Previously, Emsi occupation earnings data was estimated using only the latest year of Occupational Employment Statistics (OES) data available. The BLS cautions strongly against treating OES as a time series; therefore Emsi did not allow prior years of OES to influence the latest year of OES.
This methodology was necessary in the past because Emsi had not built out a methodologically reliable way of using OES as a time series. The result of using only one OES year at a time was a large amount of volatility in earnings as each new OES year came in without the benefit of the context provided by prior years. This frequently led to users asking about dramatic differences in earnings between dataruns using different years of OES.
Emsi has finished building a time series of OES, which means that all years of earnings (including current-year) now have the benefit of being influenced by earnings data from other years in the dataset. The result is a much more stable, accurate earnings set. Fluctuations year-over-year will decrease, and users will see less of the wage and employment volatility that used to characterize yearly OES updates. Users will still see volatility between the 2019.2 and 2019.3 dataruns because of the improved methodology and the introduction of 2018 OES. Additionally, users may still notice volatility in employment and earnings for small geographies where employment and earnings are heavily suppressed. Emsi will be conducting further research into improvements for unsuppressing and estimating data for these regions.
Currently Emsi tools (with the exception of our core LMI API) only display latest-year OES earnings data. Our newly-developed historical occupation earnings data will be integrated into our tools in the near future. Current-year earnings will continue to be displayed as they always have, but they’re now the result of our OES time series rather than a single year of OES data.
Further documentation describing Emsi’s new OES time series is available here.