Figure 7
By treating the employment distribution in each U.S. city as a “text” and each occupation as a “word,” we employ a generalized word shift graph to compare the differences in employment between the 15 cities with the most diverse employment distributions and the 15 cities with the least diverse employment distributions, as measured by the Shannon entropy. We use the average employment diversity across cities’ workforces as the reference value (\(\Phi ^{(\text{ref})}\)). Occupations are relatively “surprising” (+) or “unsurprising” (−) depending on their surprisal in each class of city