Rational actor-based sociological models (such as two-sided logit and probit models) face a fundamental data problem which seems to limit their applicability for addressing certain questions about large populations. Specifically, outcomes involving voluntary, mutual choice -- such as the formation of marriage, employment or network connections -- are generated conditionally on the particular social contexts available to actors, that is, on the presence of other actors available as potential partners for marriage, employment or networking. While it is feasible to enumerate all potential partners in a workgroup or other small organization, it seems impossible to obtain data on all the potential marriage, employment, or networking partners that a typical adult may have had available prior to the formation of observed partnerships in a large population. Yet this information is critical to understanding which partnerships will actually form.
This presentation will consider what should be done to apply and validate rational actor models in samples from large populations. (Previous statistical work on these models has assumed the population is observed.) The discussion will be motivated by emphasizing the important questions that available data sets seem to call forth, and the interesting and plausible answers that rational actor models provide to these questions under simplifying assumptions. I will argue that a suitable strategy is to begin by assuming that samples from large populations effectively represent local social contexts of choices, for reasons that are perhaps not obvious and are certainly debatable. How this assumption can be tested and refined will be considered. Though a limited set of sampling results from simulated populations will be presented, the discussion will be largely nontechnical.
