According to Barack Obama, 600 thousand Americans have lost their jobs since January. Actually, he’s wrong: something like 20 million Americans have lost their jobs since January. It’s just that most of them found new jobs. Probably the new jobs generally weren’t as good as the ones they lost. And almost certainly, more than 600 thousand of them were unable to find new jobs, because many of the new jobs created were filled by new entrants to the labor force or by people who were already unemployed when the year began.

Like almost everyone else I’ve ever heard, Senator Obama is making the mistake of using a net job loss figure with language that, if taken in its plain sense, clearly implies he is talking about gross job loss. And it seems to me that gross job loss is the appropriate concept: losing your job is a pretty serious bummer, even if you are able to find a new one after a few months.

knzn is right to highlight the distinction between gross and net employment outflows. Obama probably was talking about gross job losses. But the reason presidents talk about “the economy” “creating” so many jobs — never mind that an economy can’t create anything — is because that’s the proper way to interpret the statistic.

It is disingenuous to claim that 20 million workers lost jobs in 2008 without also talking about the 19.4 million persons who found new jobs. It would be equally disingenuous for Bush to claim he “created” 19.4 million jobs in 2008. That brings the discussion back to a net flow of 600 thousand.

Composition effects are large and important, as are the welfare losses from unemployment, but it was the 20 million number that struck me — as too small.

I discussed earlier that about 5 percent of the population (10 million persons) moves into and out of employment every month. Assuming outflows account for roughly half of that, we get 5 million persons a month. Over 8 months that comes to 40 million persons, twice knzn’s estimate.

To be fair, knzn’s figure is a rough estimate based on last year’s data and we all know the risk of projecting current trends into the future — just ask LTCM, Bear Stearns, Lehman Brothers, etc. And I suspect that I’m taking knzn’s forecast more seriously than he does but since I’ve already hauled out the sledgehammer, let’s find that fly.

knzn arrives at his estimate of 20 million lost jobs by extrapolating the “gross job losses” series from the BLS’s Business Employment Dynamics (BED). (This is a different source than the one the 600,000 figure comes from, but that doesn’t matter.) The BED reports quarterly data and with a considerable lag — no 2008 data are available yet — so knzn extends the quarterly rate from 2007 (about 7.4m per quarter) by assuming losses of 8m in the first 3 quarters of 2008 and rounding down.

There are better data available. Using Current Population Survey (CPS) data, I calculated the total number of persons who left employment each month through August 2008 (the latest available). These data, called gross flows, show that over 46.9 million Americans lost jobs in the first eight months of 2008! On the bright side, during that period over 46.4 million Americans were hired. The net change is -511,000, close to the -605,000 figure we get from cumulating the change in nonfarm payrolls from the establishment survey.

The first 3 columns of the table below show the monthly gross flow of persons into and out of employment as well as net change in employment (inflow — outflow). The right-most column shows the change in non-farm payrolls calculated from the CES.

So why does knzn’s estimate gross job losses differ so much from mine? The answer is time aggregation. knzn uses the BED, which has a quarterly frequency. I use CPS data, which has a monthly frequency. And the frequency of observation matters when measuring gross flows.

My research (link to come) shows that increasing the frequency of observation from monthly to weekly yields a 20-percent increase in measured transitions. Although I can’t extrapolate the weekly-to-monthly number into a monthly-to-quarterly effect, it is clear that the discrepancy would only intensify.

Indeed, it must. Below is a graph of quarterly employment outflows from 1990-2008. The quarterly CPS series is the sum of the monthly gross flows for that quarter. The BED understates the quarterly CPS job losses by more than a factor of 2.

Source: CPS and BED data.

In addition, the CPS data understate the true magnitude of gross job flows because they do not include direct job-to-job transitions. The CPS only records when a person moves from employment to nonemployment (unemployment or out of the labor force). Many workers transition directly from one job to another, without an intervening spell of unemployment. Fallick and Fleischman (2004) estimate that 2.6 percent of workers change employers each month (3.8m in 2008), more than twice as many as from employment to unemployment.

Whether the job-to-job changers should count as “job losers” is another question, but if Obama wants to talk about the total number of American workers who lost jobs in 2008, that number is at least 47 million.

We use the term gross flows to distinguish such measures from net flows, which are the change in the stock from one month to the next. It is the net change that gets reported in government statistics. For example, if 250,000 persons found a job this month and 100,000 lost a job, the net change is +150,000 persons. The total gross flow, however, is 350,000 persons.

Gross flows are considerably larger than net flows. Over the last 30 years the U.S. economy has added about 150,000 jobs a month (net) on average. During that same period the average gross flow into and out of employment is over 10 million persons a month! To put that figure in perspective, 5.4 percent of the U.S. working-age population moves into and out of employment every month. Accordingly, the net changes reported in the monthly Employment Situation release do not capture the true dynamism in the U.S. labor market.

Although not directly germane to my research, I recently became interested in the long term trend in gross worker flows. The figure below plots the trend in worker flows for 1976-2007. The series shown is the sum of all flows into and out of employment, expressed as a share of population. Since I am interested in only the trend in the series, I remove the substantial month-to-month variation by seasonally adjusting the data and then smoothing the seasonally-adjusted data using a local weighted least squares regression. Shaded bars indicate recessions as determined by the NBER.

There has been a dramatic decrease in gross flows over the past 30 years. This decrease is consistent with evidence from Bleakley, Ferris, and Fuhrer (1999) and Fallick and Fleischman (2004), who also observe declines in gross flows.

In the late 1970s, flows averaged about 5.8 percent of the population a month. This fell more or less steadily throughout the 1980s and early 1990s until bottoming at 5 percent at the end of 1996. Flows grew to just over 5.3 percent a month in 2001 after which they fell off a cliff, stabilizing briefly in 2004-05. By the end of 2007, the total gross flow was below 5 percent of the population. To put a number to this trend, the decline in gross flows means that about 1.7 million fewer persons a month move into and out of employment today compared to 1977.

Identifying an empirical regularity is only a first step in research. Without understanding the causes of this trend it is impossible to draw any meaningful implications from the data. In fact, it is difficult to say whether the secular decline in total gross flows into and out of employment is “good” or “bad”. On the one hand, if the decline represents decreased employment volatility for U.S. workers — increased job stability — then it may be a positive development. On the other hand, if the observed decline in gross flows results from decreasing allocative efficiency — labor market “sclerosis” — then the trend may be worrisome. Other possible explanations include “job lock,” where employees cannot easily change jobs because of employer-provided health insurance, and increased efficiency in employer-employee job matching, resulting in fewer low-quality matches and thus less job turnover.

The examples above show that both explanations and implications can be conflicting and contradictory. This is why economists write models. Also why Truman wanted a one-armed economist.