Demonstration of BLS Separation Rate Methodology Change C. Brett Lockard Employment Projections Program June 9, 2015
Overview Measuring Historical Separations Labor Force Exits Occupational Transfers Projecting Separations Limitations of Separations Method 2
Measuring Historical Separations 3
Measuring Historical Separations Ideal data source would be longitudinal Existing longitudinal surveys (NLS, PSID) are too small to provide reliable data for detailed occupations Separations method uses CPS data (as does replacements method) 4
Measuring Historical Separations Uses longitudinal aspects of CPS monthly data and supplements Directly measures workers who separate from an occupation Measures labor force separations and occupational transfers separately 5
Labor Force Exits 6
Calculating Historical Labor Force Exits Use longitudinal characteristics of the CPS to look at historical labor force exits CPS measures individual household members for 4 months, then for 4 months again after an 8 month break CPS respondents are matched based on household ID and demographics 7
Labor Force Exits Excludes records where individuals moved in and out of the labor force within either 4 month period Intended to remove short-term exits and workers marginally attached to labor force Also removes noise from the data Measures people in the labor force in months 1-4, but not in months 5-8 separated from the labor force for at least 4 months 8
Occupational Transfers 9
Calculating Historical Occupational Transfers CPS Annual Social and Economic Supplement (ASEC) asks respondent about labor force activity in the previous calendar year Conducted every March 10
Why ASEC Supplement? Other possible sources less suited to this analysis If Matched monthly CPS occupation data Values a year apart are independently coded Job Tenure/Occupational Mobility supplement More seasonally affected Only conducted biennially 11
Occupational Transfers ASEC respondents are asked whether their longest job in the prior year was the same as their current job If not, the occupation of their longest job from the prior year is coded Prior year occupation compared with current occupation to measure transfers 12
Occupational Transfers If respondents were employed in a different SOC major group in the prior year, they are considered to have transferred from their prior occupation Major group transfers used because: Major group coding is more reliable Eliminates transfers between comparable occupations, like retail salespersons and parts salespersons 13
Projecting Separations 14
Regression Approach Rather than extrapolate from historical data, a regression model is used to project separations Accounts for demographic factors that affect separations Age, sex, educational attainment Better handles small occupations and changing occupational classifications 15
Labor Force Exits Model Binary dependent variable: does an individual leave the labor force? P(Exit = 1) = Φ( β 0 + β 1 Sex + β 2 Age + β 3 Sex Age + β 4 Occupation + β 5 Education) 16
Projecting Labor Force Exits Regression model run on multiple years of historical data (currently 2004-12) generates coefficients for each age, sex, occupational group, and education level If Coefficients applied to base year (2012) CPS demographic data 17
Projecting Labor Force Exits Resulting probability for each record is multiplied by observation weight to get a count of exits and non-exits For example, if an observation has a 5% probability of separation, and a weight of 100, then 5 workers are recorded as exiting and 95 workers as staying Results are aggregated by occupation, and then scaled to a 1 year rate 18
Occupational Transfers Model Binary dependent variable: does an individual leave their current occupation and find employment in a new occupation? P Transfer = 1 = Φ( β 0 + β 1 Sex + β 2 Age + β 3 Occupation + β 4 Education + β 5 Unemployment rate) 19
Projecting Occupational Transfers Regression model run using multiple years of supplement data (currently 2003-12) Implementation is same as labor force model: Coefficients applied to base year (2012) for each detailed occupation Results scaled to one-year rate 20
Separation rate Z-Score Raw Data vs. Regression Results If 6% 5% 4% 3% 2% 1% 0% LTHS HS SCND AD BA MA Doc Education category 0.2 0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 Separation rate Z Score 21
Example Results Age Sex Occupation Education P(Exit=1) P(Transfer =1) Registered Associate s 32 Female Nurse Degree 1.7% 2.1% 45 Male Registered Nurse Bachelor s Degree 0.4% 1.3% 46 Female Registered Nurse Master s Degree 0.7% 1.4% Registered Nurses 1.7% 1.6% 22
Combining the Measures Applying the regression coefficients to current workforce data yields rates for labor force exits and occupational transfers One-year rates are assumed to be annual averages over the projections period Rates are applied to the average of base and projected year employment to generate annual average separations 23
Limitations of Separations Method 24
Concept Limitations Demand-side measure only Does not provide any information on supply Does not differentiate between opportunities for different types of entrants 25
Method Limitations Applies CPS occupational separation rates to OES-based matrix employment data Assumes historical patterns of separations for given characteristics Assumes no change in occupation demographics over projection period Applies in-sample patterns to all workers 26
Contact Information C. Brett Lockard Economist Employment Projections Program www.bls.gov/emp 202-691-5730 lockard.brett@bls.gov