Marijuana is the most commonly used illicit drug in the U.S. Approximately 46% of college students report having tried marijuana, 27% report use contained by the past year, and 16% report previous 30-day use (1); thus, a significant proportion of college students use marijuana. Furthermore, problems associated with use are, unluckily, not uncommon. For example, short-term cognitive impairments and impairment contained by educational behaviour have be associated with bulky marijuana use (2-4). Given this, identification of individuals who use marijuana and report associated problems is an far-reaching issue. However, despite the relative prevalence of use, many college students hold never tried marijuana or do not currently use it. As a result, an assessment of marijuana-related problems will yield a mixed distribution near a high number of zero-values. Some respondents will report not experiencing problems because they in reality did not use marijuana at all during the assessment pane and those who do use the drug will report a range of problems including, for some users, nought. Identifying variables that predict nonusers as well as identify predictors of the number of problems experienced among users are both of interest. Zero-inflated Poisson (ZIP; e.g., Lambert (5)) and a subsequent generalization of ZIP, Zero-inflated cynical binomial regression (ZINB; e.g., Heilbron (6)) are two statistical techniques that allow one to accomplish both of these objectives within a single analysis. They, thus, represent parsimonious procedures that allow one to examine effects over the full distribution. These mixed distributions are a common element in research investigating risk behaviors near low base rates (6).
Distributions of risk behaviors surrounded by general populations frequently will hold a large number of zero-values. That is, a big proportion will not engage within the targeted risk behavior, while a smaller proportion of at risk individuals will report varying levels of the risk behavior and associated consequences. Such distributions pose difficulties for adjectives statistical methods based upon average distributions. ZINB regression models are one method for analyzing such data. ZINB models assume two distinct populations; one where the target behavior is always elsewhere, the other in which the target behavior can be any integer, including zilch (6). More specifically, in the current study, one population would other score not anything on a marijuana problems measure because they did not use marijuana during the assessed time spell (current nonusers) and the other population could score any plus, including zero, because they are adjectives in the risk behavior (users--who may or may not experience problems). Thus, the prediction of counts is conditional upon the probability of the values one from a hypothetical subsample of participants that are predicted to "always" gain zero on the weigh. The model allows one to either use different sets of predictors to predict the two criterions (i.e., other zero-values and counts) or to utilize the same predictor set and evaluate whether variables are differentially associated next to the respective criterions (7).
For the current study, we employed the same predictor set to predict zero-values (i.e., current nonusers) and counts (i.e., number of problems among expected users) within order to examine the differential predictive power of the variables of interest. Such differentiation is of supposed interest in substance use research. That is, identifying both the types of variables that are associated beside use initiation or low-level experimentation as well as the types of variables that are primarily associated beside the prediction of use problems represents a common aspiration in substance use research. One model of substance use proposes that social-environmental variables are associated primarily near use initiation or low-level use while psychobiological variables are more strongly associated with use-related problems (8, 9). The role of psychosocial variables, such as use motives, expectancies, or perceived use utility do not clearly fit into this dichotomy. Indeed, these psychosocial variables hold relations with both use and use-related problems (10-12). This study examined one social-environmental unsettled (social norms), one psychosocial variable (perceived marijuana use utility), and one biopsychological unstable (impulsivity). Social norms are a social-environmental undependable consistently associated with marijuana use (13, 14). Perceived use-utility is a psychosocial changeable more associated with marijuana use than problems (12). In contrast, impulsivity is a biopsychological irregular more associated with problematic use (15).
Social norm may be defined as either the actual or perceived behavior of individuals in social networks as in good health as the group member's attitudes toward target behaviors (i.e., whether group member think one should absorb in the behavior). In the present study, social norm are represented by both the perceived marijuana use behavior and attitudes of peers. Social normative variables frequently are associated with marijuana use (13, 14, 16). Although the basis for this relationship traditionally has be attributed to the influence of social networks on use behavior, recent research on social norms and alcohol suggests that screening effects (i.e., choosing social networks with similar use practices) may also be of exigency (17). Indeed, selection of marijuana using peer groups may be influenced by variables such as sensation seeking (13) and house relations (14). Such selection effects suggest that affective, cognitive, and social normative variables are not independent of respectively other. Individuals may be choosing social networks that not only hold similar marijuana use practices but likely share adjectives beliefs about the costs and benefits of marijuana. The potential interdependence of affective, cognitive, and social normative predictors of marijuana use make their concurrent assessment of interest. In a study on drug refusal, peer influence is cited as a stronger influence on drug use decisions by low-level users while heavier drug users are more imagined to cite emotional determinants and seldom cited peer influence as a factor (18). Thus, social norm may be more strongly associated with use initiation and low-level use a bit than use-related problems.
Marijuana use, like heaps behaviors, may be influenced by the perceived costs and benefits of use. Evaluation of marijuana and other drug use has be operationalized in diverse ways, range from the global evaluation of attitudinal constructs (19) to specific expectancies of drug effects (20) to subjective expected utility models explicitly examining cost and benefits (21). An extramural way of evaluating the perceived utility of drug use is to examine it in relation to personal goal (22). For instance, personal strivings are ongoing goals that individuals are characteristically trying to get done through their behavior (23). Drug use is expected to increase to the extent that it is congruent with the attainment of such valued goal. Previous cross-sectional research has indicated that perceived conflict/utility of marijuana use in achieve personal strivings is associated with marijuana use initiation as okay as frequency and problems (12). This study seeks to somewhat replicate this finding in a multivariate context.
Impulsivity is related to difficulty with the restraint of one's own behavioral and excited responses (24). Impulsivity, although commonly referred to as "behavioral" undercontrol, also may be described as an over-reliance on affective rather than cognitive cues (25). Impulsivity have well-documented relations with substance use problems (8, 26, 27). Impulsivity have evidenced direct relations with marijuana-related problems above and beyond use frequency (15). Thus, impulsivity may probable be associated with use-related problems among expected users.
PURPOSE
The purpose of this study is to examine associations between social norm, impulsivity, perceived use utility and marijuana-related problems in a indication of undergraduates. Zero-inflated negative binomial regression is used to predict the current nonusers from the users in the indication, as well as the number of problems for the predicted users. Based on previous research, social norm are hypothesized to predict current nonusers, while impulsivity is expected to be associated with the number of problems experienced by the predicted users. Perceived use utility is hypothesized to be a significant predictor of both current nonusers as capably as number of problems.
METHOD
Participants
Participants included 292 students at a small state university; all participated within research for partial fulfillment of course requirements. Women made up 70% of the token. The sample range in age from 18 to 26 (M = 19.69, SD = 1.56); 94% be White, 1% Black, 2% Asian, 1% Native American, and 1% multiracial.
Measures
Marijuana Use and Problems
Lifetime marijuana use was assessed by a 7-point anchored rating scramble (0 = no use, 6 = more than 300 days). Marijuana use in the closing 30 days was assessed by a 9-point anchored rating scramble (0 = no use, 8 = more than once a day).
Marijuana-related problems in the ending 30 days were assessed by 23 items. Items be rated on a 5-point scramble ranging from 0 (never) to 4 (more than 10 times). This go up was designed for adolescents and, thus, is appropriate for this population. This amount is internally consistent (alpha = .86) and has evidenced expected relations next to marijuana use in previous research (11, 27). Sample items included "not competent to do your homework or study for a test," and "feel physically or psychologically dependent on marijuana."
Impulsivity
Eysenck's Impulsivity Scale (29) includes 24 items assessing lack of control over behavior; respectively item is dichotomous. The alpha coefficients for men and women exceed .82 (29). Two items dealing specifically with drug use be excluded, yielding a 22-item size. The alpha coefficient in this taste was .78. Sample items included "Do you habitually do and say things lacking stopping to think?" and "Are you an hot-headed person?"
Social Norms
Social norm were assessed by the following three items:
1. Number of friends who use marijuana: 7-point anchored rating ascend none (1) to all (7).
2. Friends' attitude toward participant using marijuana once a month or smaller amount: strongly disapprove (1) to strongly approve (5).
3. Friends' attitude toward participant using marijuana twice a month or more: strongly disapprove (1) to strongly approve (5).
The mean of the standardized items be used (alpha = .90).
Strivings Assessment
Personal strivings are "goals that sprawl directly behind individuals' behavioral choices (i.e., what an individual is characteristically trying to do)" (30). In the personal strivings assessment, participant first listed 10 personal strivings, near the instructions describing a personal striving as "an objective you are typically trying to accomplish." Participants be given examples such as "trying to be physically attractive," "trying to seek out contemporary and exciting experiences," and "trying to avoid being notice by others." Participants were instructed to focus of actual instances of their behavior and to base their results on the actual intention of the behavior. Personal strivings be found to be stable in college students; 45% of the strivings tabled at initial assessment were scheduled again 18 months later (31). The remainder of the assessment focused on the five strivings that the participant identified as most descriptive of themselves.
To assess perceived conflict/utility between strivings and marijuana use, the five strivings were enter into a matrix that included five columns to represent levels of marijuana use: (1) prudence, (2) at least once a year but smaller amount than once a month, (3) at least once a month but smaller number than once a week, (4) 1-3 days a week, and (5) most every day. The participant rated the extent to which respectively level of marijuana use would help out or hinder the attainment of respectively personal striving using a 5-point scale (-2 = especially harmful effect, +2 = terribly helpful effect) and a gain was enter in respectively cell. A marijuana use-striving conflict/utility score be created for each personal striving (reverse scoring the thriftiness column). Finally these sums were combined into a single marijuana use-strivings conflict/ utility (i.e., use utility) chalk up (alpha = .92). Higher scores correspond to greater perceived utility of marijuana within achieving goal. Lower (more negative) scores correspond to greater perceived conflict between marijuana use and aspiration attainment.
Procedure
Participants completed questionnaires online in small groups beside adequate space to ensure privacy within a computer lab under the supervision of a research assistant. Previous research supports the reliability and truthfulness of Internet-based assessment of drug use (32). All participants provided written informed consent. Participants generate a unique code for themselves and did not place their describe on the questionnaires, thus, ensure their anonymity. The assessment session lasted approximately one hour. Two previous manuscript focusing on alcohol use have be derived from this dataset (33, 34).
RESULTS
Descriptive Statistics
Approximately 49% of the sample reported have used marijuana at least once contained by their lifetime and 21% reported use in times gone by 30 days. Average use in the second 30 days among those who had tried marijuana be 1-2 days (rating scale M = 1.48, SD = 2.25). The be set to on the problems measure for those who have tried marijuana was 3.41 (SD = 6.49). Thus, a colossal percentage of participants reported no marijuana use and the penny-pinching number of problems was outstandingly low. ZINB models are designed for examining this type of distribution. Table 1 presents the summary statistics and correlation matrix for the predictors.
ZINB Model
The ZINB regression model was estimated beside the ZINB command in Stata 8.0 (35) which solves for parameter estimates using maximum chance estimation. ZINB models have two sets of predictors, one set is used to predict zero-values (current nonusers surrounded by this case) and one is used to predict counts among the predicted users. All cases are used in both analyses but are weighted base on the results of the logistic component of the model (see below). In this manner, the model is predicting a zero-score to be generate from one of two populations. More specifically, one set of predictors is used in a logistic model, within which the likelihood of the inspection being a current nonuser is computed, and a second set of predictors is used within a negative binomial model that predicts the count of expected problems, which may be zilch or some positive integer. Thus, the probability that an observation is other zero is modeled by probability, [omega], and the probability that the supervision follows a negative binomial distribution near mean [lambda] is (1-[omega]). More specifically,
P(Y = O) = [omega] (1)
P(Y ~ Negative Binomial ([lambda], [alpha])) = (1 - [omega]) (2)
conceding the following distribution of counts:
P(0) = [omega] + (1 - [omega]) x F(0|[lambda]) (3)
P(k) = (1 - [omega])) x F(k|[lambda]) (4)
where F represents the quotation distribution (negative binomial with fixed parameter [alpha]), [omega] represents the predicted probability of self always-zero, modeled by the logistic component of the model, and [lambda] represents the predicted mean of the refusal binomial component of the model. While the data modeled contained by this study are not true count data, this analytic technique is appropriate for two reason: first, the data are distributed exclusively on the nonnegative integers and tend to show heteroskedasticity (exactly resembling true count data); second, the data appear to be a true mixture model (thus, the need for zero-inflation). As such, even though the notes technically are not generated by a count process, the resultant distribution have the important characteristics expected of a count process and, thus, a count model is appropriate.
Gender, social norm, use utility, and impulsivity were included as predictors in both components of the model (i.e., prediction of zero-values as capably as the number of problems among the predicted users). Thus, the two-part model was parameterized as
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)
The odds ratio for the full ZINB model was [chi square] (9) = 109.47, p < .001; maximum possibility [R.sup.2] = .31, indicating that the overall model was significant. The maximum-likelihood [R.sup.2] is a standard of fit that is analogous to the coefficient of determination ([R.sup.2]) in familiar least squares (OLS) regression (e.g., Hardin and Hilbe, (36)). Both the logistic component of the model (LR [chi square](4) = 41.21, p < .0001) and the glum binomial component of the model (LR [chi square](4) = 22.25, p = .0002) were significant, indicating that the prediction of current nonusers and the prediction of marijuana-related problems be both significant.
Furthermore, support for the ZINB model over other possible count-data models was strong. The Vuong question paper for nonnested models supported the use of a zero-inflated model over a standard negative binomial model, z = 3.51, independent p = .0002, and the LR test for overdispersion also be significant (LR [chi square](1) = 152.96, p < .0001) demonstrating that a ZIP model would be inappropriate.
With regard to the hypothesized predictors in the ZINB regression model, just social norms predicted zero-scores (i.e., expected current nonusers). Perceived use utility, impulsivity, and femininity were not significant predictors of zero-scores. In contrast, perceived use utility be significantly positively associated with number of problems among expected users. Social norm, impulsivity, and gender be not significant predictors of number of problems among expected users. Full results of the regression analysis are presented in Table 2.
DISCUSSION
The results demonstrate that social norm and perceived use utility are related to nonuse and marijuana-related problems, respectively, among college students. The primary strength of this study is the use of ZINB regression to simultaneously predict current nonusers as well as the predicted count of marijuana-related problems among expected users. Differential results emerge in language of statistical predictors of nonuse versus predicted marijuana-related problems; in focused, social norms differentiated expected nonusers from users, while perceived marijuana use utility predicted the number of problems contained by users. These results generally are consistent next to models that propose social-environmental variables as being more associated next to use initiation and low-level use and biopsychological variables as being more associated next to use-related problems. However, it is important to document that such a result would not have be observable within the framework of the more traditional OLS regression analysis. Thus, the statistical modeling employed here study allowed for the emergence of theoretically consistent results.
Examining respectively set of predictors more closely, both social norms and perceived use utility be hypothesized to predict current nonusers. However, only social norm and not perceived use utility was a significant predictor of current nonusers. In previous research, use utility have been significantly associated beside use initiation (12). In the current study, use utility and social norms be fairly outstandingly correlated, which may have contributed to the observed difference surrounded by the results. Much like traditional OLS regression, ZINB regression is susceptible to problems beside multicollinearity, and these findings may be due to such a result.
As hypothesized, perceived use utility was associated beside number of marijuana-related problems among expected users. Previous cross-sectional research has observed significant association between perceived use utility and marijuana-related problems (12). The current study provides a partial replication of this relationship contained by a multivariate context.
Impulsivity was hypothesized to be significantly associated beside problems among expected users and to not be a significant predictor of current nonusers. However, impulsivity was associated marginally beside both the number of problems among expected users and the prediction of current nonusers (p's < .07). Thus, the pattern of relationships did not conform to the hypothesis.
Several limitations should be noted. Marijuana use be quite low within the sample. Although the rates be equal to that reported in national sample, the extent to which these relationships hold among samples near higher rates of use and problems will stipulation to be determined in future studies. Also, the token was predominantly women and White, and sweeping statement of the results to populations with different demographic characteristics desires to be determined. Finally, the cross-sectional analysis precludes causal interpretations. For example, longitudinal studies are needed to take to mean the relation between social norms and marijuana use behavior over time and, thus, examine the relative strength of social influence versus social test effects.
The current study employed ZINB regression to predict marijuana-related problems in a mixed distribution of current users and nonusers within a sample of college students. The analysis approach provides a parsimonious agency to analyze risk behaviors with low bottom rates. Furthermore, the analyses allowed for a theoretical separation of prediction of users versus nonusers, and predicted marijuana-related problems among predicted users. Results indicated that social norm predicted nonusers, while perceived use utility predicted the number of problems reported by expected users. Results generally be consistent with theories of the differential association of social-environmental variables and biopsychological variables near use and problems, respectively.
No comments:
Post a Comment