Codex has taken the pain out of finishing a data-heavy book. Here’s a paragraph about it. Being able to talk to your book and data files feels Star Trek-like. Link to my agent instructions below. ...

The prison population has fallen from its peak. But not nearly fast enough—and it rose in 2022 and 2023. At the 2023 level, every additional year means another 1.25 million years under prison authority. Other wealthy democracies imprison far fewer. We can too. 3/4

I just added up the number of years Americans have spent in prison 1979-2021: 48.5 million. In the figure below, each square represents a million years spent in prison, an unfathomable number. 1/4

To understand the mechanism, we conducted a large survey of internal white non-Hispanic. We find evidence consistent with the culture-of-honor explanation, a set of defensive strategies that kept people safe in dangerous places but put them at risk in safer ones. 10/

We also find these homicide persistence patterns despite migrants being better educated and higher income than the people they leave behind and the people they live nearby. And we find it within migrant group education and income categoreis. 7/

With fixed effects, we find homicide persistence among every group, including married women and the elderly. E.g. 70-year-olds from Kentucky are still at higher risk of homicide in say Illinois than Massachusetts born in 70-year-olds also living in Illinois. 6/

Since we have individual-level death certificate data for millions, we examine whether this homicide persistence holds up with precise fixed effects to rule out alternative explanations, e.g., only comparing migrants within county and within age groups, sex, and marital status. Here are the Ns. 5/

We find a similar pattern for 1979-1992 and 2000-2017. Here is the 2017 data. 4/

Bivariate findings. Left: people who stayed put. Right: migrants wherever they ended up. Migrants from historically violent states remained at much higher risk even after moving. Kentuckians stand out. They remained at higher risk wherever they went (mainly the safer Midwest). 3/

We measure state historical homicide rates, our independent variable, in the 1930s. Kentucky, the Deep South, Arizona, New Mexico, and Nevada stand out. Since we have state of birth on death certificates starting in '59, we can look at persistence of these high rates in later periods. 2/

New paper in PNAS: Your homicide risk follows you around the country. If you’re born in the historically violent Wild West, Appalachian Highlands, or Deep South, a higher risk of violent death trails you wherever you go. 🧵1/14

The follow-the-leader tendency is largest for WHO membership. Among Trumpers who supported membership in Aug. 2020, and who never learned Trump’s stance on it, support remained near 100% after the election. By contrast, among those who learned Trump’s stance, support fell to 45%.

Did those voters defect when they learned? Since so few learned Trump’s stance, we lack the power to shed much light on this question, and it’s a hard causal one. Nevertheless, we don’t see much sign of defection among those who realized they were out of line with Trump.

Despite billions spent on the campaign, Trumpers mostly didn’t learn that Trump was out of line with them on COVID. On mandatory masks in public, for example, the % saying Biden supported this more than Trump increased from 19% to only 40%.

Not only were Trump voters out of line with Trump on COVID, they also mostly didn’t realize they were out of line. Among Trump voters who wanted mandatory masks in public, only 19% thought Biden was more supportive of this policy than Trump.

Trump’s pollsters repeatedly warned him he was out of line with his own voters. The RNC chair and some of his advisors pleaded with him to change his stance. Trump ignored them.

In the 2020 campaign, at the height of the Covid-19 pandemic, Trump took a stance out of line, not only with the public in general, but even with his own voters. It’s been forgotten, but even Trumpers were scared and wanted masks and aggressive gov’t action in 2020.

New paper. What happens when a presidential candidate steps way out of line, not only with the public, but even with his own voters? And on a highly salient issue? Candidates typically don’t do that—they’re too strategic. But Trump isn’t. He’s good for social science. 🧵

Here is the same period for the 1984 election, morning in America.

An underappreciated reason why Harris lost was flat real disposable income growth in the election year. RDI was above trend, but voters care about growth in the election year itself. Between Jan. and Nov. 2024, nothing happened.

Been working on the unicorn problem for years only to read it’s been solved by GPT-4. AGI must be close.

An example of the ever present tendency of elites to extract from non-elites. Workers are being forced to pay thousands of dollars to quit. www.nytimes.com/2023/11/20/m...

Political Science should beat other fields to this obvious next step in improving transparency and reproducibility. Pre-register or justify! From a nice Data Colada post: datacolada.org/115

GPT with vision is like having an 85+ percentile tour guide accompanying you. Check out how it makes sense of the cultural references in a Día de Muertos event poster.