Explore further

Errata and Notes

• Some inconsistencies between text/PDF/exhibits
  There are a few places where the numbers in the text don't exactly match the numbers in the figures/tables or the PDF version. This is an inevitable thing that happens when papers are in progress and underlying datasets are being updated and corrected. There are also currently small discrepancies between the digital and PDF version for similar reasons. These should all be reconciled by the time the paper is published.

• Some conflicting occupations in the raw data explorer
  In the raw data explorer, there are some cases where the occupation described by Gemini is different from what we used in the paper. We checked every discrepancy but have not yet had time to annotate all of them here. Reasons for discrepancies are fathers with multiple occupations, ambiguous or conflicting historical data, or different treatment of children with absent or multiple fathers.

• Running replications in Claude/ChatGPT web windows
  We were surprised that the web-based ChatGPT and Claude don't seem to be willing to download our 750-row CSV and run analysis on it. For now, you can either download the CSV dataset and paste it into the web window, or point an agent on your computer at this site.