3/4
Each linked dataset uses different sources and approaches for record linkage.

Consider source material used for linking, methodological biases, accuracy, and representativity to determine which linked dataset is appropriate for your specific research question.

2/4
Married women experienced life cycle of kin beyond the household. Younger married women lived closest to siblings (own or husband) & husband’s parents. By age 60, shifts to living near children and their own siblings. Kin propinquity rate ⬆️ ~60% when accounting for both familial lineages.

Found this out in my office. My three year old has good taste in books.

Agreement rates between sample data (more person hours verifying the data) and the full count are generally acceptable, but there are areas for improvement, specifically occupation. Differences tend to be minor (e.g. one year of difference in age) but worth considering for your research design.

While errors occurred during microfilming, our impression is these were minimal and random. Transcription is a larger issue. Some detail was lost and some transcriptions are just garbage. You can tell because for occupation, 15% of the uncoded strings represent 2% of the people between 1910-1930.

Census Bureau altered data when clerks edited responses/ amended instructions based on real & perceived data quality issues. This is most apparent with the 1910 overcount of women’s paid labor. 



Comparing published counts w/ microdata a good data quality check but numbers will not match exactly.

The stages of data collection/processing can be broadly grouped into 6 stages.



The first errors occurred more than 100 years ago as respondent error or enumerator error. Basically someone provided the wrong information or the enumerator recorded/interpreted the information incorrectly.