Measurement Error in U.S. Federal Census, 1850-1930

Sam Hwang, University of British Columbia
Munir Squires, University of British Columbia

We document measurement errors in the U.S. Federal Census between 1850 and 1930, focusing on age and names. These variables are important because they are used to link records across censuses. Using representative samples linked to crowd-sourced data, we show that measurement errors in these variables are non-classical: they are correlated with one’s literacy and occupations. Individuals with measurement errors in names or age are less likely to be linked and more likely to be linked incorrectly, even after taking into account their observable characteristics. Finally, the impact of measurement errors on measures of intergenerational mobility is explored (in progress).

See extended abstract

 Presented in Session 14. Measurement Issues in Historical Data