A composition effect in earnings growth and education attainment levels in Canada

I came across this post by Mickey Kaus a while ago, on trends in US earnings broken down by education attainment levels. From about the mid-70s to the mid-90s, earnings growth diverged sharply: increasing strongly for those with high levels of education, and falling for people with lower levels of education. Earnings growth has been more balanced since then, but the gaps carved out by this divergence have not been filled in.

This isn't an exercise we can repeat for Canada; this kind of earnings data only start in 1997. But I was curious to see what it would look like, and I found as neat an example of composition effects as you'll ever see.

Here are average weekly earnings of full-time workers with varying levels of education, scaled so that 1997=100. All groups saw increases, and you don't see the divergence seen earlier in the US - for example, earnings growth for university graduates were not any higher than they were for those with less formal schooling.

 

 

What's curious is the red line - average weekly earnings for all full-time workers. Ordinarily, you'd expect the line for the total to lie somewhere in the middle of its components, but the red line lies above all the other curves for almost the entire sample. How could the growth rate of the average be higher than the growth rates of all its components?

Of course, these are growth rates: the chart for levels is exactly like what you'd expect: earnings increase with education, and the population average is somewhere in the middle:

 

 

What's going on, of course, is that the population weights have changed: the percentage of full-time workers with university educations has been increasing, and the share of workers with high school or less has been falling: 

 

 

This is the composition effect at work. Even if earnings for each education level had stayed constant, the fact that more workers now have university degrees and fewer have high school or less would have been enough to increase the average in the population as a whole.

To my eyes, the shift has been astonishingly large and quick. In 1997, one full-time worker in five had a university degree; now one in three do. And in 1997, one in six did not have a high school diploma; that's now only the case for one full-time worker in fourteen. Of course, the shift cannot be entirely explained by an increase in education attainment rates in Canada: many low-skilled Canadians have found themselves squeezed out of the labour market and no longer have full-time jobs. 

You can even see the composition effect in median earnings:

 

 

(Of course, talking about composition effects with medians is tricky. The mean of a mixture is the weighted average of the means of the individual nodes, but there's no clean link between the medians of the components and the median of the whole.)

Anyway, there it is. The data don't go back far enough to pick up the divergence in the earnings, but they do turn out to be interesting all the same, thanks to a composition effect.

Comments

[Ian]

Do you happen to know how they classify a bachelors degree vs a university degree? I see StatsCan lists them as:

1) some postsecondary; 2) trades certificate or diploma from a vocational or apprenticeship training; 3) Non-university certificate or diploma from a community college, CEGEP or school of nursing; 4) University certificate below bachelors degree; 5) Bachelors degree; and 6) University degree or certificate above bachelors degree.'

I'm not sure what would count as a degree or certificate above bachelors that wasn't some sort of graduate degree.

[Kevin]

Good example of composition effects. You can go back to 1971 for earnings on an annual basis.

[Kent Charters]

There are other factors at work here that could be significant: e.g the age distribution of workers with different levels of qualification. With such a rapid increase in the proportion of more highly educated workers, the age distribution of those with lower levels of education increasingly "shifts upward" while the age of distribution more highly educated workers shifts downward, towards workers with lower levels of experience (below their maximum earnings age level, but increasing). As the older less educated workers age they also decline in earnings ability - probably more quickly than more highly educated workers. Credientialism is another possibly major factor in the decline in earnings of less educated workers, as educational qualification levels for most occupations have been shifting upwards for decades, so that able workers find themselves blocked from employment and advancement by more highly credentialed workers. The effects of credentialism likely result in overestimates of the value of education and limit the ability of older workers to reintegrate into the market following job losses.

[cm]

Kent: Credentialism doesn't lead to overestimating the value of education, only the value of the credential papers.

From my perspective in tech, "skill"/specialization requirements *have* increased. But stated in high-level qualitative terms, in highly "skilled" (and often highly specialized) fields, maybe 10-20% of the work require the high specialization, and 80-90% are bog standard stuff that competent generalists can do.

However, most (and I don't want to guess how many) work tasks of a size that measures in maybe days, or even hours, are "contaminated" with just enough of the 10%+ specialized content that they require an expert, or at least consulting with an expert. An additional issue is that it is often not clear upfront whether and when along the process this will be required.

With increasing complexity, also traditionally (perceived as) "lower level" tasks like testing, document preparation and review, customer assistance, product installation/maintenance etc. have started to require higher "skills" and specializations. In "high tech", the support staff often gets stuck in their job and in turn needs to escalate to product developers, either because of lack of specialized knowledge (and in software, notoriously lack of documentation) or because servicing the product simply cannot be done by non-experts anymore.

As a result of all this, companies were to some extent forced to hire more specialized and college educated or trained in the respective fields staff even for roles that are nominally "lower level" with the corresponding prestige penalty. To some extent this has in turn led to such roles being classified with higher titles ("title inflation"). A related trend has been to hire people in e.g. highly-classified product development positions but then expect them to "also" do support. This is very common in software, which suffers a lot from the described issues.

But of course there is the aspect that without unions and company pensions, and tenuous job security, work has become a "pay as you go" affair, with little mutual investment. So companies don't want to train people more than minimally necessary.