Conley and Kelly almost killed an entire literature. But the Spanish Inquisition survived!

I first heard about this paper over dinner with a grading committee recently. Somewhere between the main course and the coffee, the conversation got really interesting when someone brought up Conley and Kelly’s The Standard Errors of Persistence (Journal of International Economics, 2025). I had somehow missed it, and I’m not sure how.

The literature in question is the “deep roots” genre: studies showing that some feature of a place centuries ago - a medieval massacre, the reach of the slave trade, an ancient institution - predicts something today, like trust or violence or income.

Reading papers like that, I rarely look at the t-stats. I probably should. Apparently, they are often spectacular: t-statistics so large the results look bulletproof. That, Conley and Kelly argue, is exactly what should worry us.

The villain is spatial autocorrelation. Simply put: things close together tend to be alike. Neighboring regions have similar weather, similar history, similar wealth, and similar omitted variables (ie the stuff that the model leaves out).

A simple but cruel method

Conley and Kelly prove it with a wonderfully cruel trick. They run regressions where both variables are random noise — meaningless data, with only the feature that nearby points resemble each other. The t-statistics come back enormous.

Then they test real published papers: replace the historical cause with noise. The fit often improves! They replace the modern outcome with noise, and the historical variable still “explains” it.

Honestly, I wonder how so many studies made it through seminars, conferences and anonymous reviews.

What papers can we no longer trust?

The published version runs diagnostics on 30 persistence studies from leading journals and reproduces one “leading” regression from each.

Ranked roughly by citation weight, the most influential papers that we should question are (I used AI to very briefly summarize the long-run causal effect that each paper claims to have discovered):

• Acemoglu, Johnson & Robinson (2001), “The Colonial Origins of Comparative Development,” AER — settler mortality → institutions → modern income. One of the most cited papers in all of economics (and already severely harmed by David Albouy)!

• La Porta, Lopez-de-Silanes, Shleifer & Vishny — “Legal Origins” — common- vs civil-law roots of modern financial and legal outcomes. Founding paper of an enormous literature.

• Acemoglu, Johnson & Robinson (2002), “Reversal of Fortune,” QJE — places rich in 1500 are poor now, via institutions.

• Nunn (2008), “The Long-Term Effects of Africa’s Slave Trades,” QJE — historical slave exports → present-day African underdevelopment.

• Nunn & Wantchekon (2011), “The Slave Trade and the Origins of Mistrust in Africa,” AER — descendants of the heavily raided are less trusting today.

• Alesina, Giuliano & Nunn (2013), “On the Origin of Gender Roles: Women and the Plough,” QJE — plough agriculture → persistent unequal gender norms.

• Voigtländer & Voth (2012), “Persecution Perpetuated,” QJE — towns with medieval pogroms saw more anti-Semitic violence in the Nazi era. (The “medieval pogroms → Nazi zealotry” study.)

• Ashraf & Galor (2013), “The Out of Africa Hypothesis... and Comparative Economic Development,” AER — genetic diversity and prosperity. Already controversial on other grounds.

• Becker & Woessmann (2009), “Was Weber Wrong?,” QJE — Protestantism → literacy → prosperity in Prussia.

• Dell (2010), “The Persistent Effects of Peru’s Mining Mita,” Econometrica — colonial forced-labour boundaries still visible in household consumption today.

And now, the million dollar question: What studies survive?

It isn’t all demolition. They also show how to recover more credible results: soak up the spatial trend and correct the standard errors. Only one study clears every hurdle:

• Drelichman, Vidal-Robert & Voth (2021), the Spanish Inquisition → long-run development (PNAS). It’s significant on the treatment test (BCH p = 0.03), passes the placebo, and is the single study that rejects the synthetic-outcome null at 5%. This is the cleanest survivor in the paper.

Some survive the headline treatment test at 5% but not the toughest synthetic-outcome test:

• Michalopoulos & Xue (2021), “Folklore” (QJE) — punishment-of-tricksters in folktales → trust. BCH/HC p = 0.03.

The marginal three (7–9%), neither clearly alive nor clearly dead:

• Voigtländer & Voth (2012), “Persecution Perpetuated” — medieval pogroms → Nazi-era violence (p ≈ 0.07).

• Giuliano & Nunn (2021), cultural persistence — climatic instability → importance of tradition (p ≈ 0.07).

• Nunn (2008), the slave trade → African underdevelopment — hangs on at roughly p ≈ 0.09 under HC inference.