Goldman economist Alec Phillips explains why just as the census has been artificially adding NFP jobs to this point, beginning in June its impact will be one of actually subtracting jobs from the NFP. This means that in the president's next NFP endorsing press release, he will, for once, focus on the pro forma impact of the census, and as NFP is now expected to swing wildly lower in June, presumably as a function of the Census program peaking in May. No matter what, at this point nobody is willing to stick their neck out and discuss the lack of improvement in private sector hiring.
From Alec Phillips
- Census workers added 411,000 to payrolls in May, but will begin to subtract from payrolls for the next several months, as operations wind down. We estimate that the decline in census workers this month could shave as much as 275,000 off of the June headline payroll figure, with additional subtractions of between 75,000 and 100,000 per month in July through September.
- Although the data are inconclusive, it is likely that at least half of the temporary census workers already had other jobs, and thus had a less significant effect on the unemployment rate than would otherwise have been the case. While it appears that a good deal of the rest may have come out of the ranks of the unemployed, census hiring has probably been only a secondary factor in the decline in continued unemployment claims over the last several months.
The number of temporary census workers surged to 564,000 in May, adding 411,000 to last month’s otherwise disappointing payroll report. But the number of census workers has already begun to decline from the early May peak, potentially subtracting as many as 275,000 jobs from the payroll report.
Most forecasters, including ourselves, consistently overestimated the effect that the census would have in the first few months of the year. As it turns out, this was due to two factors. First, a quirk in the calendar meant that the payroll survey week (i.e., the week that includes the 12th of the month) tended to fall a several days later in those first few months this year than it did ten years ago. With workers being added each week, this meant a lower reported number each month than what the 2000 trend would have implied, which we and most others used as a baseline for this year’s effects. Second, despite survey week irregularities, the ramp to the peak in this year’s census has been steeper than in 2000, presumably as computerization has led to somewhat lighter preparatory efforts despite a greater number of census takers at the peak. This led to smaller than expected payroll gains in the first few months of the year, but a larger gain in May.
Now that census payrolls have peaked—this appears to have occurred in the first week in May—the census will begin to weigh on the monthly jobs report. We expect the level of census workers in June to dip just under 300,000, which would put it slightly below the 2000 level. This would result in a decline of around 275,000 in census payrolls, as noted above.
Additional declines are likely in July through September, when several programs wind down. The most significant is the “non-response follow up” program, for which the vast majority of hiring was done. This ends officially on July 10, but most of the temporary census-takers that have been hired will have already completed work well before that date. Smaller programs end later in July and in August. We expect additional declines of between 75,000 and 100,000 through September as those final verification efforts wind down. The census effect should fade completely by December.
Regarding the unemployment rate, the question is whether the census workers hired over the last few months were previously unemployed, not in the labor force, or already had a job other than with the census. It appears that a good number of them may fall into the latter category. The year over year change in multiple jobholders in May increased from -676,000 to -4,000, a change of 672,000, compared with a year over year change in census workers of 495,000 (see chart below). These are noisy data and the timing of the monthly change over the last few months (not shown) raises doubt as to how many of these were census jobs, but our standing assumption that about half of the census workers come from someplace other than the unemployed still seems reasonable. The Labor Force Status Flows report, which estimates the flow of workers among unemployed, employed, and not in the labor force categories, doesn’t provide much additional information, though it is worth noting that the number of workers moving from “not in the labor force” directly to employment has declined every month since January on a seasonally adjusted basis, so this seems a less likely source of census workers than the unemployed (the change in the flow of unemployed to employed has, if nothing else, at least been directionally consistent with increased census hiring over the last couple of months).
A second question is whether the census is at least partly responsible for the recent decline in total continued claims (including state and federal emergency/extended benefits), which have dropped from a peak of 10.55 million in January to 10.01million in mid-may. For the most part, this decline seems unrelated to census hiring. First, the timing of census hiring and the drop in continued claims doesn’t match up, although peak census hiring occurred around the same time that unemployment benefits temporarily lapsed, so the data from that period are even noisier than usual. Second, many census workers may still be filing claims, since those who earn less than their weekly unemployment benefit are generally able to collect the difference, and those who earn more may still file claims to maintain eligibility for the future, even if they aren't receiving payments now. If we assume that half of the census workers were previously unemployed (this may be a slightly generous assumption, since some of them probably also came from outside the labor force), and of those unemployed, half stopped filing claims, this would mean that around 140,000 of the decline in continued claims is due to census hiring, a fairly small share of the 540,000 million drop in continued and extended/emergency benefit claims since census hiring began several months ago.