Munir Squires

This paper introduces "marital distance," a network measure quantifying social integration using genealogical records of 50 million Americans (1750-1950). Marriages with short network distances reflect unions within close-knit social circles, while longer distances bridge separate social spheres, integrating society. This distance increased sharply during this period, with the steepest rise during the Second Industrial Revolution (1870-1915). Throughout, people married far closer in the network than random matching would predict. Using county-level analysis and a shift-share instrument, we establish that manufacturing employment led to greater social integration. Sibling fixed effects show that brothers marrying more distant spouses were likelier to migrate, leave farming, and enter skilled occupations. These findings demonstrate how economic modernization transformed American society by fostering greater social integration.

Kinship and Marriage: A Model of Social Network Formation

Why do societies vary dramatically in kinship intensity and marriage restrictions? I develop a model where individuals choose between marrying distant partners (gaining new connections) or close relatives (strengthening existing ties). Endogamy creates dense networks where ostracism is particularly costly because individuals lack outside connections, and this vulnerability enables higher cooperation through collective punishment. The model explains how economic development weakens kinship groups by providing alternatives to family-based cooperation, accounting for both the persistence of strong kinship systems in some contexts and the historical transition in others toward more individualistic social structures.

Improving Historical Census Transcriptions: A Machine Learning Approach (with Christian M Dahl, Sam Il Myoung Hwang, and Torben Johansen)

Historical census records allow researchers to track individuals over time, but linking records across censuses is difficult for minority and immigrant populations due to transcription errors in handwritten names. Our machine learning approach improves name transcription in historical U.S. census records. Human transcribers disagree on 30% of names, especially for foreign-born individuals. Our machine transcriptions increase linking rates by 147% for contested records while improving match quality by 38%. These improvements expand sample sizes for traditionally under-linked groups—foreign-born, non-white residents, and those without formal education—where additional linked records are valuable for research. Validation against genealogical records confirms these improvements represent genuine accuracy gains rather than spurious matches.

Long-Term Mortality Impacts of Temperature Shocks: Evidence from Historical and Genealogical Data (with Patrick Baylis)