interval estimation approach. Multivariate Behavioral Research, 25(2), 173−180. Steinberg, L., Graham, S., O'Brien, L., Woolard, J., Cauffman, E., & Banich, M. (2009). Age differences in future orientation and delay discounting. Child Development, 80, 28−44. Stephens, R. S., & Curtin, L. (1995). Alcohol and depression: Effects on mood and biased processing of self-relevant information. Psychology of Addictive Behaviors, 9(4), 211−222. Stewart, S. H., & Kushner, M. G. (2001). Introduction to the special issues on ‘anxiety sensitivity and addictive behaviors’. Addictive Behaviors, 26(6), 775−785. Stewart, S. H., Samoluk, S. B., & MacDonald, A. B. (1999). Anxiety sensitivity and substance use and abuse. In S. Taylor (Ed.), Anxiety sensitivity: Theory, research, and treatment of the fear of anxiety (pp. 287−319). Mahwah, NJ: Lawrence Erlbaum Associates Publishers. Tucker, L. R., & Lewis, C. (1973). A reliability coefficient for maximum likelihood factor analysis. Psychometrika, 38(1), 1−10. Vandenberg, R. J., & Lance, C. E. (2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3, 4−70. Verdejo-García, A., Pérez-García, M., & Bechara, A. (2006). Emotion, decision-making and substance dependence: A somatic-marker model of addiction. Current Neuropharmacology, 4(1), 17−31.

46

M. Krank et al. / Addictive Behaviors 36 (2011) 37–46

Weston, R., Gore, P. A. J., Chan, F., & Catalano, D. (2008). An introduction to using structural equation models in rehabilitation psychology. Rehabilitation Psychology, 53(3), 340−356. Whiteside, S. P., & Lynam, D. R. (2003). Understanding the role of impulsivity and externalizing psychopathology in alcohol abuse: Application of the UPPS impulsive behavior scale. Experimental and Clinical Psychopharmacology, 11(3), 210−217.

Woicik, P. B., Conrod, P. J., Stewart, S. H., & Pihl, R. O. (2009). The Substance Use Risk Profile Scale: A scale measuring traits linked to reinforcement-specific substance use profiles. Addictive Behaviors, 34(12), 1042−1055. Zuckerrnan, M. (1979). Sensation seeking: Beyond the optimal level of arousal. Hillsdale, N.J.: Erlbaum.