Here's the best way to understand what Gabriel Zucman is up to on the California billionaire tax, and why it's a problem: Suppose you were tasked with assembling a complex historical time series of an important economic indicator, but also suppose the data needed to build it is of widely varying quality. There are gaps in some years when components you need simply weren't measured. In other years the stats you do have are not as reliable. Or maybe the source changed the way it recorded things, or changed the underlying definition. And on top of all that, your indicator contains components that are simply hard to measure with any precision. So what do you do? You'll inevitably have to make some discretionary calls on how to handle the challenges you encounter. Zucman's inequality stats all face this problem. They draw on data that are often incomplete, and go back almost a century in some cases. Reliability varies, as do underlying sampling techniques that generated them. There are also gaps from things that aren't easily measured where you have to make a defensible judgement call on how to handle it. And each and every one of those calls affects your end results. A rigorous researcher's judgement calls would follow certain defensible principles or rules that were clearly articulated and replicable. The effects of those calls would also distribute in different directions. Some would increase the final output number, others would decrease it. In Zucman's work, we typically find the opposite. He makes judgement calls out of necessity, but they almost always go in the same direction, which is to show (a) rising inequality and (b) declining tax burdens on the rich. They always skew in a nonrandom direction, and that direction conveniently aligns with the political story he's telling about inequality and wealth taxes. Or think of it this way: if you're assembling a century-long measure of income concentration it's quite plausible you may have to make 50 different judgement calls on how to handle gaps and uncertainties in the source data. If 30 of those cause your results to show higher inequality in cumulative, but 20 of those reduce inequality in cumulative, that suggests a fairly even distribution is at play. In Zucman's work (as well as his coauthors Piketty and Saez) though, you'll find situations where 50 out of 50 discretionary calls all happen to cause the headline measure of inequality to increase. That's not random - it means that whenever they must make a choice on how to handle data uncertainty, they choose the path that confirms the political story they wish to tell. And this is why the various Piketty-Saez-Zucman measures of inequality are *always* on the high side compared to alternative measures of the same thing. Overlay that with the fact that all three of them are partisan political activists who advocate for wealth taxes, and you see the problem. Their empirical academic work ends up skewing in favor of their political objective, even though it's masquerading as a data-driven exercise. And that's why their empirical claims should be checked against other sources and heavily discounted when they do not align - which is frequently the case.

🧵In 2019, California billionaire tax crusader Gabriel Zucman made an astonishing claim to the NY Times: billionaires pay a lower avg tax rate than poor people in the US. I noticed something fishy though: Zucman's NYT chart did not match his own published data from a 2018 paper