TECHNICAL APPENDIX After Everything - Projections of Jobs, Education, and Training Requirements through 2031

• In-sample forecasting: We estimate the results of our model using only a portion of the sample, and then using the result to predict outcomes for the remainder of the sample in order to test the extent to which the model accurately predicts known events. [...] We have chosen to use the CPS instead of the much larger American Community Survey (ACS) due to the longer history of the CPS data set.3 Our time-series methodological framework benefits from our use of the longest-term data set possible with information pertaining to the educational and occupational characteristics of the population.4 The relatively lengthy period of observations in the CPS data. [...] The latter are derived from the Current Population Survey (CPS) and reflect the occupational distribution of the CPS and the American Community Survey (ACS). [...] Our approach allows us to estimate the progression in educational demand for every year of the 10-year forecast instead of only at the beginning and end of the forecast horizon. [...] The RMSE is a good measure of the extent to which the model predicts the response, where lower values of RMSE indicate better fit in a nested context.18 We find that the model fit improves with 17 The changing mix of workers within each industry over time will also influence the educational requirements of the occupations within that industry.