Occupational Social Status Modeling with Affect Control Theory

Occupational status is an essential concept for stratification scholars. It is a central part of structural inequality, delineating power differences between groups of people and influencing interactions between professions [16]. Understanding how occupations are mentally organized is important to answer questions including how individuals navigate social situations [16], why people choose particular careers [1], and where scarce resources will be allocated [13]. To this end, numerous occupational status polls have asked respondents to rank jobs based on their social standings [2]. These rankings are culturally stable between people even across time. The Harris Poll rating is a prominent poll that will be used in this paper. It asked approximately 1000 adults from 2000 to 2009, representative based on factors including age, gender, education, race, and region, to rate the status of 23 occupations [12]. The final ratings were the mean percentage of respondents over five years who selected “very great prestige”. However, despite the polls and importance of occupational status, there has been little convincing work explaining how participants derive occupational status rankings. One major approach uses a macro stratification framework that focuses on economic power measures, which does not adequately explain how professions like farmers have much higher status than stockbrokers [2]. Using economic power instead of cultural evaluations of esteem, worthiness, and value to society leads to a lack of construct validity. Instead, we use the conceptualization of status as a network of societal deference relations [4] which has been shown to be a more theoretically grounded operationalization of status [3]. According to this model, if one profession repeatedly defers to another, the other profession has the higher status. In this paper, we present a computational derivation of status scores based on the socio-psychological Affect Control Theory and improvements based on probabilistic BayesACT and graph centrality information propagation. These derivations correlate better with status than other approaches and maintain construct validity. This paper specifically demonstrates the graph centrality measure as an extended version of the deference scores presented in [3].