Deviations from cultural consensus about occupations: The duality of occupation meanings and Americans’ meaning communities

We examine ratings of 642 occupations by a national online sample of U.S respondents in 2019 (Freeland et al., 2020). We analyze the respondents’ ratings of occupations on three dimensions of cultural meaning—evaluation (good versus bad), potency (powerful versus powerless), and activity (lively versus quiet). We take deviations of respondents’ individual ratings from population evaluation, potency and activity estimates, focusing on deviations from consensus rather than consensus itself. Drawing on Breiger’s (1974) work on duality, we examine two projections of the initial rectangular matrix of correlated deviations. Our two projections represent (1) the cultural communities that people form when they differ from consensus in similar ways, and (2) the clusters of occupations that move in similar ways across those subcultures. Correlations among the residuals at the person level are indicators of shared subcultural differences from the mainstream—different ways of meaning-making about what is valuable and worthy about occupational work. At the occupation level, the structure represents schemas for which occupations share common elements and move together when those elements are evaluated differently. We use dyad models to investigate what metrics of occupation similarity predict similarity in deviations from consensus. We find that similarity in affective meaning (evaluation, potency and activity), material requirements, rewards, and work characteristics all predict clustering at the occupation level. Demographic composition of occupations is less important. We find that older respondents, White respondents, and higher income respondents tend to discriminate more between occupations on evaluation and potency. Respondents who are more similar in age have more similar patterns of deviations. However, occupation-level variables are in general much stronger predictors of residual structure than respondent-level variables.