John quotes Jason Sorens as writing:
Even when you control for overall state ideology (Democratic states have higher taxes, and really Democratic states have much higher taxes), union density increases tax rates. Increasing union density from 10% of the workforce, as in Nebraska, to 25%, as in Hawaii and New York, increases the tax burden by about one and a quarter percentage points of state personal income.
I’m bothered by the causal language here, the jump from a statistically significant regression coefficient to the claim that “union density increases tax rates.” What’s particularly scary here is that the causal leap is implicit. Sorens doesn’t write: Here’s a correlation, and one possible interpretation is causal. Rather, he reports the regression result as if it is direct evidence of causation.
John does it right. His headline is “Do unionized states have higher taxes?”, not “Does union density increase tax rates?”
You might say I’m being picky here–everyone knows that a regression does not in general directly answer a causal question (especially when you control for an intermediate outcome such as voting patterns). But . . . here’s what Sorens writes: “the evidence suggests that policies discouraging collective bargaining help a state reduce its tax burden in the long run . . .” He’s definitely giving a causal interpretation.
I agree that observational data are relevant to causal questions–much of my career would be a sham if I didn’t believe this–but I also think we’d do best, as social scientists, to clearly distinguish between evidence and speculation. In this case, the data show that states with higher taxes have higher rates of unionized workers. The speculation is that a state policy that changes the rate of unionization would have a certain expected effect on taxation. That’s the part I don’t see the evidence for.
I’m not trying to pick on Jason Sorens here. Gathering data, making graphs, and running regressions of historical data–telling us what’s happened in the past–is a contribution in itself. And then if you want to make some claims from there, go for it–but please separate these claims speculations from the hard data.