Alberta’s opposition NDP have been hammering the government recently with the accusation that the province has lost “50,000 jobs” under the UCP’s watch. The data, however, doesn’t bear that out.
While employment appears to have been on a downward trend since last summer, the scale of job losses is almost certainly being overstated by NDP MLAs who keep repeating the “50,000” figure.
How so? It has to do with the way Statistics Canada’s data is understood — or misunderstood.
Every month, the federal agency publishes the results of its Labour Force Survey, which forms the basis for the news reports you hear on things like the unemployment rate. But there are some nuances to the numbers that are often lost, especially when figures are taken in isolation.
It’s important to remember that these monthly numbers are estimates of employment, and those estimates come with a pretty hefty margin of error.
That error can be compounded when you simply pick one month and compare it to another month, which is why Statistics Canada advises looking at longer-term trends when trying to figure out what’s actually happening in the economy. To get an even better picture of reality, the national data agency also advises using the Labour Force Survey in combination with other jobs measures.
And that’s exactly what the NDP is not doing when they claim “50,000” jobs have been lost since last summer.
This, it should be noted, is not unique to the provincial party. Politicians of all stripes in every part of the country have cherry-picked numbers from the Labour Force Survey to bolster their positions or attack their opponents.
But let’s look at this particular claim, to see why these types of claims should generally be taken with a grain of salt.
The evolution of the ‘50,000’ claim
The figure became a part of the NDP messaging in January, after new Labour Force Survey data came out showing an estimate of full-time employment that was nearly (but not quite) 50,000 workers below the estimate from the previous June.
The UCP’s corporate tax cuts started on July 1, so this offered an opportunity for the opposition to tie the employment numbers to the policy change.
Senior members of the NDP caucus started quoting that figure and saying the job loss happened “since Kenney’s $4.7-billion corporate giveaway.” Former NDP finance minister Joe Ceci called the government’s record on jobs “catastrophic,” citing the figure as evidence. NDP Leader Rachel Notley cited the 50,000 number as a “fact.”
As time went on, references to the “50,000” figure continued, but the NDP started to drop the “full-time” qualifier before the word “jobs.”
“You’ve lost 50,000 jobs since being elected,” finance critic Shannon Phillips tweeted at Premier Jason Kenney, a claim she has repeated over and over on social media.
This made the claim more misleading because, even by the method the NDP is using, the latest data doesn’t show 50,000 fewer workers in January than in June. When looking at all types of workers, both full-time and part-time, it’s more like 36,000.
But there’s another issue with the method, itself.
The trouble with comparing limited data points
The inherent uncertainty in the monthly estimates can magnify the appearance of swings in employment, even if there’s been little change in reality, when you compare a limited number of data points.
That’s because the margin of error in these estimates is larger than many people realize.
So while the January survey gave us an estimate of 2,318,500 workers in Alberta, that comes with a margin of plus or minus 31,400 at the 95-per-cent (or “19 times out of 20”) confidence level. (The level you are likely familiar with from political polling and public-opinion surveys.)
The error can also compound if you pick just two months to compare, says John Santos, a Calgary data scientist who specializes in polling.
“If you just take two arbitrary data points, it is possible that one is reading high and one is reading low,” he said.
“Depending on which one is reading high and which one is reading low, you might actually get a more accurate picture of reality — or an even more distorted picture of reality. So that is definitely a danger there.”
A motivated person could also deliberately look at a dataset and choose two months to compare that make the difference look especially large or especially small, depending on their motivation.
To get a more accurate picture of reality, StatsCan suggests looking for trends across a larger number of data points.
“It may be interesting to look at specific periods, but we also recommend our users not only look at a specific point in time but tend to focus on long-term trends,” Vincent Hardy, an analyst with the agency’s labour-data division, previously told CBC News.
Looking for the underlying trends
To that end, StatsCan also publishes a “trend-cycle” estimate of employment, which provides a longer-term view that smoothes out much of the “noise” in the monthly estimates.
“These smoothed data make it easier to identify periods of positive change (growth) or negative change (decline) in the time series, as the noise of the irregular component has been removed,” Statistics Canada explains. “This allows for a more accurate identification of turning points in the data.”
Looking at the data this way, it does appear that employment has likely been trending downward since last summer. But the figures do not demonstrate that 50,000 jobs have been lost.
If you’re looking for harder numbers, the lack of precision in the Labour Force Survey data may be frustrating.
Fortunately, it’s not the only way to measure employment.
Another dataset, with its own pros and cons
There is another way that Statistics Canada measures jobs.
Known as the Survey of Employment, Payroll and Hours, this dataset is garnered from multiple sources, including payroll-tax data, which is used to count the number of jobs in various industries across the country. The upside to this measure is that “there is no statistical uncertainty associated with the employment estimates.”
The downside is that it doesn’t measure all types of work. It only counts payroll jobs, meaning many farm workers and people who are self-employed aren’t figured into the data.
Another key difference is that the Survey of Employment, Payroll and Hours counts jobs, while the Labour Force Survey makes estimates about the number of people working. So someone with two part-time jobs would have both those jobs count separately in the former, but would only count as one worker in the latter.
One other difference is that it takes longer to count all those payroll jobs than it does to conduct a survey, so the data is about a month behind. We only have payroll jobs numbers up until December right now.
Here’s what that data looks like, over the past five years:
As you can see, it looks less dramatic than what the NDP has been describing.
While things have been trending downward in recent months, jobs haven’t fallen off a cliff. The decline has been steady since August, and there were about 8,800 fewer jobs by December.
Of course, while this count is highly accurate, it’s important to remember that it doesn’t include all jobs and that it double-counts people who hold two jobs. So it’s not a perfect measure, either.
“Reality is much more complex than any single statistic,” said Santos, the data scientist.
“Data never speak for themselves.”
Statistics and politics
With respect to Alberta jobs numbers, in particular, Santos noted the political parties have switched roles compared to when the NDP was in power.
Back then, it was the opposition UCP that was routinely using monthly employment estimates to hammer the government of the day, whenever the numbers showed a job loss. (And being notably silent when the figures showed a month-to-month job gain, which is when the NDP typically piped up.)
Now, Santos said, “the shoe is on the other foot.”
“So I’m not saying that the NDP is being disingenuous, but I think any political actor — whether it’s government, the opposition, other parties, advocacy groups, etc. — is always going to portray statistics in the light that is most favourable for the particular argument they’re trying to advance,” he said.
“And I think the job of disinterested observers — whether it’s economists, data scientists or journalists — is to point out: ‘Hey, look, you might be reaching a bit far with that particular claim.'”