Break out the party hats and kazoos, Canada. As suggested last month, the April 2014 Labour Force Survey (LFS) release marks the 100-month anniversary of the survey collecting data on but not mentioning immigrant labour market outcomes in the monthly report.
Justifiably, many readers are still surprised to learn the LFS even collects data on immigrants’ labour market outcomes. Neither Statistics Canada’s official monthly releases nor Canadian Labour groups’ respective ‘analysis’ ever mention immigrants (but don’t get the latter started on temporary foreign workers…). That said, the immigrant labour market data collected is unreliable, broader LFS sampling methodology contributing to / exacerbating the problem.
The limited LFS immigrant cross-tabulations StatsCan does provide are found in CANSIM tables 282-0101 to 282-0108.
While CANSIM tables showing underemployed and discouraged workers (neither factored into the unemployment rate) by age and gender are readily available, that info’s not available for the only other core demographic variable in the LFS – immigrant status. (Last month’s post covered the curious omission of race from the survey.)
That’s usually not a problem for most economists, as more detailed cross-tabulations can be produced using public-use microdata files (PUMFs). Unfortunately, the Labour Force Survey Microdata File doesn’t include all the LFS variables. While the LFS PUMF is updated every month, it’s never included immigrant status variables – as noted, that data’s been collected since January 2006.
Asked why that specific set of variables continues to be omitted, StatsCan doesn’t readily have an answer – although it’s quick to note those variables are among the set being considered for addition to the revised LFS PUMF. So ‘maybe’, in 2018. Not terribly encouraging.
The US and UK, in addition to asking about race, also include both race and immigrant status variables in their public-use microdata files. And one needn’t be an economist or proficient with statistical software to use their data. Curious readers can try the US BLS Data Finder, US Census Bureau Data Ferret or UK Data Service.
‘Cost Recovery’, immigrant labour market data
StatsCan’s default reason for not doing things these days is budgetary restraint. In some cases, that argument has merit. Where the agency wilfully withholds data it had received funding to produce is not one of those cases.
Seeing as StatsCan has chosen to omit data on underemployed and discouraged immigrants both from its published tables and public-use microdata, the only way one could access it (short of receiving a SSHRC grant and RDC access) is by special request to StatsCan.
Unfortunately, StatsCan only provides the data on a substantial ‘cost recovery’ basis for the ‘custom’ work. While the requested (monthly) series were never provided, after more time and effort than it was worth – but no payment, on principle – StatsCan did provide the following annualised data
Part-time employment by reason for part-time work (based on new definition), age group, sex, immigrant status, Canada, province, annual average 2013
Persons not in the labour force by reason for not looking for work by Immigration status, sex, age, Canada and province, annual average 2013
Not worth the effort, sampling issues
A quick review of the data raises concerns the survey sample wasn’t representative of the Canadian immigrant population. For example, the data shows little difference between unemployment of immigrants arriving within the preceding 5 years and preceding 10 years. Most of the research suggests there’s a fairly significant difference in their respective labour market outcomes. Practically speaking, it’s the difference between labour outcomes before obtaining citizenship (typically a five year process), and after.
A quick peek at the sample distribution (also obtained by special request – minus the ‘cost recovery’ solicitation) shows that in fact the sample isn’t representative.
Unweighted counts of individuals in the LFS, by immigrant status and CMA/CA, average for 2013
A quick review of the immigrant sample for the major urban centres of Montreal, Toronto and Vancouver shows immigrants are under-represented (relative to their population distribution). Immigrants are slightly over-represented within the three CMA’s samples – 24.6%, 49.7% and 42.5%, respectively. However, the three cities are remarkably under-represented within the total sample – 4.3%, 5.5% and 4.4%, respectively. (In terms of (estimated) total population: Montreal 11.3%, Toronto 17.0% and Vancouver 7.0%).
What you end up with is immigrants comprising 16.1% of the 2013 LFS sample when actually accounting for an estimated 23.5% of the Canadian population (non-institutional, age 15+). Concurrent general under-representation of immigrants and specific under-representation of the three major urban centres where the vast majority reside is a recipe for bad data.
Under-counting immigrants in large urban centres – where their rate of unemployment tends to be higher, while over-counting them in smaller urban centres – where their unemployment tends to be lower (immigrants tend to move away from large centres only when employment opportunities arise), would understate immigrant unemployment, underemployment, wage gap and other labour market indicators. Likely by a lot.
The same large urban centre under-representation could also be significantly understating poor employment outcomes in general. Toronto and Montreal have significantly higher unemployment rates than many other major urban centres.
Consequence of doing stats on the cheap
To be fair, there’s a reason StatsCan does its sampling this way, and it’s one of those not reasonably justified on the basis of cost efficiency.
In addition to reporting household labour market attachment, the LFS is also used to provide the unemployment data for HRSDC’s EI Economic Regions, used to determine EI qualifying hours and weeks entitlement. In an effort to get more representative samples of less populated regions without jacking up the overall LFS sample size – which would cost more, StatsCan resorts to this skewed sampling methodology. It’s worth noting the US doesn’t (ab)use its labour force survey in this manner (unemployment benefits qualification and entitlement are based on state payroll data, not the monthly US labor force survey).
As the old adage goes, you get what you pay for. In the case of the LFS, that’s a non-representative sample and unreliable employment and wage data.