How are we to interpret unexpected income values – for example, in the 2022 one-year ACS there i $760,000 in earnings attributed to a young school teacher . Are these likely errors, is there some sort of value swapping that goes on for privacy purposes or is there something else at play?
I can confirm that this record has the same earnings, age, and occupation as in the file released by the Census Bureau. The ACS does not swap values between respondents and instead applies top coding for high earners; our user note describes the top coding procedure and provides the threshold values for each sample. For Massachusetts (the state this record is located in) in 2022, the threshold wage and salary earnings for top coding is $474,000. Anyone who reported annual earnings above this threshold is assigned the median earnings of the group exceeding the threshold, which in this case is $760,000. What this means is that this individual reported earning at least $474,000. Note that INCEARN is the sum of INCWAGE (wage and salary income) and INCBUS00 (business and farm income). For this individual, the entire value for INCEARN is coming from INCWAGE.
The ACS is a self-administered questionnaire, so it is certainly possible for a question to be misunderstood resulting in earnings being misreported. IPUMS does not have a recommendation of how to handle such cases. Depending on your research question, you might consider using data from the Current Population Survey (IPUMS CPS) or the Occupational Employment and Wage Statistics (OEWS) program as an external benchmark.
Thanks!.. I would have expected maybe the ACS would have some validity checks on answers? Like, if this respondent intended to report 50,000 and it was interpreted as 500,000 and triggered the top code value of 760,000, there’s no algorithm to identify this is as a likely outlier based on the other characteristics of the respondent? My understanding is that if someone says they were born on Planet Venus, there is a mechanism for dealing with that as an invalid response…