Discrepancy in 2022 PERWTs

Hi IPUMS support–thank you so much for all that you do! I’m so thankful that we can consult experts on using your valuable datasets.

I am using 2022 ACS data for a project in which I am trying to get joint distributions of race, sex, and occupations of people on a county and state basis. I like this dataset because I can construct detailed conditional probabilities (for example, getting the distribution of occupations for white males in Georgia).

Admittedly, I am new to this field of data analysis. In order to check whether I was using the data correctly, I looked into race distributions according to state, and was wondering why there might be certain discrepancies against census data or whether I am doing something totally wrong. For example, I am getting this for this percent distribution in Hawaii.

IPUMS 2022 ACS PERWTs:*
White: 87.0%

Census 2023 1-year estimate ACS:
White: 21.9%

*For context, I am basically summing up all the PERWTs for all residents for a given state. Then, I am summing up all the PERWTs for all residents of a given race in that state, and dividing that by the state’s PERWT sum to derive the percentage.

Please let me know if there are data-collection-related issues, or it is likely that I am calculating these percentages incorrectly. Maybe I should look at an alternative dataset? Thank you so much!

It sounds like you are trying to apply person weights to your analysis using 2022 ACS data. A good place to start would be our blog post that answers some common questions on using weights with IPUMS data. Since you are using IPUMS USA’s ACS data, I would recommend reading our user note on weights and we also provide data training exercises for a variety of stats packages, most of which include exercises on applying weights to your analysis (See IPUMS USA’s Exercise 2 for weight-specific exercises using US data).

Without knowing your specific approach to analysis or how you are analyzing your data (Stata, SPSS, SAS, CSV?), I can only provide some general feedback. In general, what you are doing sounds reasonable to me, so I am not sure why your numbers are off. Using 2022 ACS data with our online statistical analysis tool, I find the distribution of race in Hawaii to be similar to the Census estimate your cited. Note that IPUMS data is the publicly available sample that has undergone editing for privacy and will not always match Census estimates exactly (which are typically made using unedited data). We do expect that they are close, as they are in this case (22.2% white in Hawaii using IPUMS ACS vs. 21.9% in the Census estimate).