Organizational Properties from Aggregate Data: Separating Individual and Structural Effects

In this paper, we argue that aggregate measures (means computed on distributions of individual scores) may be valid indicators of organizational properties which also enable an investigator to determine whether statistical relations among organizational measures arise from organization-level, as opposed to individual-level, causal processes. We review certain statistical aggregation issues as these pertain to organizational analysis, and we propose Hauser's path analytic model of analysis of covariance as a device for separating individual and structural effects. These methods are then applied to data gathered in a survey of 20 social service organizations. We specify and estimate a causal model wherein administrative intensity, lateral communications, and decentralization of decision making are endogenous variables. We find that a number of "total" effects on these properties mask quite different--and in some cases contradictory--processes at individual and organization levels. Among other inferences, we suggest that the organizational- and individual-level influences we observe on decentralization raise questions regarding certain widely accepted interpretations of this property. We recommend that analysts working with aggregate measures adopt similar procedures in order to fully exploit their data for the insights that may be gained into multilevel organizational processes.