Economy adds 82,300 jobs in March, as unemployment rate dips to 7.2%
Julian Beltrame, The Canadian Press April 5, 2012

Refering once again to CANSIM Table 282-0085 (Labour force survey estimates (LFS), supplementary unemployment rates by sex and age group, unadjusted for seasonality), the unadjusted R4 (the ‘official’ rate) and R8 (the slightly more realistic rate) are practically unchanged from February to March 2012.

Skipping forward a couple of tables to CANSIM Table 282-0087 (Labour force survey estimates (LFS), by sex and age group, seasonally adjusted and unadjusted), it is readily apparent that the increase in employment and dip in the unemployment rate in March 2012 is a product of seasonal adjustment.

The drop in unemployment rate seems to largely stem from the March 2012 seasonal adjustment which saw the estimated size of the labour force raised by 211.7 thousand and the number of jobs by 288.3 thousand (relative to the unadjusted figures).  The February 2012 seasonal adjustment saw the estimated size of the labour force raised by 252.6 thousand and the number of jobs by 290.1 thousand.  That difference between the seasonal adjustments for February and March 2012, 37.5 vs 76.6 thousand, accounted for practically the entire ‘positive’ change over the same period.  (Adding the difference of 39.1 thousand to the numerator  in the adjusted March 2012 unemployment rate calculation gives you an unemployment rate of 7.4%, the adjusted February 2012 rate.)

Yes, warmer weather may have played a part in kickstarting the construction season a bit earlier this year.  However, the March 2012 LFS report spreads the  gains across industry and occupation sectors.

Unfortunately, it appears this month’s strongly positive LFS jobs gain is entirely a statistical artifact.

Looking at the trend since October 2008…

The reported LFS figures involve seasonal adjustment not only to the number of jobs, but the size of the labour force.  It is a nice way of smoothing out the numbers and allowing for a consistent trend line.  Unfortunately, as was the case in March 2012, sometimes monthly movement simply stems from the seasonal adjustment and not any real change in the labour market.  For comparison, one can see from the preceding chart that for the same two-month period in 2011, the unadjusted unemployment rate jumped from 8% to 8.2% while the seasonally adusted rate dropped from 7.7% to 7.6%.  For the same two-month period in 2010, the unadjusted rate jumped from 8.5% to 8.8% while the adjusted rate held steady at 8.2 –  a 0.6% gap between the adjusted and unadjusted rates for March 2010.  That seasonal adjustment gap was 0.7% in March 2009.

From the chart, the direction of seasonal adjustment for the February-March period is obvious.  Perhaps there is encouraging news in this month’s numbers, but it does not stem from the monthly figures so much as the recent historical trend.  From 2009 to 2011, the February-March seasonal adjustment smoothed out jumps in the unadjusted unemployment rate of 0.5%, 0.3% and 0.2%, respectively.  As noted above, the unadjusted rate held steady at 7.7% between February and March this year.  Unfortunately, unemployment rate R8 remained high (11.3%), indicating that the trend in the official rate (R4) continues to mask labour market problems.

To the point, Statistics Canada, and the media who take their lead from its daily reports, should refrain from trumpeting positive monthly changes in the LFS, especially when it is obvious they are largely if not entirely a product of statistical adjustment.  As suggested by a few friends of this blog, including Toby Sanger from CUPE, perhaps more attention should be paid to actual job creation trends instead of estimated monthly labour and employment figures.

Statistics Canada undertakes another employment survey,  the Survey of Employment, Payrolls and Hours (SEPH) that is more useful for the purpose of counting jobs and related data (e.g. salaries, wages, industry classification, geographic distribution).  The Labour Force Survey (LFS) is more useful for collecting socio-demographic data about workers in the labour market (e.g. age, gender, immigration status, education, union representation, reason currently un(der)employed).

LFS is a sample survey, sample size <= 0.5% of the 15+ population in Canada.  Each monthly sample is not completely independent; rather it is a quasi-cohort survey, where respondents screened into the monthly survey remain in the sample for six months.  Without even mentioning the differences with the SEPH (which is not a sample survey), these two facts alone indicate why the LFS tends to be less accurate at any level of disaggregation (e.g by province,  industry) and why its monthly figures are notoriously unreliable at picking up trends.

That’s not what the LFS was meant for.  Yet articles like the one cited at the top of this post use it expressly for that purpose, and make no mention of the useful socio-demographic information it was designed to collect.