KAZUO YAMAGUCHI Unversity of Chicago DENISE B. KANDEL Columbia University and New York State Psychiatric Institute* Similarity between spouses may result from prior similarity (selection) or interpersonal influence (causation) or both. We investigate spouses' mutual influences on marijuana use in a two-wave longitudinal cohort of 490 married pairs, using data obtained twice from each spouse over a 5 1/2 year interval. To estimate processes during marriage free of sample selection bias, we also include marriages that dissolved during the interval, and we analyze the impact of divorce on the drug use of the spouse who was reinterviewed. We test hypotheses to disentangle causation effects of spouse (or event) from selection effects involved in assortative mating (or divorce), using models with and without controls for latent individual propensities to use marijuana. We find that marital selection effects predominate over causation effects and that divorce affects spouses' continued marijuana use. We discuss the implications of the findings for understanding the persistence of drug use in adulthood, gender differences in the relationship of substance use with marriage and divorce, and the study of interpersonal influences. Key Words: assortative mating, divorce, latent-trait model, log-linear analysis, marijuana, spouse influence. Participation in a deviant behavior, such as illicit drug use in adulthood, is related to the timing and quality of participation in the social roles of family and work. In previous studies, we have documented that marijuana use leads to higher rates of premarital cohabitation and divorce and to delays in marriage and parenthood (Yamaguchi & Kandel, 1985a, 1985b). Marriage itself is associated with a decline in drug use (Yamaguchi & Kandel, 1985a). And although marijuana use greatly increases the rate of becoming divorced, divorce does not significantly affect the cessation or resumption of marijuana use. The identification of these relationships leaves many questions unanswered. In particular, the impact of being in a role or leaving a role, i.e., causation, needs to be differentiated from the impact of the characteristics of individuals who remain in or leave the role, i.e., selection. The term causation refers to the socialization effects of being in a role, as well as the termination of these effects when leaving the role. The issue of selection versus causation is fundamental to an understanding of behavioral similarity between individuals in close relationships. Whether in a friendship or a marriage, the same conceptual issues arise. Thus, issues of causation (socialization) and selection also have been raised in relationship to peer influences in adolescence (Aseltine, 1995; Ennett & Bauman, 1994; Kandel, 1978). Similarity observed between members of a dyad could reflect similarity prior to the formation of the relationship, socializing influences as the result of continuing association, or both. To answer these questions requires longitudinal data on interacting dyads. Such data are rare. In a prior phase of this research, we investigated the influence of selection versus socialization on marijuana use in marriage by analyzing crosssectional marital pairs for whom we had retrospective data on age of first drug use (Yamaguchi & Kandel, 1993). We examined the spouses' joint histories of use before marriage and within the last 12 months before the interview, and we determined the extent to which marijuana use was associated over time for each individual and, contemporaneously as well as over time, between spouses. We concluded that wives influenced their husbands' marijuana use more than husbands influenced their wives' use. However, control for unobserved heterogeneity in latent characteristics that affect drug use suggested that the marital effects might be due to selection effects into the marriage. Having subsequently followed these pairs for almost 6 years, we now have prospective longitudinal data on both spouses over time for pairs that remained married and data on one spouse for pairs that dissolved during the interval. …