The snakes-and-ladders model, again

They say the first rule of editing is “kill your darlings”. The first rule of science, however, is “sacrifice your darlings humanely in accordance with the research ethics committee guidelines, but keep their brains for further investigation”. So it is with this post. Per Mason, apparently the GFC looks a lot different to past recessions and it’s important to include the full span of the ECEC data back to 1986. So here goes.

The ECEC civilian workers series doesn’t go back to 1986, and neither does all workers, so I picked on the series for “private industry” – after all you’d expect public sector employment to be less responsive to the business cycle by definition – which does. The orange dots on the chart mark ECEC data points, while the blue ones mark the composition-weighted ECI series for private industry. ECEC after 2002 is quarterly, and is averaged to give an annual figure.


The big orange outlier is 2002. ECEC was issued as a slightly different series in 2002-2003, so perhaps we should exclude that one. I’ve plotted regression lines for the two series, orange and blue respectively, and for ECEC excluding 2002, black.

As you can see, ECI is still more cyclical than ECEC (R^2=0.33 vs 0.03 – ten times as much). Excluding 2002 helps a bit, but not enough (R^2=0.33 vs 0.11, three times as much). The correlation between the change in ECEC for the private sector and the output gap is 0.19, and that between private sector ECI and the output gap is 0.58. The blue outlier is 1985-1986.

Purely visually, it looks like the difference between the two series is in fact greatest in the 1980s, but any effect is accounted for by three data points (’86, ’87, ’88) and in any case, it seems to have disappeared 30 years ago. Alternatively, the BLS has changed the method it uses to collect ECEC several times in that period and this might be an artefact.