360 likes | 517 Views
Exploring Differences in Employment between Household and Establishment Data. Katharine G. Abraham, University of Maryland and NBER John Haltiwanger, University of Maryland and NBER Kristin Sandusky, U.S. Census Bureau James Spletzer, U.S. Bureau of Labor Statistics. Overview.
E N D
Exploring Differences in Employment between Household and Establishment Data Katharine G. Abraham, University of Maryland and NBER John Haltiwanger, University of Maryland and NBER Kristin Sandusky, U.S. Census Bureau James Spletzer, U.S. Bureau of Labor Statistics
Overview • Employer survey (CES) employment grew faster than household survey (CPS) employment from 1997 through 2001, but CPS employment outpaced CES employment from 2001 to 2003 • Similar cyclical pattern observed in other periods • Discrepancies have provoked much discussion but remain a puzzle • We use CPS records matched to UI wage records for the same individuals to explore sources of discrepancy between employer-reported and household-reported employment
CPS employment Number of people Includes wage and salary, self-employed, unpaid family workers Worked 1 hour, or with a job but not at work Week of 12th Person totals controlled to Census population estimates CES employment Number of jobs Includes only non-farm wage and salary workers On payroll Payroll period including the 12th Employment counts benchmarked to administrative data CPS versus CES employment
Chart 2. Ratio of establishment survey employment to household survey nonagricultural wage and salary employment, 1948-2004
What explains recent discrepancies in CPS and CES employment trends? • Sampling error in the two surveys • Persons under age 16 and members of the institutionalized population excluded from the CPS employment counts • Possible issues with the treatment of government-subsidized jobs • Incomplete accounting for multiple jobs in the adjusted CPS employment series • Adjusted series ignores jobs beyond 2nd job • Adjusted series ignores secondary civilian jobs held by those in the Armed Forces
What explains recent discrepancies in CPS and CES employment trends? (cont’d) • Benchmarking of the CES estimates • Population controls used for CPS estimates • Classification of CPS jobs as wage-and-salary employment versus self-employment • Missing marginal jobs in CPS • Missing “off-the-books” or non-standard employment in CES • Pro-cyclical turnover that affects number of jobs during longer CES payroll periods relative to single CPS reference week
Measurement framework Number of persons employed in UI X2 + X4 Number of persons employed in CPS X3 + X4 Difference in number of persons employed (UI - CPS) X2 – X3
Measurement framework Number of multiple job holders in UI Y2 + Y4(+ part X2 ) Number of multiple job holders in CPS Y3 + Y4(+ part X3 ) Difference in number of multiple job holders (UI – CPS) Y2 – Y3(+ part X2 – part X3 )
Research strategy • Use UI and CPS data to study levels and changes over time in number of people by employment status (X2, X3) and job count classification (Y2, Y3) • Are aggregate movements consistent with our hypotheses? • Examine characteristics of people and jobs in different cells • Are personal and job characteristics of people in different cells consistent with our hypotheses? • Use information on changes in person and job characteristics over time to simulate movements in X2, X3, Y2, Y3 series • Do simulated series reproduce the discrepancy that motivated our study?
Linking CPS and UI records • Census Longitudinal Employer-Household Dynamics (LEHD) program has UI wage record data for 17 states from 1996 to present • CPS data monthly and UI data quarterly • Need to construct quarterly CPS records for comparison with quarterly UI records for same individuals • Protected Identity Key (PIK) based on SSN available for 70-80 percent of March CPS supplement responses and all UI wage records
Analysis sample • Analysis sample consists of March CPS respondents age 16 and older who live in 16 states covered by LEHD data (17 states minus Maryland) • Maryland dropped because more than 15 percent of residents work in another state or DC • Because quarterly information required for comparisons with UI data, sample limited to those with CPS responses for January, February and March • Because CPS records must be matched to the UI wage records, sample limited to CPS records with a PIK • Propensity score methods used to adjust CPS weights to account for sample restrictions
Constructing quarterly employment records • In both data sets, in-scope employment includes individuals with a non-agricultural private sector, state government or local government wage-and-salary job • Information on job changes and multiple jobs held simultaneously used to categorize people as holding one in-scope job or two plus in-scope jobs in CPS • Most certain a job change has occurred if question asked and answered directly, but not always asked • Multiple job question asked every month, but class of second job asked only in outgoing rotation group • Will discuss results for more restrictive of two criteria • Number of jobs in UI data based on number of wage records
Trends in national CES versus CPS, linked sample UI versus CPS
Discrepancies in job count status, restrictive CPS classification, 1996-2003
Trend in Y2 and Y3, more restrictive CPS multiple job definition
Which people would we expect to find in X2 and X3? • Expect people in X2 to hold jobs they consider marginal • Personal characteristics: young (students), older (retired) • Job characteristics: short duration, low hours, low earnings • Expect people in X3 to hold “off-the-books” or non-standard jobs • Personal characteristics: older, less than high school education, college or higher education • Job characteristics: short duration, low hours, low earnings; types of work (industries and occupations) in which there are many non-wage-and-salary workers (self-employed, contractors, consultants)
Factors affecting probability in-scope UI worker not an in-scope CPS worker (X2)
Factors affecting probability in-scope CPS worker not an in-scope UI worker (X3)
Which people would we expect to find in Y2 and Y3? • Expect people in Y2 to hold marginal second jobs and/or have two jobs counted when they change employer • Personal characteristics: young (high turnover) • Job characteristics: short duration, low hours, low earnings • Expect people in Y3 to hold “off-the-books” or non-standard jobs • Personal characteristics: older, less than high school education, college or higher education • Job characteristics: short duration, low hours, low earnings; types of work (industries and occupations) in which there are many non-wage-and-salary workers (self-employed, contractors, consultants)
Factors affecting probability UI multiple job holder has only a single CPS job (Y2, restrictive)
Factors affecting probability CPS multiple job holder has only a single UI job (Y3, restrictive)
Actual and predicted Y2 and Y3, more restrictive CPS definition
Predicted shares of UI multiple job holders with a single CPS job (Y2)
Predicted shares of CPS multiple job holders with a single UI job (Y3)
Large discrepancies at the micro level between employment and job count status for same individuals in UI and CPS data Characteristics of those in off-diagonal cells generally consistent with expectations No single story about recent divergence between UI (employer reported) and CPS (household reported) employment Growth in number of multiple job holders in UI not measured by CPS important 1996-1999 (marginal 2nd jobs, employee turnover) Growth in number of workers measured in CPS but not UI important 2001-2003 (off-the-books or non-standard jobs) Summary
Comments and suggestions? • This is work in progress and we would appreciate your thoughts about what we’ve done and next steps we should take