Appendix 1: Demographic Data and Analysis
As discussed in section B, in order to disentangle the effect of population dynamics on the participation rate, the chapter adopted a two-pronged strategy. First, we considered a ‘demographic’ approach that relies on disaggregated population and participation data by age group (10 groups) and gender to estimate the demographic component of the decline in participation rates. And second, to investigate the behavior of specific age groups, we considered a shift-share analysis. This Annex describes these methodologies and the data used in detail, compares our results to similar studies, and discusses additional simulations on population and immigration growth based on the US Census forecasts.
We used data on labor force by gender and age groups (16-19, 20-24, 25-34, 35-44, 45-54, 55-59, 60-64, 65-69, 70-74, 75+) from the Household Employment Survey of the Bureau of Labor Statistics (BLS), for the period 1981 to present. Population data, including forecasts of population for 2014-2019, were obtained from the BLS, while the data on immigration used in the simulations described in section II of this Annex are from the US Census Bureau.
Appendix 2: State-Level Regression Model
Prepared by Ravi Balakrishnan, Mai Dao, Juan Solé, Jeremy Zook (WHD). The authors are grateful to Robert Arnold, Roberto Cardarelli, Nigel Chalk, Bruce Fallick, Andy Levin, Robert Shackleton and Mitra Toosi for helpful discussions and comments.
The decomposition uses data on population and labor force from the Household Employment Survey (cf. Annex I for more details).
The endogeneity is much more evident in the difference between OLS and 2SLS using household employment (see Table A2 in Annex II). This is not surprising, household employment, comes from the household survey and encompasses self-employment, which is more responsive to labor supply variation than payroll employment.
Due to data availability by age groups, this section relies on the ‘unemployment rate’ model instead of the ‘employment gap’ model discussed above. We still instrument to avoid endogeneity.
These authors show that during recessions, the unemployment pool is composed relatively more of workers of higher skill and wages compared to normal times (as a big shock hits workers of all ranks). As these workers also have stronger labor market attachment, the average rate of transitioning into non-participation declines during recessions.
Indeed, reverting to pre-Great Recession average levels of school enrollment and employment rates for students would increase the youth participation rate by around 7pp from the current level of 54¾ percent.
Even those denied benefits can often spend one to three years out of the labor force until the appeals process is exhausted
The census also produces three alternative population forecasts based on different migration assumptions. As we show in Annex (II), this makes little difference to the path of the aggregate LFPR, but can make a substantial difference to the path of labor force growth.
The adjustment suggested by Citibank (2014) is followed. Specifically, the part time adjustment is the product of: (i) the change in part time workers due to slack work or business conditions relative to the average for 1997-2007; and (ii) (1-the ratio of average part time hours/average full time hours). This adjustment is added to the unemployment rate (i.e. weighted by the trend LFPR).
The Census Bureau produces these three scenarios as immigration is very difficult to forecast. The different Census scenarios maintain the same methodologies and assumptions on fertility and mortality, and differ only in the levels of net international migration assumed under each scenario.
Bill S.744, Border Security, Economic Opportunity, and Immigration Modernization Act