The Getting Started With the Public Use File document discusses this.
Such discrepancies can occur for many reasons. The summary tabulations are weighted using different weighting variables, depending on the survey year and whether or not the variable was asked of the full sample or half the sample. If your tabulations use the incorrect weighting variable, there may be small discrepancies. In addition, some PUF variables are subject to topcoding or other disclosure avoidance techniques to protect respondent confidentiality. The published Summary Tables are based on the internal use file (IUF), which does not have these protections. If your tabulations depend on geographic variables, you may find differences as well. Because the AHS is longitudinal, the PUF's geographic variables are "frozen" as of the date when the sample was drawn. The published Summary Tables use more current geographic definitions.
To find summary table variables that do not match the PUF, see the Table Specifications and PUF Estimates for User Verification. This product is organized to correspond exactly to each AHS summary table that is currently available in the AHS Table Creator. Each row provides the SAS code logic needed to replicate most summary table estimates, plus the Table Creator estimate for the U.S., the PUF estimate for the U.S., and a column noting the reason for any difference between the two. To locate the table specifications, go to the Data section of the Census AHS website, choose the year of survey you want, click on the link to the PUF, and then choose the National PUF or the Metropolitan PUF.
For more information about topcoding and other confidentiality measures, see Disclosure Avoidance Techniques: 2015 to 2023.