Today’s Labour Force Survey release is bad news: employment fell by 54,000 between September and October. But look at last month’s release: an increase of 61,000 between August and September. Both numbers are large: the typical monthly change in employment is about 20,000 jobs.
It seems likely that at least one of these numbers is a statistical glitch, possibly due to changing seasonal employment patterns in sectors such as education.
Statistics Canada’s estimate for the standard error of its LFS employment numbers is 27,200, so there’s roughly a one in twenty chance that a given employment number could be over- or under-estimated estimated by up to 54,000 jobs in a given month. The odds of seeing two consecutive errors of that magnitude are very small, so which month is the one-in-twenty outlier?
Another important LFS series is hours worked -- a series that is even more strongly correlated with GDP growth than employment. Hours worked fell even more sharply than did employment during the recession; instead of layoffs, many employers simply cut back on hours.
Hours worked jumped sharply in July and peaked in August, which is why we’re expecting a reasonably good GDP number for the third quarter. But since then, hours worked has fallen slightly. It’s hard to reconcile September’s strong jump in employment with falling hours worked.
My reading of these numbers suggests that much of September’s increase of 61,000 was statistical noise and that October’s number set the record straight. The two-month increase of 15,900 between August and October -- an average increase of less than 8,000 per month -- seems more consistent with the other data we’re seeing.
Follow Economy Lab on twitter
Stephen Gordon's recent posts and Twitterfeed can be viewed here.