Over the weekend, a critical media report, looking at the apparent contradiction between the July 2013 Labour Force Survey (LFS) and Survey of Employment Payroll and Hours (SEPH) releases, was brought to our attention. As noted (repeatedly) on these pages, analysing the month-to-month changes reported in Statistics Canada employment survey releases is a fool’s errand. The report deserves an ‘A’ for effort, for at least casting a critical eye on the figures. It gets an ‘incomplete’ otherwise.
LFS, SEPH methodology and sampling
We’ve discussed the LFS methodology briefly before, but perhaps it bears repeating. The LFS sample size means every respondent represents on average 200 Canadians age 15+. Respondents are part of a quasi-cohort, screened into the survey for six consecutive months at a time. Respondents’ profiles are only updated when they report a change in employment status (given the length of time it takes to complete the survey, there’s actually a disincentive to do so). Finally, Statscan loses a significant number of LFS respondents over the course of a six month cycle. Effectively, if there’s any real change in the Canadian labour market beginning in a given month, its full impact won’t show up in the LFS until six months out when the sample is completely refreshed.(The US Bureau of Labour Statistics uses quasi-cohorts screened and cycled in on a staggered schedule. However, the Current Population Survey, on which the US monthly labour reports are based, has a relatively much smaller sample size than the LFS.)
Also, the LFS is a household survey. Some of the information the LFS attempts to collect from household respondents would be more reliably collected from businesses. This would include industry classification, actual number of employees on business payrolls and the hours these employees worked. Enter the Survey of Employment Payroll and Hours. On the face of it, using Canada Revenue Agency (CRA) Business Number Registry (BNR) seem to be a fairly simple and straight-forward way of collecting accurate information for the labour market variables just mentioned. Here as well Statscan uses a sample survey, the Business Payroll Survey (BPS). The BPS sample size means every responding business represents on average 60 businesses. Most responding businesses are also part of a quasi-cohort, screened into the survey for twelve consecutive months at a time. It’s unclear how many responding businesses Statscan loses over the course of a twelve month cycle. Finally, for reasons left unexplained, Statscan does not use the BNR for industry classification, instead relying on its own Business Registry (BR) for the industry classification. Apparently, the BR has been remarkably unreliable in collecting industry classification info over the last 5 years, but more on that in a moment.
One last point worth mentioning is that the data reported in both the LFS and SEPH monthly releases is seasonally adjusted.
July 2013 SEPH and Statscan revisions
All that said, what exactly happened in July 2013? Given the preceding, there are a lot of variables that could account for a major discrepancy between the two surveys in any given month. In this instance, it looks like seasonality played a significant role in the Construction employment jump noted in the July 2013 SEPH report.
Seasonally adjusted, Construction employment jumped by an estimated 26,869 jobs between June and July 2013, nearly half the total estimated month-to-month increase in employment. However, between May and July 2013, seasonally adjusted Construction employment increased by only 6,176 jobs. The seasonally adjusted figures showed a significant drop in month-to-month Construction employment, -20,693 jobs, between May and June 2013.
The seasonally unadjusted figures showed a more consistent rise in employment from April through to July 2013, normal given the industry’s seasonality. However, even the unadjusted figures showed a slightly lower than normal bump in Construction employment between May and June 2013, +15,819 jobs. For context, the May-June 2012 bump was +39,502, while the May-June 2011 bump was +53,383.
Seasonal adjustment is based on historical data. The lower than normal bump in the seasonally unadjusted data between May and June 2013 translated to a significant drop in seasonally adjusted Construction employment for June. In contrast, the return to trend in the seasonally unadjusted data for July 2013 resulted in a major bump in seasonally adjusted Construction employment for July.
How much, if any, of the July 2013 SEPH employment boom represents real economic activity and how much owes to survey sampling and methodology issues is hard to tell. As noted in the monthly SEPH releases, Statscan reserves the right to revise the data, and does so frequently. Sometimes these are major revisions (See Note to Readers at the end of the linked post). Which is why reading anything into the monthly movements of these labour surveys is ill-advised.
‘Unclassified businesses’ booming as data withheld
As mentioned in the post linked immediately above, there’s a much more interesting story not being reported: The ballooning number of ‘unclassified’ jobs in the survey. According to the SEPH, ‘Unclassified businesses’ accounted for two in five Canadian jobs created over the past 5 years, and almost all of the jobs created in Canada’s largest province over the same period. Outside of Construction, ‘Unclassified businsesses’ apparently accounted for the greatest bump in July 2013 payroll employment, creating 7,314 jobs that month.
To add to the mystery, apparently Statscan is refusing to release the data to labour groups curious to see what exactly is going on with the boom in ‘Unclassified business’ hiring.