Tuesday 30 August 2011

The younger cohort drives it [updated]

Stats NZ has very helpfully provided some disaggregated HLFS data with results for 15, 16, 17, 18 and 19 year olds separated out. StatsNZ rocks. And so I re-ran things splitting the 15-17 cohort, who experienced a rule change in 2008, from the 18-19 cohort, who've been subject to the adult minimum wage for much longer.

The graph below plots the residuals from the very simple regression I've been running that predicts youth unemployment as a function of adult unemployment.

So, what do we see here? The blue line traces how youth unemployment outcomes for the 15-19 age group as a whole differ from predictions based on a model estimated on the period prior to the change in the youth minimum wage. The red vertical line marks the period break.  The green line tracks residuals for the 18-19 cohort; the red line for those aged 15-17.

As expected, there's a much bigger spike for the younger cohort who became subject to the new rules than for the older cohort who had previously been at the adult minimum wage. Outcomes for 18 and 19 year olds are worse as well, which I'd attribute to this cohort not having experienced this kind of labour market since they became subject to the adult minimum wage and to more eighteen year olds coming into age eighteen unemployed rather than in employment (note that the red line jumps higher and, importantly, earlier than the green line).

And, we can run a few other fun regressions.

Here, I take as dependent variable the number of employed persons in the age category (thousands) as a function of the population in that age category, the adult unemployment rate, an indicator variable equal to one for periods subsequent to the minimum wage change, and an interaction term between the adult unemployment rate and the indicator variable. For the 18 and 19 year olds, the indicator variable is insignificant and the interaction term is only barely significant at the 10% level. But the interaction term is significant at the 1% level for every age cohort from 15-17.

Each specification uses OLS with Newey-West standard errors for autocorrelation. (Newey in Stata, two quarter lag).

Recall that adult unemployment in the current quarter is 5%. So the interaction term (and the insignificant shift variable) for 17 year olds says that, after June '08, a 5% adult unemployment rate correlates with 11,100 fewer 17 year olds in employment than would have been the case prior to June '08 (17,500 fewer in employment from the interaction term, 6,400 more from the shift variable).

Employment is substantially lower for younger age cohorts - and the difference is statistically significant. For 19 year olds, there is no statistically significant difference in the post 2008 era. Eighteen year olds have a drop in employment, but the effect is smaller than for 15-17 year olds. And this is all about what we'd expect with a policy change affecting 15-17 year olds. In the prior period, some 17 year olds would carry through employment to age 18 and so fewer 18 year olds would be out on the market for the first time; employment among 18 year olds in the current era is then lower as well despite their not being directly subject to the change in policy.

There's more work yet to do. With the disaggregated data, there's now enough to make it worth writing up properly.

While we're talking youth unemployment, I'm going to be charitable and interpret John Key's assertion that, in the absence of minimum wages, youth pay rates would drop to a couple of dollars an hour as his just opening up room on the right for ACT. Employers do have to compete with each other for employees.

Update: 15 year olds are not subject to minimum wage legislation. Specifications looking at the unemployment rate for 15 year olds as a function of the adult unemployment rate find that the adult unemployment rate has no predictive power for the 15 year old unemployment rate except when we're looking at the period post the change. If the adult unemployment rate affects the number of 15 year olds in employment (second set of regressions) but not the unemployment rate for 15 year olds, it's doing it then through labour force participation rates: when there are no jobs going, the 15 year olds don't enter the labour market. And, as we'd expect that employers would worry about a massive wage hike when the 15 year old turns 16, that also directly depressed employment of 15 year olds subsequent to the change even if the change doesn't nominally affect 15 year olds.


  1. Hi Eric,

    It's been over ten years since I did labour economics, but obamas new economic advisor sounds familiar. I seem to remember he wrote one of the more controversial papers at the time.


  2. Hi Eric

    Long-time reader, first-time caller...

    I have three questions:

    1. In the second set of regressions, why do we seem to care only about the interactive term and not about the constitutive term for the date indicator? I would have thought the youth minimum wage explanation would imply BOTH a slope shift for young people relative to adults, AND also an intercept shift, meaning consistently more youth unemployment at any level of adult unemployment. How would interpret the non-performance of the constitutive term?

    2. I notice the cohort size term is much weaker for 15 and 16 year olds than for the other, I guess reflecting that most people in those younger cohorts are in school. Are goodness of fit measures also substantially weaker for those groups than for the older cohorts? If so, does that lead you to worry a bit about the overall performance of those two models in particular?

    3. How can we be confident that the effects on the indicator and indicator interaction terms are picking up the the imp[act of minimum wage changes rather other concurrent events such as, say, the global recession? I think you can make a pretty reasonable argument that massive recessions hit new workers especially hard (because of broadly LIFO personnel management in many sectors), and those new workers tend to be young. The younger they are, the newer they are likely to be. That argument would, I think, also be consistent with your empirics.


  3. @Michael: Alan Krueger is an awesome choice. He has indeed written papers finding no effect on US restaurant employment of changes in minimum wages over a period in which minimum wages were low relative to average wages. I believe his results as far as they apply. But if labour demand is less elastic in restaurant trade than elsewhere, and if we don't really expect minimum wages to bite hard if they're very low, then I don't see much reason to expect the results to apply to NZ where minimum wages are half the average.

    @Rob: I'm just going to go ahead and hoist this one into another post. 'Till I get around to that though, a couple points:
    1) In we look at specifications with unemployment rates instead of number employed as dependent variable, both the shift term and the interaction term are significant for 16 & 17 year olds but neither is significant for 18 & 19 year olds. But, the shift variable is negative.
    2) No huge differences in R-sq when looking at OLS specs rather than Newey
    3) I'd say not global recession because the adult unemployment rate already picks up recession effects. Prior recessions in NZ were far worse for adult employment and had far better youth outcomes. But you're right; I can't definitely exclude alternative coincidental things. But the differential responsiveness of 16-17 year olds to 18-19 year olds suggests minimum wages.