Phantom jobs and job losses

Public Interest, Wntr, 2005 by Tim Kane

AMERICANS are working. But how many of them? The answer to that question can have profound political consequences. Employment was a big issue in the 2004 presidential campaign, as George W. Bush and John Kerry sparred over how many jobs had been created or lost, and how to create more, and better, jobs in the future.

Yet as the contenders traded barbs over outsourcing and tax increases, hardly anyone except for a handful of economists paid much attention to trying to answer the most basic question of all--how many Americans are working right now? It sounds like a simple question, and there are certainly many possible answers floating around in the media, such as payroll statistics or the unemployment rate. But measuring employment and unemployment turns out to be a ticklish business, and only becomes more so once the experts turn their attention from statistical questions of samples and methods and start seeking explanations for the numbers they have counted.

So perhaps it shouldn't be surprising that many of these experts have developed methods and models that are incomplete. There is one statistic in particular that fueled the economic pessimism that marked the presidential campaign--total non-farm employment as measured by the government's payroll survey. But cracks have been appearing in this statistic's facade. The method used to produce the number has not kept pace with the rapidly changing American economy. It is high time to reconsider the faith politicians and policy makers place in current payroll data.

Yesterday's methods in today's labor market

Employment is measured by the government's Bureau of Labor Statistics (BLS), a division of the Department of Labor. The BLS produces two different numbers intended to paint a picture of the overall labor market--the payroll survey and the household survey. The payroll survey yields data on the actual number of jobs gained and lost in the survey period (which is one month for both numbers). The household survey produces an unemployment rate that captures the percentage of the labor force that is out of work, as well as an overall employment estimate.

Conducting two different surveys might seem redundant, but the payroll and household numbers are produced in very different ways. The payroll survey has a larger sample size, and tabulates roughly 400,000 employer records (although usually only about half of the surveyed employers return the questionnaire in time to be tallied in the preliminary release on the first Friday of every month). Certainly, the large sample size--it aspires to capture about one third of the work force--is one of the payroll survey's assets. And yet this vast sample lulls economists into a false sense of confidence about payroll sample quality.

For starters, job growth estimates in the payroll survey have a potential margin of error of 200,000. Over 15 percent of preliminary payroll job growth estimates--the ones that garner the most media attention and are most quickly available to policy makers--are incorrect. This is partly a result of employer foot-dragging, but is significantly tied to a second, even more significant, structural limitation of the survey itself.

The payroll survey only measures traditional jobs, where a worker is hired onto a formal payroll. The survey involves an indirect sample of employment records from the unemployment insurance system. It was a smart way to measure employment in the age of "organization man," when the labor market was more rigid, most people who worked did so full time, and people tended to stay at one job for a long time. But payrolls are not so adept at measuring employment in today's labor force of flexible moms, part-time students, and early retirees. Consultants are not counted, nor are the self-employed, nor many partners in the new breed of limited liability corporation (LLC), nor real estate agents. So, for example, the rise of working parents who are flexibly or marginally attached to the labor force is largely missed by the payroll method of measuring employment.

There is yet a third quirk of the payroll survey that is worth noting: It asks employers how many people have been on the payroll at any time during the survey period, which can lead to double-counting. Consider a worker who quits her job at the end of the second week of January to start in a better position at a different firm. She will be counted on the questionnaire submitted by her former employer, since she was on the payroll for at least part of the survey period. She will also be counted by her new employer, for the same reason. Considering that, on average, up to 3 percent of workers changed jobs every month during the late 1990s, it is easy to see how this methodological flaw can significantly skew perceptions of the size of the job market. In a work force of 130 million people, roughly four million people were being counted on two payrolls per month due to turnover. In general, the payroll survey will overstate employment during times of high job turnover in the labor market, making employment seem smaller by comparison during periods of relative stability.