Individuals and Social Structure

Treatments of contextual effects in the social science literature have traditionally focused on statistical phenomena more than on social processes. Typically, the existence of contextual processes has been inferred on the basis of "group-level" effects (as contrasted with "individual-level" effects). This article seeks to redress that imbalance by focusing on underlying processes through which social structure and social interaction may impinge upon individuals. Four alternative generating mechanisms for contextual effects are discussed, along with their implications for model specification and estimation; only one of these, however, is found to be compatible with the notion of modeling individual outcomes as a result of group processes (endogenous feedback). Methods for consistent parameter estimation in endogenous feedback models are presented, based on maximum likelihood estimators for models with lattice autoregressive structure and their extension to models with lattice autocorrelated error structures. Moreover, since the latter often arise by failure to specify contextual processes explicitly, conventional statistical definitions of contextual effects are shown to be confounded with the resulting specification bias.