Statistics Canada publishes two sets of numbers for employment. The one that makes the headlines is from the Labour Force Survey of households, and it is released on the first or second Friday following the month in question. The other series is the Survey of Employment, Payrolls and Hours of employers, and it is released about 6 or 7 weeks after the LFS numbers are published.
In the US, the household survey and the establishment survey are released at the same time, and my understanding is that US analysts believe that the payroll numbers are more reliable. But I'm unaware of any reason to prefer the SEPH numbers over the LFS, quite possibly because the SEPH only started in 1994.
July 2009 poses a bit of a puzzle for those of us who follow these things (such as Andrew Jackson). According to the August 7 LFS release, employment declined by 0.26%, but the September 30 SEPH numbers showed an increase of 0.51%. What are we supposed to make of that?
For reasons unknown to me, the LFS survey generates numbers that are something like 15% greater than those produced by the SEPH. So here is a plot of both series scaled so that March 1994=100:
Both series track each other well up until the end of 2000, but the SEPH records higher employment growth after that date. (The US numbers show the same pattern).
In the shorter term, the two series show remarkably little coherence. Here is a scatter plot of the monthly growth rates of the two employment series:
The raw correlation of the monthly growth rates is only 0.24 (the corresponding number for the US data in this period is 0.51), which also happens to the the fraction of observations in which the reported growth rates have opposite signs. The July 2009 numbers are somewhat of an outlier, but they are not the biggest departure from the 45o line.
Here is a graph of how the two series have tracked the current recession:
The decline in the LFS numbers started to slow several months ago, but the SEPH number continued to decline sharply through June. Instead of being a puzzle, the contradictory July numbers may simply be a correction of a short-term divergence of the two sets of employment numbers.
But that is clearly not the end of this story. Is the LFS the release to look at, and the SEPH is a late, meaningless distraction? Or should we be ignoring the LFS and waiting for the SEPH release? When should we expect the two series to agree or disagree? Etc?
After Nick's spectacular success in this post, I'm hoping that someone out there knows more about the nitty-gritty details of the two surveys than I do, and is willing to explain them for us.
I believe that total LFS employment is about 15% higher because it includes self-employment, whereas SEPH is only payroll employment.
Statistics Canada includes a section comparing LFS and SEPH near the bottom of its SEPH release.
Posted by: Erin Weir | October 04, 2009 at 09:10 PM
I'm no expert, but the web link Erin helpfully provided implies the difference is because the same question is being asked of different people, so you get slightly different answers: "The information source is the key distinction between the two surveys: SEPH provides information related to occupied jobs based on a census of administrative data from businesses whereas LFS provides information on the employment characteristics of individuals based on a survey of households. As a result, from time to time, the estimates from the surveys can differ on a monthly basis. However, over the longer term, the estimates move in similar directions." However, the link also notes the two surveys treat the self-employed differently.
As for which one is better, StatsCan offers this unhelpful advice: "Since each of these surveys addresses different needs, the choice of data depends on the users' goals."
Posted by: David | October 05, 2009 at 12:36 AM
Self employment kicked in at the same time as the dot bust?
Are there that many self employed web designers?
If not that, what changed then?
Posted by: www.facebook.com/profile.php?id=611985302 | October 05, 2009 at 12:50 AM
Ah, thanks - I had thought of the self-employed, but I couldn't figure out why US analysts prefer to work with the establishment survey, which would also exclude them.
Posted by: Stephen Gordon | October 05, 2009 at 06:13 AM
Good question. One I ought to have thought more about.
Why does it matter? Because we want to predict future employment. Which series forecasts best?
I wish I understood VARs better (I wish i understood econometrics better). Isn't there some sort of test one can do to tell which series has better forecasting power? In other words, does LFS help SEPH predict future SEPH? Does SEPH help LFS predict future LFS?
My eyeballs seems to think that SEPH looks noisier than LFS. Both are surveys, and so both should have some white noise error associated with sampling error.
Posted by: Nick Rowe | October 05, 2009 at 07:21 AM
I also find the two paragraphs contrasting SEPH and LFS to be insufficient. Is anyone aware of a more thorough treatment?
Posted by: MJ | October 05, 2009 at 09:45 AM
The problem with counting self-employment is that it adds to “employment” many jobless workers who are forced into marginal self-employment and/or prefer to report self-employment rather than unemployment. Most of the self-employment total is genuine, but I suspect that most of the surge in self-employment over the past year has been involuntary.
Andrew compared LFS employees to the SEPH total, so that neither figure included self-employment. This approach, which approximates an apples-to-apples comparison, reveals an even larger July discrepancy than noted by Stephen: a decrease of 0.56% versus an increase of 0.51%.
Posted by: Erin Weir | October 05, 2009 at 10:16 AM
On the noise, LFS considers people on strike as employed while SEPH considers them unemployed. Might be small, but still a factor.
Posted by: Mark | October 05, 2009 at 10:21 AM
Another factor might be the underground economy. One of the sources for SEPH is CRA data on payroll etc, which for obvious reasons wouldn't include people that are employed, but not on record as such for tax or EI purposes.
Posted by: Jim Sentance | October 05, 2009 at 02:23 PM
I wonder if you could use the divergence between the two surveys to measure the size of the underground economy? Since the underground economy tends to include more services (child care, house cleaning, restaurants) and construction it might be more cyclical (but perhaps sex and drug sales would be more counter-cyclical, I don't know).
The LFS records employed and self-employed separately, I think, so that should be easy enough to net out.
Posted by: Frances Woolley | October 06, 2009 at 08:39 AM
"On the noise, LFS considers people on strike as employed while SEPH considers them unemployed. Might be small, but still a factor."
Wasn't CUPE Toronto and Windsor on strike in July?
Posted by: Andrew F | October 06, 2009 at 10:26 AM
If that was it, you'd expect the LFS to hold steady while the SEPH decreased. But the movement in July 2009 went the other way: the SEPH rose while the LFS fell.
Posted by: Stephen Gordon | October 06, 2009 at 10:31 AM