On the causal bit: So, you'd probably be better off skimming the paper for the details, but the gist is that they estimate the effect of unemployment on skills

On the causal bit: So, you'd probably be better off skimming the paper for the details, but the gist is that they estimate the effect of unemployment on skills requirements in a few ways, and the results all come out about the same. They use state-to-state variation to take out some of the bigger macroeconomic factors. They use the return of a huge number of troops when we pull out of Afghanistan for another change that is mostly uncorrelated with economic conditions and do some other clever stats to make that assumption more solid. (I briefly skimmed it hours ago when I made the parent comment, so I might be a bit off now, but that's the gist of what stood out to me). Showing causality is generally about finding apples-to-apples comparisons and finding sources of randomness causally upstream of what you want to measure the effect of, to get a kind of mini-experiment. They do this. It's not proof, but it's some decent evidence. ------------ On your question about percent vs absolute: Your idea is totally plausible and doesn't contradict the study, as far as my skimming goes. There's no answer to to "why" just a decent argument if I suddenly made unemployment higher, more of the job postings you see would have higher skill requirements. My pet theory on "why" is that filtering out the noise gets really hard when tons of people start applying, so they up the bar.