Deciding to Dissolve: Individual- and Relationship-Level Predictors of Roommate Breakup

Abstract

The desire to change roommates served as a model of nonromantic relationship dissolution within 115 college roommate dyads. We measured personality, mental health, social/communication patterns, and academic achievement attitudes over the course of a semester, and we used multilevel modeling to estimate individual-level and relationship-level predictors of dissolution. Self-characteristics (e.g., one's own depression), roommate characteristics (e.g., roommate's poor communication), and relationship characteristics (e.g., similarity in attitudes toward competition) each increased desire to end the roommate relationship. In these data, unique contributions from one's own psychological health, one's own and one's roommate's social/communication style, and roommates' attitude similarity predicted relationship dissolution.

ACKNOWLEDGMENTS

This project was supported by Grant R01MH058869 from the National Institute of Mental Health. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Mental Health or the National Institutes of Health.

Notes

¹Certainly contextual factors matter greatly in relationship dissolution, including social support, income and employment, physical distance, social tumult, and so on (Leslie, 1982), but these are generally conceived of and measured externally to the partners. Our analysis is necessarily limited to the kinds of factors we can identify with self-report data.

²There were no significant effects of sex on any of the variables. There were no significant differences between same-race (n = 74) and mixed-race (n = 42) dyads on any of the variables.

³Participants were also informed about how to complete weekly surveys on issues unrelated to this study.

⁴The alphas for Time 1 and Time 2 for the various scales are as follows: Center for Epidemiological Studies Depression Inventory = .86, .89; Anxiety = .91, .94; fear of failure = .73, .86; trust = .81, .88; Dysfunctional Attitudes Scale = .89, .91; self-esteem = .88, .91; unsupportiveness = .84, .94; tolerance = .90, .92; defensive responses = .85, .94; roommate communication = .91, .95; responses–clarifying = .49, .72; responses–root causes = .76, .89; approach achievement = .89, .93; avoidant achievement = .73, .70; CSW–competition = .86, .90; CSW–competence = .78, .86; self-validation = .95, .98.

Note. RSE = Rosenberg Self-Esteem Inventory; CSW = contingencies of self-worth.

*p < .05. **p < .01. ***p < .001.

⁵Varimax rotation was used to maximize the likelihood that our analysis would reveal different patterns of effects (e.g., partner effects, but not actor effects, as predicted by the improvement model hypothesis) for different classes of variable.

⁶This model can be written in general equation form as

where the response yij is the desire to change roommates, x1ij is the actor predictor variable, x2ij is the partner predictor variable, and w1j is the absolute difference score between x1ij and x2ij. A limitation of using multilevel modeling techniques with dyadic data is that random effects may only be estimated for one less than the number of level-1 units (Kenny & Cook, 1999). In the case of dyadic data, only one random effect can be estimated. Hence, models were estimated with random intercepts and fixed slopes. We chose models with random intercepts rather than random slopes because overall model fit improved when the intercept was allowed to vary across dyads compared to when the slopes were allowed to vary across dyads.

Note. Actor effects reflect self-characteristics, partner effects reflect roommate characteristics, and similarity effects reflect relationship characteristics. All intercepts are significant. RSE = Rosenberg Self-Esteem Inventory; CSW = contingencies of self-worth.

†p < .10. *p < .05. **p < .01. ***p < .001.

⁷One could argue that the lack of partner effects is because low self-esteem, anxiety, and depression cannot be detected in normal social interaction, but in fact they can (e.g., Coyne, 1976).