Phantom jobs and job losses

Public Interest, Wntr, 2005 by Tim Kane

Aggregate data from the Census Bureau show that job-changing from one employer to another, which had averaged 3 percent per month in the 1990s, declined by about 0.2 percentage points per year after 2001, settling at 2.4 percent in 2003, where it remains today. This seemingly small change meant that during the campaign season, roughly one million fewer workers were changing jobs each month. The result was that roughly one million fewer workers were being double counted on payrolls, a statistical change that the payroll survey registered as one million "lost" jobs.

Given the obvious flaw in the payroll methodology, the only curious thing is that this effect was not quantified this neatly before 2004. The BLS had noticed the divergence between the payroll and household surveys, and had even started publishing a "reconciliation" each month in an effort to explain the gap. But when it began issuing this report, the bureau initially neglected to mention job-changing as one of the possible causes. Only in August 2004, after pressure from outside economists, did the BLS issue a paper discussing the issue. This despite ample anecdotal evidence that the economy was experiencing decreased turnover.

Not only that, but the BLS itself produced a new data set that, in hindsight, clearly points to a sharp drop-off in turnover after September 11. In 2003, the bureau launched a new quarterly report intended to go beyond the static snapshot methodology of the payroll and household surveys and look at the internal dynamics of employment. The report, called "Business Employment Dynamics" (BED), measures the gross flows of new jobs created and new job losses every quarter, and has been calculated as far back as 1992. This study provides a stark image of the slowdown in turnover after 2001.

A measure of gross job flows can be instructive because it provides a clue to the degree to which the demand side in the labor market--the employer--is providing opportunities for the "churn" that causes turnover and the resulting double counting on the payroll survey. BED demonstrates that 8.2 percent of all jobs were created anew each quarter in the 1990s, while roughly 7.5 percent were destroyed. In other words, in any given month 7.5 percent of American jobs would be lost, while a number equal to 8 percent would be created. So the net job creation (jobs created minus jobs destroyed) might have appeared modest, but it was masking a much greater amount of movement in the job market--movement that also had the potential to exacerbate the payroll survey's double-counting problem. The rate of job gain started to drop in 2000, then rebounded in the second half of 2001 before dropping again throughout 2002 and 2003, when it settled at 7 percent. Yet while the rate of job losses spiked during the 2001 recession, reaching 8.4 percent in the third quarter of that year, it dropped after the September 11 attacks. The job loss rate remained lower in the second half of 2003 than at any other time in the previous decade. In other words, the labor market is in a period of relative stability, with a much smaller amount of churn than was seen in the 1990s, and thus much less potential for payroll double counting. This provides a demand-side explanation for the divergence between the payroll and household surveys.