Will Wilkinson asks about inequality and the recession. I don’t know of anything ongoing data or analysis wise. Macro-economics blogger Will Ambrosini steps up and looks at 2008 growth -vs- 2007 income. That’s not strictly inequality, so I want to post a quick graph.

This is the 1 year increase in unemployment on the y-axis – not the unemployment, but the increase in unemployment to try and control for “natural rates” – against the 2006 Gini-coefficient, a measure of inequality, on the x-axis, by U.S. state and DC. It is positive, and almost statistically significant (p-value of 10.8%, t-stat 1.63). The rightmost outlier is District of Columbia.

Now of course the gini coefficient could just be a function of human capital – college educated workers have less unemployment. It is also probably high where there are a lot of employees in finance and/or where the real estate market was booming.

I’m not sure what’s the best way to set-up regressions like this, so I regressed the Gini-coefficient along with the percent of college graduates in the state, to mop-up human capital issues, along with the percent of total workers in the finance industry and percent of total workers in the real estate industry as good proxies for unemployed directly related to the boom – real estate turned out to be a cleaner estimate than construction (data from Census).

This is a down and dirty estimate of course. I may try this again at the county level later when I have more time. What I want to note, however, is that the gini coefficient is the only variable of those listed that comes close to clearing a statistical hurdle. We want to get that p-value ideally under 5%, though 10% is used sometimes. (Taking gini coefficient out doesn’t help the others’ p-values.) Note also that the percent of workers in finance has a negative sign. What are ways to make this regression better?

And what is it about inequality that could cause extra unemployment in a downturn, holding other things equal?

### Like this:

Like Loading...

*Related*

[We want to get that p-value ideally under 5%]

hi ho, hi ho, it’s off to data-mine we go.

Leave it, dude. The regression is telling you the same thing as the scatterplot, which is that there’s some reationship there (or at least, some degree of association), but that it’s small compared to the variance of the data. Even spotting you the 5% cult (which everyone belongs to to a certain extent), torturing the model to give you more significance than there really is, is bad practice. Leave this one how it is.

d^2,

Ha! If I hack off enough limbs, this patient will survive!

I’m curious if there are accurate ways to do U3 unemployment projections with any variables. Does it have to be done at the MSA or county level? States are probably too unwieldy.

He’s probably thinking of the Piketty/Saez data, which show inequality falling during the Great Depression and WW2. Calamity is leveling, I guess: http://elsa.berkeley.edu/~saez/piketty-saezAEAPP06.pdf

As for inequality causing extra unemployment, I know of two models that predict that. 1) Jamie Galbraith sees inequality and unemployment as two sides of the same coin. You see it in megacities in developing countries. Urban wages are much higher than rural wages but city jobs are scarce. So in the city you have high unemployment because people are hustling for scare jobs.

2) Akerlof and Yellen’s fair wage-effort hypothesis also predicts higher unemployment for low-wage workers in the presence of a wage gap. This one is a little over my head.

Nice Chris. Also: “fair wage-effort hypothesis”? Sigh, looks like I’m hitting the googles.

I agree with dsquared; “almost statistically significant” = there is no there there. Human brains are great at seeing patterns, but sometimes we see patterns that aren’t really there. That’s what statistical significance testing is *for* and we abandon (or fudge) it at peril of being deceived by our own pattern-seeking brains.

There is, so far, nothing to explain here. However emotionally unsatisfying that may be to narrative-loving hairless apes, it’s what the data show.

I dunno; a 10% significant p-value says that there’s some there there, doesn’t it? There’s nothing sacred about the 5% value. You can divide it into boxes (ahh, my favourite subject, robust estimation) and note that no state with a Gini above 0.45 had an unemployment increase less than 2%, and so on.

My suspicion is that if I dig lower than the state level, say at the MSA or county level, it’ll light up. State level is going to have a lot of noise and contradictory impulses, so if it has a bit of an itch it is probably because the real story is underneath. I’ll probably get to it at a later time.

You can divide it into boxes (ahh, my favourite subject, robust estimation) and note that no state with a Gini above 0.45 had an unemployment increase less than 2%, and so onNo state with a Gini above 0.45 had an unemployment increase higher than 6%, either. With only 50 states I don’t think empty boxes mean much.

The high p-value means that a shotgun loaded with 50 pellets would make a pattern at least that strong about one time in ten.

I suspect that inequality is the wrong question. Some industries are more recession-vulnerable than others, but there may not be much (if any) relationship of the wage level of an industry and its recession vulnerability. And sufficiently geographically dispersed industries won’t show up at all on state or even county level data – every county has fast food joints roughly proportional to its population.