Background: A rich layer of empirical research detailing the effects of social networks on attitudes and behavior has accumulated in demography over the last 20 years. But there is limited data on social interaction in general, and odd inconsistencies between network data and observational data in terms of the level of heterophily.

Objective: To document actual social interaction and identify covariates of two types of heterophilous interaction - with a person of the opposite sex and a member of a different extended family.

Methods: Over a nine-hour period, all social interaction of a sample of 48 adult men and women in a single village in Malawi’s Central Province was logged (N=1811), along with interactant characteristics. Analysis focuses on heterophilous interaction across gender and kin lines. Two-level mixed models explore covariation between types of interaction and characteristics of interactants and setting.

Results: There is heterogeneity in interactional patterns across individuals, lengthier interactions with kin (than non-kin) and unrelated confidants, a crowding-out effect of large families on interaction with non-kin, fewer but longer interactions among men than women, and more frequent cross-gender than same-gender interaction.

Conclusions: Differences between reported social networks and observed social interaction raise questions about how well standard conversational network data capture actual patterns of social interaction, whether observed patterns may analytically substitute for, modify, or interact with conversational network effects. The collection of actual interactional data should be scaled up to address these questions.