[this is a link post to the preprint Persistence - a Critical Review, by Jaime Sevilla]
EDIT: For a more informal and less dry piece read my takeaways here.
IN SHORT: I review, replicate and extend the analysis from seven econometric papers studying how events that happened to and values held by our ancestors affect their descendants several generations afterwards (intergenerational persistence). I argue that together the papers provide moderate evidence of the existence of long term causal effects mediated by parentage.
KEYWORDS: persistence, cultural persistence, economic history, multiple hypothesis testing, post design power analysis, spatial autocorrelation bias, causality, natural experiments, instrumental variables.
Intergenerational persistence is an important topic for Effective Altruism, because it can help us understand how our actions today can affect many generations after. I undertook this research to help us shed light on whether cultural interventions (like increasing the degree at which present people value truth and cooperation) can be an effective way of affecting the long-term future.
The papers I review are:
- The long term effects of Africa’s slave trades (Nunn, 2008)
- The slave trade and the origins of mistrust in Africa (Nunn & Wantchekon, 2011)
- On the Origins of Gender Roles: Women and the Plough (Alesina et al., 2013)
- The Church, intensive kinship, and global psychological variation (Schulz et al., 2019)
- Persecution perpetuated: The Medieval Origins of Anti-Semitic Nazi Violence (Voigtländer & Voth, 2012)
- Trade, Institutions, and Ethnic Tolerance: Evidence from South Asia (Jha, 2013)
- Long-term persistence (Guiso et al., 2016)
- I discuss a gold standard for cultural persistence studies, covering how to (1) identify robust long term correlations via regression studies under different sets of controls, (2) identify causal effects via natural experiments and (3) identify whether culture is a significant mediator via children-of-immigrant studies. More
- I find that many of the papers manage to find statistically significant results. A naive aggregation of the estimated correlation effect sizes suggests that future correlational studies might find effects of around β ≈ 0.28 (0.13) standard deviations per standard deviation of exposure variation. That is, future studies in similar topics should expect to find that one standard deviation of variation on an event would predict ~28% of variation in long term outcomes. However it is hard to rule out spurious correlations due to issues such as spatial autocorrelation or outliers. More
- Some of the papers attempt to study causation via natural experiments. While a couple of such papers arguably succeed in identifying a causal effect, we cannot discard that subsequent robustness checks will cast doubt on the results. A naive aggregation of the estimated correlation effect sizes suggests that future causal studies might find effects of around β ≈ 0.11 (0.02) standard deviations per standard deviation of exposure variation. That is, future studies in similar topics should expect to find that one standard deviation of difference on an event would cause ~11% of variation in long term outcomes. More
- I find that children-of-immigrant analysis suggests the possibility of long term persistence of variation mediated by parentage. The authors of the papers tend to explain this persistence in terms of cultural variation, relying mostly on historical accounts as evidence. More
- Whether long term persistence of variation usually stays constant, wanes or increases with time is an open question. Studying better these dynamics of persistence would be critical to understand the very long-term impact of cultural interventions today. More
This work was funded by the Forethought Foundation, and helped inform the decision of whether to rely on results about cultural persistence in William MacAskill’s forthcoming book.
Thanks to David Rhys Bernard, Nathan Nunn, Aron Valinder, Luisa Rodriguez, William MacAskill, Florian Penner, Ehud Reiter, Faatima Osman, Pablo Villalobos, Ronja Lutz and Neel Nanda for feedback and constructive criticism.
Leticia García provided invaluable research assistance, including carefully reviewing all the analyses conducted in the paper. All remaining mistakes are solely my fault.
Pablo Villalobos, Ronja Lutz and Neel Nanda assisted me in acquiring and cleaning the data I needed for the analyses, for which I am incredibly grateful.
David Roodman and Julian Jamison provided a blind peer review of this paper.
Nuño Sempere provided comments on this post.
I welcome comments and feedback on the piece. Read the rest of the piece.