Accountability Employment

UK Office for National Statistics responds to concern over self-employment. StatsCan? Not so much.

Chart 1a Earnings, income by type of work and class of worker (usual hours)
Chart 1a Earnings, income by type of work and class of worker (usual hours)

Source: Census of Population, 2006 [Canada]: Public Use Microdata File. Individual File: Canada. 2.7% sample, Statistics Canada.

To the UK…

Self-employed workers in the UK – 2014
Office for National Statistics August 20, 2014

Readers unfamiliar with the Bank of England quarterly should check out the press conference that followed release of the August 2014 Inflation Report. The increasingly embattled Bank of England (formerly Bank of Canada) Governor Mark Carney found himself facing reporters asking whether the Bank was ‘”clueless” and debating its “degree of cluelessness”.

A particular target of their ire was questionable LFS employment and wage growth data produced by the ONS and cited by the Bank to bolster its case for a sustained economic recovery. A recovery increasingly fuelled by the self-employed, from whom the UK LFS – like its Canadian counterpart – does not solicit self-employment income information.

UK economists and media have been letting their government hear it over the record high 15.1% self-employment share and the lack of sufficient self-employment data in the UK Labour Force Survey (LFS).

While the ONS deserves credit for its timely response, the report it released today will likely continue to fuel concerns. The report notes already poor self-employment earnings dropped 22% since 2008/9 as the number of respondents identifying as self-employed rose 16% to 4.6 million over the same period, the highest it’s been since the Office began tracking four decades ago. A recent Mindful Money post by economist Shaun Richards delves into UK concerns over rising self-employment.

… and back

As folks in the UK are in fits over record-high 15.1% self-employment, Canadians seem either oblivious or indifferent to the fact their LFS reports the same self-employment share.  It’s actually a bit down from late 2009 when it was 16.3%, or the late 1990’s when it was at 17.4%.

Further muddying the waters are StatsCan’s exceptionally poor efforts at reporting self-employment:

Like its UK counterpart, the monthly Canadian LFS does not collect earnings info from respondents indicating self-employment as their primary labour market activity. One rationale is It’s impossible to determine the share of earnings retained by the entrepreneur as personal income and the share re-invested in the entreprise on a monthly basis. Another is that, since the survey involves one individual responding on behalf of all household members, it would be difficult for such respondents to provide accurate self-employment and earnings info by proxy. (The second point is courtesy of an agency analyst – it’s worth noting the accuracy of proxied survey responses is a general problem, not exclusive to the LFS or self-employment.)

StatsCan could do like the US Bureau of Labour Statistics (BLS) and collect the self-employment earnings info once a year, addressing the first (and only material) rationale for not collecting self-employment earnings data. It doesn’t (although it kinda did with SLID, but more on that below).

The most recent (according to same agency analyst) StatsCan ‘report’ analysing self-employment earnings was published in 2009. Worse, it covers all of two years, 2000 and 2005. Worse still, it’s a couple of rather uninformative pages in a 127-page document titled The Canadian Labour Market at a Glance 2007.

The most recent StatsCan survey focusing on self-employment was the one-off Survey of Self-employment (SSE), an LFS supplement StatsCan conducted on behalf of HRSDC in April 2000. The little info provided on the survey’s methodology suggests it was designed to be uninformative and unreliable: Voluntary survey, 6623 individuals (sampled only once), excluded all respondents working less than 11 hours per week (despite LFS counting respondents who works as little as 1 hour as employed), 60.6% response rate, interviewer error botched 28% of responses.

That leaves administrative data from Canada Revenue Agency, compiled but not made publicly available by StatsCan (no PUMF); the Survey of Labour and Income Dynamics, which was also a less than reliable (overestimated 2005 self-employment income by a quarter) voluntary, small sample, low response survey before it was cancelled after its final 2011 release; and the long-form Census, which was cancelled and replaced by the previously referenced, wholly unreliable 2011 NHS. The mandatory long-form survey’s cancellation continues to cast a long shadow.

Not that StatsCan put much effort into analysing self-employment income prior to the mandatory long-form cancellation. To underscore that point, the following is all the self-employment analysis provided in the 51-page 2006 Census Income ‘Analysis Series’ (PDF) report:

On the other hand, self-employed individuals accounted for slightly more than half of the 47,360 gain in the number of those receiving $150,000 or more. In 2005, self-employed individuals represented 14.0% of all full-time full-year earners.

A bleak picture of Canadian self-employment

Chart 1a illustrates the self-employment income data from the 2006 Census PUMF, broken down by employment type (full-time, part-time) and class of worker (employee, self-employed). The chart intentionally cuts off at $100,000 to help focus on the very low part-time / self-employment earning, income figures. The lines represent wages for those who primarily worked as employees and self-employment income for those who were primarily self-employed.

Limitations off the top: The 2006 long-form Census collected earnings and income info for reference year 2005 – actually an advantage when it comes to analysing self-employment income (addresses the monthly collection issue mentioned earlier). Likewise, it asked respondents whether the majority of weeks they worked in reference year 2005 were full-time or part-time. However, for reasons not entirely clear, it queried respondents primary class of work only for the survey (May 2006) reference week.

That aside, there’s also an interesting little StatsCan quirk: While a respondent can record a legitimate “0” for earnings – for example, a self-employed respondent could either make no income, or break even and round their response to “0” – StatsCan only counts positive and negative values in calculating its income stats (means, medians). See footnote 1 below this NHS income table.

Those 0’s make a huge difference when it comes to self-employment income. While mean full-time self-employment income was pulled up to a meagre $21,465 by a few relatively high earners, those 0’s rendered median income practically non-existent at $1000. Even lower still was part-time self-employment income, with mean income at $6967 and median at an actually non-existent $1 (the 2006 PUMF metadata notes: “Values that would have been rounded to zero have been replaced by 1″).

As the chart (hopefully) illustrates, half of the 2006 Census respondents indicating self-employment as their primary labour force activity had no self-employment earnings. The issue with counting them among the employed in the LFS are obvious. Counting that many people without any income from their primary labour force activity as otherwise gainfully employed significantly overstates employment, while excluding their non-existent incomes from the data significantly overstates both the reported earnings figures and gains.

To address the obvious criticisms of such a basic analysis: Obviously those identifying as self-employed without self-employment income did not get by without any income. There are a number of possible scenarios: self-employed non-earners could have registered businesses for non-employment reasons, drawn on savings or investments, received other income from third parties, received government transfers, lived with other household members with an income,or possibly earned employment income from a second job.

All but that last one are irrelevant to the discussion of whether / how the self-employed should be included in the LFS. Regarding second jobs, according to the LFS there are about a third as many multiple job holders as there are self-employed in Canada, and most respondents reporting full-time hours at a primary job work limited hours at second jobs (keeping in mind the Census self-employment earnings illustrated are for those reporting their primary labour market activity as self-employment).

The LFS is supposed to be a measure of household labour market attachment, which includes earnings gained from labour market activity. On that count, it’s really tough to make the case for including a group of workers, many with no employment earnings from their main labour market activity, most with rather negligible earnings (for example, only the top 20% of the part-time, self-employed earned $10,000 or more).

When the LFS counts as fully employed a significant share of self-employed earning no self-employment income it understates the unemployment figures used to determine employment insurance qualifying hours and benefit entitlements. When at the same time the LFS excludes the earnings of all self-employed – including the significant share earning no self-employment income – it overstates wage / salary levels and gains used in the public and private sectors to assess income and transfer payment indexation.

As a number of economists have suggested, StatsCan needs to figure out how to get its employer payroll survey out at the same time it releases the LFS household survey, as the US BLS does. (Not that Canada’s payroll survey has been without its issues, up until recently reporting nearly half a million ‘unclassified’ jobs.)

If it’s going to continue reporting self-employment in the LFS, StatsCan should at the very least publish a regular, somewhat reliable report on self-employment earnings.

Note to readers:
 While the 2011 National Household Survey (NHS) PUMF has been out for a couple weeks now, readers are likely aware it’s garbage data (but don’t just take our word for it) that shouldn’t be cited.

Update 25/08/2014

Chart 1b Earnings, income by type of work and class of worker (non-zero, usual hours)
Chart 1b Earnings, income by type of work and class of worker (non-zero income, usual hours)

Source: Census of Population, 2006 [Canada]: Public Use Microdata File. Individual File: Canada. 2.7% sample, Statistics Canada.

Given the peculiar long-form Census issue referenced earlier – where income, hours and weeks worked are collected for the preceding year while employment activity / work class are collected for the preceding week – some analysts (including StatsCan’s) prefer to exclude zero incomes from the analysis. The reasoning is that a respondent who indicated being self-employed in May 2006 and reported no self-employment income in 2005 may not have been mainly if at all self-employed that year. Which may be true, especially given the volatility inherent in self-employment. The obvious downside is the exclusion of legitimate zero incomes.

Chart 1b illustrates the same data as 1a, less the reported zero incomes. It does, however still include the “values that would have been rounded to zero (that) have been replaced by 1″. Even with the more conservative assumptions, nearly a third of those reporting self-employment as their main labour market activity reported earning no (<$1) self-employment income. Also as before, summary statistics belie the extremely skewed self-employment income distribution. Excluding zero incomes, mean full-time self-employment income was $25,186, while median income was just $8000. The respective figures for part-time self-employment were $9616 and $4000.

Leave a Reply

Your email address will not be published. Required fields are marked *