University of California, Riverside. The simplex ..... Episcopalian (0.9%), Methodist (0.4%), other affiliations (3.5%), and ag- nostic or ...... protection; hospital indemnity: accident and travel: Keogh retirement: office overhead: and student/school,.
Journal of Personality and Social Psychology 1987. Vol. 52, No. 1,134-144
Copyright 1987 by the American Psychological Association, Inc. 0022-3514/87/JOO. 75
Structural-Equation Models of Current Drug Use: Are Appropriate Models So Simple(x)? Ron D. Hays, Keith F. Widaman, M. Robin DiMatteo, and Alan W. Stacy University of California, Riverside The simplex and common-factor models of drug use were compared using maximum-likelihood estimation of latent variable structural models in two samples: a sample of 226 high school students, using ratio-scale measures of current drug use, and a sample of 310 industrial workers and 811 college students, using ordinal-scale measures of current drug use. Latent variables of alcohol, marijuana, enhancer hard drugs, and dampener hard drugs were specified in a series of structural models. Contrary to previous findings with cumulative drug-use data, the common-factor model provided a more acceptable representation of the observed current-use data than did the simplex model in both samples. In addition, the similarity of results across both of these samples supports recent contentions by Huba and Bentler (1982) that quantitatively measured variables are not necessarily superior to qualitative, ordinal indicators in latent variable models of drug use.
One of the clearest findings of the extensive research to date on involvement with drugs is that substantial covariation exists
vorable to marijuana use (marijuana should be legalized, regu-
among the different forms of drug use (Jessor, 1982; Kandel, 1982). Individuals who use one drug are significantly more
juana-using peers were most important for predicting mari-
likely to use other drugs, and the heavier their use of one drug the greater tends to be their involvement with the other drugs.
contact with drug-using peers were most important for hard-
lar marijuana use is not harmful) and association with marijuana use; and poor relations with parents, depression, and drug use.
Recognition of the covariation among various types of drug use
Although conceding that there does tend to be an ordering of
has prompted interest in the developmental progression of drug involvement.
drug involvement that is consistent with the stage theory (Donovan & Jessor, 1983), a particular sequence of drug-use initia-
Two alternative theories have been advanced to describe drug use: the stage theory and the common-influence theory. Advo-
types of drug use are hypothesized to be intercorrelated, but
tion is not required by common-influence theory. Different
cates of the stage theory propose that individuals initiate drug use in a fairly invariant order: first alcohol, then marijuana, and
the sequence and patterning of drug use is believed to depend
finally hard drugs (Huba, 1983). Use of a drug classified at one
diate social context" (Jessor, 1978, p. 77). Thus, several differ-
level or stage of the sequence implies use of those drugs at an earlier but not a later level (e.g., use of marijuana implies use
ent drug-use patterns are consistent with common-influence
of alcohol but not necessarily use of hard drugs). Furthermore, stage theorists propose that involvement with alcohol, mari-
same factors rather than unique factors explain involvement with drugs at each stage (Donovan & Jessor, 1978; Donovan,
juana, and hard drugs is best predicted by different antecedent variables (Kandel, 1980). For example, among a larger set of
Jessor, & Jessor, 1983; Jessor & Jessor, 1980). Within the con-
statistically significant predictors, Kandel, Kessler, and Margu-
different drugs has been found to be related to the same pattern
lies (1978) concluded that prior involvement in minor delinquency and the use of beer or wine and cigarettes were most
of psychosocial variables: personality unconventionality (e.g.,
important for predicting hard-liquor use; beliefs and values fa-
ance of deviance), perceived environment unconventionality
primarily on "differential support and availability in the imme-
theory. In addition, common-influence theorists argue that the
text of problem-behavior theory (Jessor & Jessor, 1977), use of
high value on independence, greater social criticism, and toler(e.g., lower parent and friends controls, more approval, models, and opportunities for deviant behavior), and behavioral unconventionality (e.g., greater involvement in deviant behavior and
Preparation of this article was supported in part by a Grant in Aid of Research to Ron D. Hays from Sigma Xi, the Scientific Research Society. Thanks are due to Angeline E. DiMatteo for data collection and Deanna DiNicola, Ralph Downey, Robert McArthur, Susan McMorris, Jennifer Wong, and Debbie Yee for coding and keypunching assistance. The comments made by two anonymous reviewers of this article and the information provided by David Dante DiNicola are greatly appreciated. Correspondence concerning this article should be addressed to Ron D. Hays, who is now at the Behavioral Sciences Department, Rand Corporation, 1700 Main Street, Santa Monica, California 90406.
less involvement in conventional behavior; see Jessor, 1984). Many independent studies have established that the stage ordering for initiation of drug use is a prevalent pattern (Donovan & Jessor, 1983; Huba, 1983; Huba & Bentler, 1983; Kandel, 1975b). The description of drug-use initiation as a sequence of stages has stimulated theoretical development to explain the process of involvement with drugs (Kandel & Adler, 1982; Newcomb, Huba, & Bentler, 1983; Potvin & Lee, 1980; Rooney & Wright, 1982). Social-learning factors such as differential support for drug use and availability of drugs have been cited as
134
STRUCTURAL EQUATION MODELS OF CURRENT DRUG USE
important explanatory factors (see, e.g., Jessor, 1978; Kandel, 1982). Drug use is said to reflect the content of the normative milieux of the primary groups to which an individual belongs (see, e.g., Krohn, Akers, Radosevich, & Lanza-Kaduce, 1982). That is, individuals who interact regularly with parents and peers who use certain drugs tend to use the same drugs themselves (Huba, Wingard, & Bentler, 1979). Membership in social networks of users is one of the strongest predictors of drug use (Kandel, 1980). Thus, the stage sequence of drug use may emerge because primary reference groups provide models and differential reinforcement for progressively greater involvement in drug use. For example, an alcohol user will experience more social pressure to use marijuana than hard drugs if his or her alcohol-using social group consists of several marijuana users but few users of hard drugs. If this individual becomes a marijuana user and interacts with a revised social group including a larger proportion of marijuana users, this network is likely to contain more hard-drug users and exert more pressure to use hard drugs. The stage sequence of drug use can be viewed as a consequence of decreasing sanctions against and increasing support of harder drug use as one progresses in drug involvement. These social-psychological factors are expected to influence the individual's norms, attitudes, and orientations (Akers, Krohn, Lanza-Kaduce, & Radosevich, 1979). An individual's latitude of acceptance and rejection (Hovland, Harvey, & Sherif, 1957) of various drugs may initially be consistent with societal norms. After marijuana, which may be just within the latitude of acceptance of many persons,' is used, more permissive norms will be experienced and one's latitude of acceptance of hard drugs may expand. Despite the increasing evidence supporting the stage sequence of drug use-initiation, a debate concerning the quality and validity of drug-use measures has recently emerged. The issues that have surfaced in this debate have important implications for the interpretations drawn from drug-use research.
The Drug-Use Measurement Debate The quality of the research data in some studies of adolescent drug use has been questioned (Baumrind, 1983; Martin, 1982). On the one hand, Martin (1982) has criticized the subjective character of the response alternatives (never used, only once, a few times, many times, regularly) used to measure drug use in some studies (such as Huba, Wingard, & Bentler, 1981). Martin (1982) suggested, instead, that categorization should be based on quantitative-frequency data: If the 7th-grade respondent is required to define for himself or herself what constitutes "many times," the researcher cannot assign an internally consistent quantitative meaning to a scale point and the reader cannot equate different researchers' use of scale points to evaluate contradictory findings, (p. 600)
On the other hand, Huba and Bentler (1982) have argued that there is little consensus among researchers about the best way of assessing drug use. They found (Huba, Bentler, & Newcomb, 1981) that the correspondence between pairs of 14 drug-use survey experts who rated the quality of 87 quantity and frequency drug-use items on a unidimensional scale (essential, very necessary, necessary, very desirable, desirable, acceptable,
135
okay, marginal, worthless, avoid) was disturbingly low (average product-moment correlation was only .12). Furthermore, Huba and Bentler (1982) suggested that even if there was a single best way of assessing a manifest variable (MV) of drug use so that it had the lowest conceptual error, it is not clear that such a variable would be demonstrably superior in a LV [latent variable] model, (p. 604)
The lack of consensus among experts in evaluating drug-use items stems from the difficulty in weighing the strengths and weaknesses of different types of questions. For example, the open-ended question, in which respondents indicate the precise number of times they have used a drug, provides a continuous measure of drug use and prevents the bunching together of cases in one or two response categories. Open-ended questions are potentially more sensitive indices of drug use than are closeended measures. However, if respondents are unable to remember the exact number of times they have used a drug, it may be better to have them provide a more qualitative assessment of their usage (Kandel, 1975a). There is some empirical support for the contention that quantitative measures may not be superior to more qualitative measures of drug use in structural models. For example, Garrett and Bahr (1974) investigated the association between selfreport rating and quantity-frequency measures of alcohol use and concluded that quantity-frequency measures were not necessarily better indicators of alcohol use than were the more simple rating measures. In other research, both ordinal rating and ratio quantity-frequency measures of alcohol, tobacco, and marijuana use were found to display rather high and approximately equal levels of valid trait variance in a confirmatory factor-analytic study of drug-use indicators (Stacy, Widaman, Hays, & DiMatteo, 1985). In addition, Huba (1983) reanalyzed data from an earlier study by Single, Kandel, and Faust (1974), which assessed cumulative drug use with quantitative closeended response alternatives (e.g., never used, 1-5 times, fi-9 limes, 10 limes or more), and found results consistent with previous studies (e.g., Huba, Wingard, & Bentler, 1981) that used qualitative indicators: The simplex model fit the data as well as did the common-factor model. Using simulated data, Johnson and Creech (1983) concluded that, in multiple-indicator models, the categorization errors created when continuous variables are measured with ordinal categories were generally "not of sufficient magnitude to strongly bias the estimates of the important parameters" (p. 406). Thus, several studies have confirmed the Huba and Bentler (1982) position that ratio- and interval-scaled measures are not demonstrably superior to ordinal-scaled indicators of drug use in latent variable models. Although the level of measurement of observed measures may have little effect on models of drug use, measures of drug use differ on another important dimension: the time frame of drug-use assessment. On the basis of results of a comprehensive evaluation of drug-use items, a National Institute on Drug Abuse expert panel recommended that drug-use surveys in-
1 Although 67.5% of the high school seniors sampled in 1978 disapproved of regular use of marijuana, only 8.1 % saw "great risk" in trying marijuana once or twice (Petersen, 1980).
136
HAYS, WIDAMAN, DlMATTEO, AND STACY
elude questions on current use (last 30 days and last 12 months) as well as cumulative use (Lettieri, 1981). Although a series of studies examining cumulative drug-use patterns has recently been published (Huba, 1983; Huba & Bentler, 1982, 1983; Huba, Wingard, & Bentler, 1981), it is not known whether structural models found to be consistent with self-reports of cumulative drug use are also consistent with self-reports of current drug use. In a recent study, Huba and Bentler (1984) re-
cally in direct proportion to the frequency of current marijuana use in one study (Single et al., 1974). These empirical results support contentions that drug users show both continued and widening use of drugs as they progress through the stages of drug use. If these contentions accurately describe drug-use behavior, then current-use data may yield a simplex patterning of relations among drug-use latent variables similar to that shown by measures of cumulative drug use. The
ported analyses that included measures of current drug use, but the simplex and common-factor models were not explicitly compared. Information about current-usage patterns is important in a clinical sense because current use influences the diagnosis made and may affect the treatment chosen. In addition, knowledge about current drug-use patterns may help to identify the factors maintaining drug-use behavior beyond initial experimentation.
correlations between alcohol and marijuana latent variables and between marijuana and hard-drug latent variables would be expected to be higher than the correlation between the alcohol and hard-drug latent variables, because the similarity of patterns of drug use would be more alike in the former two cases than in the latter. But if there are significant deviations from
Relation Between Initiation and Current Drug-Use Patterns An examination of initiation of drug use requires a determination of how an individual without a prior history of use first comes to try a drug. This is a different issue from determining what maintains drug-using behavior on a regular basis (cf. Akers, Burgess, & Johnson, 1968). Kandel et al. (1978) have suggested that situational and interpersonal variables are most important for initiation, but intrapsychic variables are most important for increased involvement. "Simple contact does not necessarily mean continued use of a substance . . . it may be inferred that additional factors are involved in continued use of substances" (Pandina & White, 1981, pp. 451-452). Nonetheless, drug use is a behavior that, once initiated, is likely to continue to some degree. For example, Kandel (1982) noted that a large proportion of the adolescents who try illicit drugs remain users of those drugs; more than half (53%) of the high school seniors who had ever used marijuana had used it in the month preceding the interview (Johnston, Bachman, & O'Malley, 1981). The relation between current and cumulative drug use does, however, vary by age and by drug. The ratio of current (last month) users to ever users has been found to be as follows (calculated from data in Miller et al., 1983) for youth (ages 12 to 17), young adults (ages 18 to 25), and older adults (ages 26 and up), respectively: alcohol (0.41,0.72,0.64), cigarettes (0.30, 0.51,0.44), marijuana (0.43,0.43,0.27), stimulants (0.39,0.26, 0.10), cocaine (0.25, 0.24, 0.14), hallucinogens (0.27, 0.08),2 sedatives (0.22, 0.14), tranquilizers (0.18, 0.11), and analgesics (0.17,0.08). In addition, Donovan and Jessor (1983) found that
the general patterns of continued and widening drug use with progression through the stages (e.g., use of alcohol is decreased as marijuana use is initiated and use of marijuana is decreased as use of hard drugs is initiated), departures from the simplex pattern may be found with significant positive or negative direct effects of alcohol use on hard-drug use.
Two Studies of Current Drug Use Because a fundamental issue in structural modeling is the robustness of a model across different samples (Fredricks & Dossett, 1983), we examined the pattern of current drug use in two samples. In one study, ratio-level items inquiring about current drug use were collected from a sample of high school students. In the second study, ordinal-level current drug-use data published by Douglass and Khavari (1978) were analyzed.
Study 1 Method Subjects.
A total of 226 high school students (93 boys, 127 girls, 6
unstated) were surveyed. (Five of these respondents did not complete the entire questionnaire, and the following percentages for demographic variables do not sum to 100% because not all individuals responded to each item.) The mean age of the sample was 16.7 years (SD = 0.8), with a range from 15 to 18 years old. The majority of the sample was white (92.9%), and 3.5% specified a different ethnic status. Twenty-four (10.6%) individuals were in the 10th grade, 99 (43.8%) were in the 1 Ith grade, and 98 (43.4%) were 12th graders. The following religious affiliations were reported: Catholic (73.9%). Baptist (3.5%), unspecified Protestant (3.5%), Presbyterian (2.2%), United Church of Christ (1.3%), Episcopalian (0.9%), Methodist (0.4%), other affiliations (3.5%), and agnostic or atheist (6.6%). Most of the respondents had never been married (96.5%), although two were married (0.9%) and one was divorced (0.4%). The modal category for combined annual parental income chosen by respondents was $30,000-39,999 (26.5%). Combined parental
Daily intake of alcohol increases linearly across all of the groups that drink, as does frequency of drunkenness and of negative consequences due to drinking. There is clearly no substitution of illicit drugs for alcohol shown here. Similarly, frequency of marijuana use increases across all four groups that have had experience with the drug; again "harder" drugs do not serve as substitutes for "softer" drugs. Rather, a deepening of regular substance use appears to go along with a widening of experience in the drug domain, (pp. 548-549)
income reported ranged from less than $3,000 (0.9%) to $110,000 or over (1.3%). Sixty-four percent of the sample perceived themselves to be middle class. Perceived social-class status ranged from upper class (1.3%) to lower class (0.4%). Procedure. A 16-page questionnaire assessing personality, perceived
2
The ratio of current users to ever users could not be computed in the
older-adult subsample for hallucinogens, sedatives, tranquilizers, and
Similar results have been reported by other researchers studying drug use. For example, the rate of liquor use increased dramati-
analgesics because the percentage of current or ever users was less than 0.5% and was therefore not provided by Miller et al. (1983).
STRUCTURAL EQUATION MODELS OF CURRENT DRUG USE environment, health-related behavior, and background variables was administered to students by a teacher at a New England high school in 1981. Students in English classes were asked to volunteer for the survey. Less than 5% of those asked to participate refused to do so. Respondents filled out the questionnaire during a 45-min class period, with most of them finishing in about 30 min. As part of the survey, each respondent was asked to indicate the number of days he or she used beer during the last month. Nine parallel questions dealt with the number of days respondents used wine, liquor, cigarettes, tranquilizers, psychedelics, stimulants, cocaine, marijuana or hashish, and heroin. Each of these questions was open-ended, requiring that respondents specify the exact number of days of use for each drug. Data-analytic strategies. The stage theory and the common-influence theory are embodied in two different structural models. The stage theory hypothesizes that a mathematical simplex model fits the draguse data; the common-influence theory hypothesizes that a commonfactor model fits the data. The simplex model specifies an invariant drug-use sequence; the common-factor model is less restrictive and allows alcohol, marijuana, and hard drugs to be freely intercorrelated. Thus, the simplex model is a special case of the more general commonfactor model, and as noted by Huba, Wingard, and Bentler (1981, p. 191), "both models argue for a unidimensional representation of involvement in drug use at the next higher order of abstraction." Although comparison of the simplex and common-factor models does not permit an evaluation of whether similar or different sets of precursors have causal impact on different types of drug use, comparison of these models does provide a test of whether a set of data is consistent with a particular, restricted model of drug use (simplex model) or is, on the other hand, consistent with a less restrictive structural model in which latent variables for different types of drug use are freely intercorrelated (common-factor model). In this study, the simplex and common-factor models of drug use were compared using maximum-likelihood estimation of latent variable structural models with LISREL v (Joreskog & Sorbom, 1981). Latent variable or structural modeling represents a method of theory testing with correlational data in which a sample of data may be assessed with respect to population models that hypothesize certain relations between constructs of theoretical interest. The goals of latent variable modeling "include both the testing of a proposed model against data and the development of models that adequately account for data" (Huba, Wingard, & Bentler, 1981, p. 182). Latent variable analysis is an advanced form of path analysis that enables the regression of unmeasured latent constructs on one another. The chi-square provided in this procedure requires the assumption of multrvariate normality of the observed variables and indicates whether the correlation matrix reproduced by a given model is significantly different from the observed correlation matrix.3 Statistical nonsignincance implies that a model cannot be rejected on statistical grounds and may provide an adequate representation of the data. Unfortunately, the likelihood of rejecting a model based on chi-square is dependent on sample size. On the one hand, models that explain essentially all of the relevant information in a set of data based on a large sample may still be rejectable on statistical grounds (Bentler & Bonett, 1980). On the other hand, too many models may appear to provide adequate fit to data from small samples if only statistical significance is relied on. To complement the results of the chi-square statistic, a measure of practical fit, delta (proposed by Bentler & Bonett, 1980), not influenced by sample size, was used in this study. Delta, a normed fit index, provides a measure of practical significance because it represents the proportion of statistical information in the data that is accounted for by a model. Bentler and Bonett (1980) suggested that models with delta values less than .90 should not be accepted, because such models can often be improved. We report values of delta for both the maximum likelihood (ML) and unweighted least squares (ULS) solutions (see Footnote 3).
137
It is possible to specify parameters for two latent variable drug-use models in such a way that one (simplex model) is a subset of the other (common factor). In previous research (Huba, 1983; Huba & Bentler, 1982, 1983; Huba, Wingard, & Bentler, 1981), three latent variables have been specified: alcohol, marijuana, and hard drugs. A causal model, equivalent to the correlated common-factor model, differs from the corresponding simplex model in that the former specifies both a direct and an indirect path (through marijuana) from alcohol to harddrug use, whereas the latter hypothesizes no such direct path. Because the simplex model requires fewer parameter estimates and is nested within the common-factor model, it is impossible for the simplex model to fit the data better than the common-factor model does, but a statistical test of difference in fit between the two models may be computed. If it fits the data virtually as well as does the common-factor model (i.e., if there is a statistically nonsignificant difference between the models), then the simplex model can be regarded as substantively correct. More extensive details concerning the procedure for comparing the simplex and common-factor models of drug use have been reported by Huba, Wingard, and Bentler (1981). It is important to note that, given the structural-equation models underlying our analyses, there are many alternate models that could be specified, and one or more of these alternate models may fit the data as well as or better than the models we considered. In this research we tested models similar to those reported in previous drug-use studies and some plausible alternate models.
Results Pearson product-moment correlations between the 10 druguse items were computed.4 These correlations are provided in Table 1. More than half of the correlations in the table were significant (28 of 45), and most were either moderate or large in size, indicating substantial covariation among different types of drug use. Model OA, a null model hypothesizing no common factors, was estimated first. The null model is a test of whether the correlation matrix for the raw data is equivalent to an identity matrix. As shown in Table 2, the null model was rejectable statistically, x2(45, N = 218) = 849.69, p < .05. Thus, the hypothesis of lack of correlation among the drug-use items was not tenable. Next, the initial common-factor model. Model 1 A, was formulated. Drawing on previous research, three latent variables were hypothesized: alcohol, marijuana, and hard drugs. Beer, wine, liquor, and cigarettes were estimated as indicators of alcohol; marijuana or hashish and cigarettes as indicators of marijuana;
3 The drug-use data in this study were rather skewed and kurtotic, and this may have affected the chi-square statistics and standard errors we report. The models presented in this article were also analyzed using the method of unweighted least squares because it does not require distributional assumptions (Joreskog & Sorbom, 1981). Results obtained from both methods of estimation were compared to examine the degree to which the results were robust across different analytic methods. The substantive conclusions from both methods of analysis were invariant. 4 Univariate outliers were defined as standard scores whose absolute value exceeded 3 for any variable. After ensuring that coding and keypunching errors were not responsible for outlying values, these outliers were recoded to a standard-score absolute value of 3. This receding preserves "the deviancy of a case without allowing it to be so deviant that it perturbs correlation" (Tabachnick & Fidell. 1983, p. 76). In this sample, this receding substantially reduced the skewness and kurtosis of the drug-use items, leading to more normal distributions.
138
HAYS, WIDAMAN, DlMATTEO, AND STACY
>0 T O O
I «,«,«.«,* •.0, 912-914. Kandel, D. (1980). Drug and drinking behavior among youth. Annual Review of Sociology, 6, 235-285. Kandel, D. (1982). Epidemiological and psychosocial perspectives on adolescent drug use. Journal of the American Academy of Child Psychiatry, 21, 328-347. Kandel, D., & Adler, I. (1982). Socialization into marijuana use among French adolescents: A cross-cultural comparison with the United States. Journal of Health and Social Behavior. 23. 295-309. Kandel, D. B., Kessler, R. C., & Margulies, R. Z. (1978). Antecedents of adolescent initiation into stages of drug use: A developmental analys\s. Journal of Youth and Adolescence, 7, 13-40. Krohn, M. D., Akers, R. L., Radosevich, M. J., & Lanza-Kaduce, L. (1982). Norm qualities and adolescent drinking and drug behavior: The effects of norm quality and reference group on using and abusing alcohol and marijuana. Journal of Drug Issues, 12, 343-359. Lettieri, D. J. (1981). Recommendations from a science administrative perspective. In G. J. Huba, P. M. Bentler, & M. D. Newcomb (Eds.),
144
HAYS, WIDAMAN, DlMATTEO, AND STACY
Assessing marijuana consequences (pp. 111-128). Rockville, MD: National Institute on Drug Abuse. Martin, J. A. (1982). Application of structural modeling with latent variables to adolescent drug use: A reply to Huba, Wingard, and Bentler. Journal of Personality and Social Psychology, 43, 598-603. Miller. J. D., Cisin, I. H., Gardner-Keaton, H., Harrell, A. Y, Wirtz, P. W., Abelson, H. I., & Fishburne, P. M. (1983). National survey on drug abuse: Main findings 1982. Washington, DC: U.S. Government Printing Office. Newcomb, M. D., Huba, G. J., & Bentler, P. M. (1983). Mothers' influence on the drug use of their children: Confirmatory tests of direct modeling and mediational theories. Developmental Psychology, 19, 714-726. Pandina, R. J., & White, H. R. (1981). Patterns of alcohol and drug use of adolescent students and adolescents in treatment. Journal of Studies on Alcohol, 42, 441 -456. Petersen, R. C. (Ed.) (1980). Marijuana research findings: 1980. Washington, DC: U.S. Government Printing Office.
Potvin, R., & Lee, C. (1980). Multistage path models of adolescent alcohol and drug use. Journal of Studies on Alcohol. 41, 531-542. Rooney, J. F., & Wright, T. L. (1982). An extension of lessor and lessor's problem behavior theory from marijuana to cigarette use. International Journal of the Addictions. 17, 1273-1287. Single, E., Kandel, D., & Faust, R. (1974). Patterns of drug use in high school. Journal of Health and Social Behavior, 74, 344-357. Stacy, A. W, Widaman, K. F., Hays, R., & DiMatteo, M. R. (1985). Validity of self-reports of alcohol and other drug use: A multitraitmultimethod assessment. Journal of Personality and Social Psychology, 49, 219-232. Tabachnick, B. G., & Fidell, L. S. (1983). Using multivariate statistics. New York: Harper & Row. Wingard, J. A., Huba, G. J., & Bentler, P. M. (1982). Psychosomatic symptoms and substance use in women. In L. Bickman (Ed.), Applied Social Psychology Annual: Volume 3 (pp. 181-215). Beverly Hills, CA: Sage. Received December 12, 1983 Revision received February 26, 1985 •
Low Publication Prices for APA Members and Affiliates Keeping You Up-to-Date All APA members (Fellows, Members, and Associates) receive—as part of their annual dues — subscriptions to the American Psychologist, the APA Monitor, and Psychology Today. High School Teacher and Student Affiliates receive subscriptions to the APA Monitor and Psychology Today, and they can subscribe lo the American Psychologist at a significantly reduced rate In addition, all members and affiliates are eligible for savings of up to 50% on other APA Journals, as well as significant discounts on subscriptions from cooperating societies and publishers (e.g., the British Psychological Society, the American Sociological Association, and Human Sciences Press). Essential Resources APA members and affiliates receive special rates for purchases of APA books, including the Publication Manual of the APA. the Master Lee lures, and APA's Guide to Research Support. Other Benefits of Membership Membership in APA also provides eligibility for low-cost insurance plans covering life; medical and income protection; hospital indemnity: accident and travel: Keogh retirement: office overhead: and student/school, professional, and liability. For more information, write to American Psychological Association. Membership Records. 1200 Seventeenth Street NW. Washington. DC 20036
