Hi. I’m creating a count of children by single year of age in Vermont in families with incomes under $175,000. Given Vermont’s small sample size, I was inclined to use the 5-year ACS summary data set, but is there a problem with this approach as some children may be counted as two different ages in different years? Should I instead go with the 1-year data, even though it has a larger margin of error?
The ACS data are cross-sectional, meaning a new sample is fielded each year. The same individual being included across adjacent years of the survey is incredibly unlikely; the ACS is fielded to 1% of the population and the same household would not be sampled in back-to-back years, though if the individual moved they might possibly be included again if their new household happened to be sampled.
Using the 5-year file is akin to pooling five years of data to increase your sample size – you will increase the precision of your estimate at the loss of some precision in the time period. Note that the 5-year files have adjusted the weights to account for the pooling of these data (i.e., weights have been divided by 5), but you will need to do this yourself if you manually pool the multiple 1-year ACS files.
You reference the summary file data, but have posted in the IPUMS USA topic, which disseminates person-level microdata. I assume that you are using the person-level microdata to create a custom table. If you are indeed interested in using the ACS geographically aggregated summary file data tables (i.e., the IPUMS NHGIS data), your options will be somewhat limited by the age and the income categories that the Census Bureau chooses to release in its published estimates (e.g., single year age groups are not available).
Thank you for this, Kari! And yes, that was a misnomer on my part re: the summary file data. I was referencing the 1-year and 5-year person-level microdata.
1 Like