UI Reemployment Pilot Analysis
To measure its Government Performance and Results Act (GPRA) and budget goal of "Facilitating Reemployment," the Employment and Training Administration (ETA) has developed a reemployment measure. It is the percentage of claimants who received an Unemployment Insurance (UI) first payment in a given calendar quarter who also had wages reported in state UI wage records in the subsequent quarter. Six states pilot-tested this measure in the summer of 2003. Each state identified claimants receiving UI first payments in each of the four quarters of calendar year (CY) 2002, and searched its wage records for "hits" on these claimants in each of the two quarters following the quarter of first payment. Two states were also able to provide crossmatch results for claimants receiving first payments in the four quarters of CY 2001.
Based on the results of the pilot test, ETA established a baseline for the measure of 51.5%. The 51.5% is the weighted average of the pilot test results, less the percentage of "hits" that appeared to reflect not reemployment but partial employment during benefit weeks.
Using standard regression methods, the pilot data on reemployment rates were analyzed for both first and second quarters after first payments were received. Following are the questions and summary findings of the analysis.
To what extent do both first and second quarter subsequent reemployment rates respond to changes in the total unemployment rate (TUR)?
To what extent do first quarter reemployment rates predict rates in the subsequent quarter?
Do the pilot test results suggest that the exhaustion rate accurately reflects reemployment rates?
Exhaustion rates are not a good proxy for Q+1 reemployment rates. When we included Q+1 exhaustion rates as an explanatory variable in the equation that models movements in the Q+1 reemployment rates, they were not statistically significant. Our primary interest is in the Q+1 period, because that is when we expect UI system and State Workforce Agency interventions to occur and to have their primary impact on reemployment of UI claimants, most of whom are by definition job-ready.
Exhaustion rates are a good proxy for Q+2 reemployment rates. In regression equations designed to explain pilot state differences in Q+2 reemployment rates, Q+2 exhaustion rates actually have a statistically stronger explanatory effect than Q+1 reemployment rates.
Adjusting the reemployment rate to remove "hits" that reflect partial unemployment raises the ratio to about 90%.
Findings: First quarter reemployment rates (denoted Q+1) are highly responsive to the TUR: a 1-percentage point drop in the TUR raises the Q+1 reemployment rate by 6.7 percentage points.
Reemployment rates in the second quarter following receipt of a first payment (Q+2) are less responsive to the TUR: a 1-percentage point drop in the TUR is associated with a 4 percentage-point rise in the Q+2 reemployment rate.
Findings: The simple correlation between Q+2 and Q+1 reemployment rates is high: 85% (that means 72% of the levels of Q+2 reemployment rates are "explained" by where they were the quarter before.)
Findings: Before we developed a reemployment rate to measure facilitation of reemployment, we used the UI benefit exhaustion rate as a readily available proxy. Because most claimants leave benefit status to take a job, we assumed that the complement of the exhaustion rate (one minus the exhaustion rate) would reasonably represent reemployment. Having actual data on reemployment rates for Q+1 and Q+2 gave us the ability finally to test the validity of that assumption. We computed the relevant Q+1 and Q+2 benefit exhaustion rates for the pilot states and examined their relationship to reemployment rates.
On the basis of the regression results, we computed the complement of the Q+2 exhaustion rates for the pilot states and compared them with the corresponding Q+2 reemployment rates. They follow the same pattern as reemployment but are lower, averaging 83% of the crude (unadjusted) Q+2 reemployment rate.
Extent that 1st quarter reemployment rates reflect changes in the TUR
This was determined by regressing the quarterly reemployment rates for Q+1 on the seasonally unadjusted TUR for the state. Each state except one (the baseline state in the regression) was identified in the regression by its own "dummy" variable. The regression assumed that the effect of differences in TUR on reemployment was the same in every state. This regression "explained" about 72% of the variation in reemployment rates. The dummy variables for all but one state were statistically different from zero at the 95% level or higher, and all were negative, indicating their reemployment rates, when adjusted for the effect of differences in the TUR, were lower than the baseline state’s. The coefficient for the TUR was also statistically significant.
The regression coefficient for the TUR was –0.067, indicating that a one percentage point decline in the TUR would raise the aggregate reemployment rate by 6.7 percentage points. Based on the CY 2002 reemployment rate for the pilot of 51.5%, which coincided with an average national TUR of 5.8%, and assuming that the aggregate pilot reemployment is fairly representative of the national aggregate, the regression suggests the following values through Fiscal Year (FY) 2008 based on the Administration’s economic assumptions:
TUR Reemployment Rate FY2003 6.0% 50.2% FY2004 5.7% 52.2% FY2005 5.4% 54.2% FY2006 5.2% 55.5% FY2007 5.1% 56.2% FY2008 5.1% 56.2%
Extent that 2nd quarter reemployment rates reflect changes in the TUR
A similar regression for reemployment rates for Q+2 was also run. It accounted for 78% of the variation, and again the relationship to the TUR was statistically significant, although the coefficient was only about two thirds as large (a one percentage point decline in TUR was associated with a 4 percentage point rise in reemployment.)
Relationship Between Exhaustion Rates and Reemployment Rates.
Exhaustion rates were inserted as an explanatory variable in the regressions explaining variations in state reemployment rates. The effect on the regression for Q+2 was dramatically different from that for Q+1:
In the Q+1 regression, adding the exhaustion rate as an explanatory variable raised the amount of variation explained by only 2 percentage points, from 72% to 74%. The exhaustion rate coefficient was very small (0.0016), not statistically different from zero, and unexpectedly positive.
In the Q+2 regression, adding the exhaustion rate made a dramatic difference. It raised the amount of variation explained from 78% to 93%. Its regression coefficient was -0.0047 and it was clearly statistically significant. This implies that a regression equation containing TUR and exhaustion rates could be used to predict reemployment rates two quarters after the first payment quarter, but not for the first quarter in which we are interested.
The difference in results for the two regression models is very understandable, for two reasons. First, based on the typical maximum benefit duration of 26 weeks, UI has always calculated the exhaustion rate as final payments for the period of interest divided by first payments six months earlier. There is thus a built-in six (6) month lag between first and final payments; this puts the exhaustion rate into the second quarter following first payment. Thus, the exhaustion rate for Q+1 uses final payments for quarter Q+1 but first payments for Q-1, putting it out of synch with the reemployment rate measure for Q+1, which is based on first payments for quarter Q.
Secondly, according to data in the ETA Benefit Rights and Experience report (ETA 218), the average benefit exhaustion occurs after a claimant has collected about 22 weeks of benefits. Only beneficiaries with very short potential benefit durations are likely to exhaust benefits in the first quarter after getting a first payment. (It is true, however, that some beneficiaries with longer durations could exhaust in the first quarter, if they received their first payment early in the prior quarter.)
Any relationship that exists between benefit exhaustion rates and reemployment rates is most likely to be observed, and measured, in the range of 20 to 26 weeks, or in the second quarter after a claimant has received a first payment. Therefore, if we want to use the exhaustion rate (or, more properly, one minus the exhaustion rate, as the percentage of claimants who did not exhaust benefits) the measure seems more naturally to relate to reemployment two quarters after a first payment was received.
Exhaustion Rates as a Proxy for Reemployment Rates
Because we have both exhaustion rate and reemployment rate data for the same time periods, the pilot allows us to do a crude calculation of how accurately exhaustion rates (or, more precisely, their complement) might represent the reemployment of claimants.
The Relationship between 2nd Quarter Reemployment Rates and the Complement of 2nd Quarter Exhaustion Rates for Pilot States, CY 2001 and 2002 State Year 2nd Quarter
One minus 2nd Quarter
Exhaustion Rate as
% of Reemployment Rate
State A 2002 60.2% 49.5% 82.2% State B 2001 78.1% 64.2% 81.9% State B 2002 73.9% 60.1% 81.3% State C 2002* 67.4% 52.5% 77.9% State D 2002 54.6% 40.9% 74.9% State E 2001 70.5% 60.6% 86.0% State E 2002 66.5% 55.8% 83.9% State F 2002 63.9% 60.1% 94.1%
The exhaustion rates average only about 83% of the unadjusted reemployment rates. The relationship is strongest in State F (94%) and weakest in State D (75%). Even after adjustment of reemployment rates to remove partial employment, it appears that in general exhaustion rates for the second quarter probably understate reemployment rates by at least 10 percent.
Although it is a good proxy for reemployment in period Q+2, the complement of the exhaustion rates in some instances understates and in other instances overstates reemployment. First, some claimants exhaust benefits who have actually returned to work previously; they return to work but neglect to inform the UI agency, and claim benefits until they exhaust benefits. The larger the incidence of these "Benefit Year Earnings" overpayments, the more the complement of the exhaustion rate understates the true reemployment rate. Second, other claimants cease claiming benefits before exhausting their entitlement for reasons other than taking a job, for example, sickness, disablement, or withdrawal from the labor force for other reasons. In these instances, the complement of the exhaustion rate overstates return to work.
Relationship Between First Quarter and Second Quarter Reemployment Rates.
A regression analysis was performed between the first and second quarter reemployment rates. Reemployment rates in the pilot states during the first quarter following quarter of first payment "explained" 72% of the variation in second quarter (the simple correlation was 85%). Including the TUR in the state during the second quarter raised the amount of variation explained to 76%. When the Q+2 exhaustion rate was substituted in the same regression for the Q+1 reemployment rate, the regression accounts for 93% of the variation. Thus, when it comes to explaining Q+2 reemployment rates, the exhaustion rate is statistically more powerful.