Dynamic Mapping of Class and Occupation: A New Measurement of Class Using Occupational Text Data

Social class is usually measured using aggregate occupational groups. However, this measure of class may encounter the problem of using a fixed occupation-to-class mapping that neglects the important changes occurring within each occupation due to technological change and organizational reform that may change an occupation’s class location over time. This study overcomes this problem by developing a new method to measure an occupation’s class location dynamically using text data and supervised machine learning models based on the neo-Marxist theory of class. Using ONET’s occupational tasks information, I evaluated the class location of detailed occupations since the 2000s, and linked the class information to the ACS and OEWS surveys to map out the class structure of the US labor force. Results suggest important between- and within-class structure changes occurred in the past two decades. First, the study finds that the American class structure has been quite stable in the past two decades when viewed in four aggregate classes (capitalists, owner-operators, managers, and non-managerial workers). However, important shifts occurred within these aggregate class categories. From 2003 to 2020, the share of proletarian workers dropped from 49% to 37%. Decomposition analysis suggests that about 30% of the shrinkage is due to changes in occupation size, while 70% of the shrinkage is caused by within-occupation class shifts where proletarian occupations experienced a "shift up" to become semi-autonomous or supervisory occupations. In other words, many of the proletarian jobs have either experienced an “up-skill” or an expansion of tasks that now require workers to take on some supervisory work.