Laboured Data – Reading the Recession Right
I purchase a monthly unadjusted Labour Force Survey data series from StatsCan that provides monthly labour force trends by age, sex, province, and type of job (full-time, part-time, by industry, and by status â€“ self-employed or employed). This is a helpful addition to the published monthly stats in The Daily, which use seasonally adjusted numbers from the Labour Force Survey. Together, you should be able to track how the labour market is transforming, given both job losses and job creation occur every month, even in the midst of a recession.
For the past few months Iâ€™ve noted some striking differences between this series and the published one. The differences are not just because of seasonality of the data.
In what follows, Iâ€™ll show you these differences in the two seriesâ€™ findings for October, but first let me first acknowledge: You really should NOT compare month to month in seasonally unadjusted series. Unadjusted series are best used to compare year over year changes.
However, the unadjusted series is what the adjusted series is based on.
It is expected that there could be agreement in direction but not scale when comparing the unadjusted versus adjusted series. Even when the differences look significant, they can be primarily driven by seasonal change.
It is altogether another matter when the trends go in contradictory directions, and the scale of the difference is significant. It is problematic if one series says the job market and/or certain sectors is growing and the other says it is shrinking.
Todayâ€™s Daily says all the employment loss in October, compared to September, was driven by part-time job losses, 60,000 of them. The unadjusted series shows the opposite â€“ overall about 47,000 part time jobs were added, 39,000 full time jobs lost, comparing October to September .
On other fronts, there is agreement. For example, both series point to the fact that prime-aged women (25 and up) lost the most jobs this month.
Here are the top notes in comparisons between the two series. The first number in each bullet point is from the custom series, the second (in parentheses) what was published in the Daily today.
â€¢ 39,000 fewer full-time jobs overall (The Daily says 16,000 more full time jobs)
â€¢ 47,000 more part-time jobs overall (The Daily says almost 60,000 fewer part time jobs)
â€¢ 27,000 more public sector employees overall (though 17,000 fewer full-time jobs, so most is part-time job creation). (The Daily says 26,000 fewer public sector jobs)
â€¢ 44,000 fewer private sector jobs (because of 51,000 fewer full time private sector jobs). (The Daily says 54,000 fewer private sector jobs)
â€¢ 25,000 more self-employed (30,000 more full-time self-employed) (The Daily says 27,500 more people are now self-employed)
Based on all workers aged 15+, the custom series indicates that, on an unadjusted basis, there were 5,000 fewer women working overall in October than in September, and 14,000 more men working overall. The Daily doesnâ€™t offer that breakout. Based on all workers aged 25+, the adjusted data indicate there are about 1,000 more men working and 24,000 fewer women working. There are also about 20,000 fewer youth, aged 15-24 working according to the Daily.
The gendered differences are important, as this recession has been labeled a â€œhe-cessionâ€ both here and in the U.S., with the lionâ€™s share of job loss being borne by men, and women propelling much of the gains in job creation. (See â€œCanadaâ€™s He-Cessionâ€ written with Trish Hennessy, and published in July 2009. Available at www.policyalternatives.ca)
This month the male/female splits seemed to reverse. The series used in The Daily does not permit such analysis, so the following is based on the unadjusted series.
Until now, women have been driving growth in self-employment during this recession, but this month men fuelled that growth. Roughly 20,000 of the 30,000 new full-time self-employed positions added to the labour market between September and October were created by men. In the domain of self-employment, women added 10,000 self-employed full-time positions to the mix, and 14,000 self-employed part time positions.
Among the self-employed this month, women were much more likely to add to the category â€œincorporated with paid staffâ€. Women are starting businesses and hiring people; men are not. (There are 13,000 more such positions among women, versus 3,500 fewer such self-employed among men.)
This month, men piled into the category â€œunincorporated self-employed, without paid helpâ€ â€“ 16,000 more men in this category, October compared to September. It could be because unemployed men have run out of EI or are just picking up whatever work there is. The biggest growth in self-employment for men in October compared to September was in agriculture, construction, transportation and warehousing. There were some increases also in business and building support services, as well as in the information and recreation sector.
For self-employed women, the biggest growth is in service sectors: information/culture, trade (presumably retail, opening up a shop), and finance/real estate sectors (real estate agents, notably). In the goods-producing industries, construction is the big winner among self-employed women. Interestingly, they are also in the self-employed categories of those without paid help. So at least in the construction sector, self-employed women are tending to do the work themselves rather than starting businesses and hiring others.
As in the past months, the over 55s is where most of the job growth is taking place. Itâ€™s hard to tell if this is demographically or economically driven. (Every month, more people trip over that line that demarcates over 55. Going forward, the numbers will only continue to escalate.)
Women aged 55 and over lost about 12,000 jobs last month as employees, uniquely driven by changes in the private sector; but there are about 20,000 more self-employed women over 55 this month compared to last. There are also roughly 20,000 more self-employed men aged 55 or over this month than last, and this group faced fewer job losses as employees (8,000, again strictly a private sector phenomenon).
Other than government-created jobs, self-employment is the big news story of this recession. But there has been almost no change among 25- 54 year olds on the self-employment front (a little more action among women taking up self-employment than men, but really mild) The 25-44 year old men lost 20,000 jobs in the private sector (but added 10,000 jobs through the public sector). Among 25-54 year old women, about 18,000 jobs lost in the private sector, about 9,000 added in the public sector.
It takes a little longer to do the analysis for youth (15-24) using these data.
Soon Iâ€™ll provide the year over year analysis.
Meanwhile, I have asked Statistics Canada staff why there are such large discrepancies, and which would be the best series to use to track developments during the recession. The first time I asked was in July, the second time was this week. Thus far no answers. I will keep you posted.
Armine, if you do not get a response from Statcan email me and maybe we can pick away at these bones a bit and find out what is going on.
I will say this up front, given the magnitude of the changes going on, seasonal adjustment using the algorithms statcan uses can make the changes at lower levels of reliability go a bit hay wire.
Think about it- if we are getting these huge movements from short term change, we know that the seasonal adjustments are most likely doing more damage to the data then they are doing good. But you better not quote me on that. It is a tricky space, and at best one most likely only deal with this seasonal adjustment “bias” in the short term, at the Canada level where most of the reliabilities are within a zone that is a bit more robust.
I guess my question would be, given all these huge changes in the short term, is any of this season data at the sub-canada level usable. Personally my opinion is barely usuable.
So then the question is, can one use the unadjusted data. Yes, at least you know what you looking at. Especially if you get into a sub-domain where seasonality is smaller in proportion. I am all for seasonally adjusted data, but not when things jump around like we are witnessing and especially sub-domains with a lot of seasonality.
Hope that helps
I would prefer to use seasonally adjusted data, if only because that is the industry standard, and permits direct comparisons to and further understanding of what the Daily reports.
However, StatCan sells custom data on an unadjusted basis for a reason – reality is seasonally unadjusted.
The problem for me is not things jumping about – I understand that surveys will produce those kinds of variations from time to time.
The problem is that one data series says full-time jobs on the rise, the other says the opposite; or one series concludes the job market is shrinking, the other that it is expanding. This has not been a one-off during the recession period.
If I get time on the weekend I’m going to calculate three month moving averages from the raw unadjusted data to see if it tracks the Daily results more closely.
I’d be more convinced something is amiss if there was a significant discrepancy between year over year changes for the month of October, comparing actual and seasonally adjusted data.
It is possible that the recession has changed factors which play into seasonal adjustment eg. there could have been bigger than norrmal reatial layoffs post Xmas, bigger than normal education sector layoffs in the Summer, much lower than normal hring of young workers in the Summer.
Yes that is the problem Andrew.
Typically seasonality has a pattern to it. The ARIMA II seasonality algorithm can make these adjustments and thereby smooth out the data.
However, when one enters into a period of dramatic change, such as the recession we are facing, the seasonality patterns can become quite “off the rails”, if you would.
This can therefore make the process of seasonally adjusting a bit of a mess. And to me that potentially is what Armine is going through with the task at hand of comparing Adjusted with Unadjusted.
However finding trends that are mentioned about going in opposite directions for extended periods might suggest something more is a foot.
I’ll try and have a look on the weekend as well.
Just wanted to clarify something-
my above comments suggests that some of what you are finding may be due to what I have suggested. However, it may only account for a small proportion of the problems you are uncovering. There could be a whole lot more going on.