Friday 23 September 2011

Unemployment and the youth minimum wage

I'd noted, initially with a couple errors, some of the differences between my findings and Hyslop and Stillman's on youth minimum wages. Where I'd found big effects of the youth minimum wage change on the youth unemployment rate, they found only small and statistically insignificant changes on the percentage of youths who are unemployed. I was a bit puzzled why they ran things on the latter measure - percentage of youths unemployed is hardly a headline HLFS result. But, thinking more on it, and subsequently clarified by email with Steven, there's really good reason for it.

Hyslop and Stillman start by asking what the effects of the youth minimum wage change are on the likelihood of being employed (recall that they're using individual-level data at a StatsNZ data centre, not aggregates). They find a reasonably large and statistically significant decrease in the likelihood of a 16 or 17 year old's being employed consequent to the change in legislation - about a six percentage point decline, or about a 9,600 person decrease in employment.

They then ask what happened to those kids - if the employment rate is down, where are they? The percentage of youths unemployed is the best way of answering that question. The likelihood of being unemployed went up a couple of percentage points, but the result wasn't statistically significant. They find instead that most of the effect was reduction in the employment rate of students who had been working part time, who would then drop out of the labour force rather than show up as unemployed or as "inactive".

As Matt Nolan over at TVHE points out, the welfare effects of this are a bit more ambiguous than the Greens have been suggesting. The Greens have been arguing that the minimum wage increase was great - no significant increase in unemployment and more kids stay in school. But that's entirely too quick. The drop, according to Table 6, has mostly been among students who combined work and study. So students are perhaps shifting to a stronger focus on their studies, but they're also going to be poorer and they're not going to be accumulating human capital in the form of experience that may well be complementary to human capital gained through education.

If our working model is that kids are short-sighted idiots (possible) who ought to be forced to accumulate human capital in the form of education only rather than in the form of work experience mixed with education, even if they have demonstrated that they believe the work experience to be valuable, then the change might not be awful.  If our working model is that kids are heterogeneous in converting education and experience into human capital, and have reasonable expectations about their individual marginal human capital increase coming from education or from work experience, pushing kids from higher valued work experience into lower valued education isn't all that great a deal.

Hyslop and Stillman also find significant negative effects on hours worked for 16-17 year olds in 2009 and 2010 on the order of 1-2 hours per week. While they're earning more per hour, they're getting fewer hours.

It's a bit fun to compare the number they get with the fancy techniques and individual-level data with what I get with my low-tech approach and aggregate data.

Again, I regress outcomes for 16 & 17 year olds on outcomes for other cohorts for the period prior to the minimum wage change then predict outcomes subsequent to the change given the performance of the comparison cohort. When I use 18 & 19 year olds as comparison cohort (Hyslop & Stillman use both 18 & 19 and 20 & 21 year olds; I haven't the latter data), I get a nine percentage point drop in the employment rate in 2010.

If you work backwards from their figures, you can piece out the expected change in the unemployment rate among 16 & 17 year olds - it looks to be about a four percentage point increase that's due to the youth minimum wage if you use average 2007 against average 2010, or five points if you use Q4 in each. Their result on unemployment isn't significant but their result on employment is; I'm not sure whether they'd then have a significant result on the unemployment rate driven by the likely significant drop in the denominator.

When I run things against 20-24 year olds, I get a 10.5 percentage point increase in the unemployment rate for 16 & 17 year olds in 2010; against 18-19 year olds, I get a 10 percentage point increase.

There are a few reasons for the difference. First, I always get larger effects when I use all adults as the baseline comparison rather than the 20-24 year old cohort. It's not crazy to then expect smaller effects if I were able to use a baseline cohort of 20-21 year olds.

Instead of conditioning a model on the early period and projecting forward, I've also run things using dummy terms and interactions for different regimes. Specify three regime periods and two regime variables. Regime 1 runs 1986 through 2001Q2. The second regime begins 2001Q2 - 18 and 19 year olds become subject to the adult minimum wage. Finally, regime three begins 2008Q2 - 16 and 17 year olds become subject to the adult minimum wage. I'm dropping the squared terms from the specifications because there are only thirteen quarters in the final regime period. And here's what I get*:

Unemployment rate, 16-17

Unemployment rate, 20-240.808***
ur2024 * Regime2 -0.284
ur2024 * Regime 3 1.471***
Regime 2 2.028
regime 3 -9.816*
Constant 8.714***

Observations 102

t statistics in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001

In 2010, when the unemployment rate among 20-24 year olds averaged 12%, the effect of being in Regime 3 was (1.471*12) - 9.816 = a 7.9 percentage point increase in the youth unemployment rate for 16 and 17 year olds. In 2010Q4, when the unemployment rate among 20-24 year olds was 11.2%, the effect was a 6.7 percentage point increase. And that's close to the 5 percentage point increase you can back out of the Hyslop and Stillman figures for 2010Q4. [Note: if I drop the regime2 variables so I can keep the squared terms on unemployment, the effect of Regime 3 gets bigger, not smaller - a ten percentage point increase.]

I'm still a bit puzzled by a couple of things. Why does the percentage of cohort unemployed stay roughly constant while cohort labour force participation plummets? In other words, why is it the employed that jumped over into education rather than the unemployed? I've been saying about 10 points of the run-up in unemployment has likely been due to changes in the youth minimum wage; Hyslop and Stillman say about five points (though that may not be statistically significant). If Hyslop and Stillman are right, what then accounts for the massive disproportionate increase in the youth unemployment rate relative to the older cohorts' rates in this recession as compared to prior ones?

* esttab in Stata is awesome. You type "esttab, label html" and it spits out html code for regression tables.


  1. Wouldn't this be an ideal setting for a differences-in-differences estimate? And isn't it time to convert the phletora of posts into a paper?

  2. Really interesting reading and thinking about the youth emplyoment changes.

    I'm not sure that I agree with you that the type of job that pays minimum wage (either youth or adult) is really adding much in the way of human capital though. It would be interesting to use some LEED EOTE data to investigate labour market outcomes from those that chose early poor employment vs. those that were only in education.