• All households are included in the upper (household level) section, and only "selected" households are included in the lower (person level) section.
• Only households which contain "selected" persons will appear in the person level section. 
• A single household may contribute toward multiple cells in the person section.

The only instance in which we would expect the household counts in the person level section to add up to match the household counts in the household level section is if all of the following 3 criteria are met:

• All households are selected
• All households contain selected people
• There are no person level breakouts in the person section

For "sanity checking", it is sometimes useful to force the above three conditions for a portion of the analysis, e.g. if you are questioning the household counts in an analysis which is based on people over 60, you could dimensionalize based on the presence/absence of a person over 60 at the household level, don't do any dimensionalization at the person level, and don't do any restrictions at either level (except that you could restrict the person level to persons over 60, because the household dimension will guarantee that at least one is present).  For the households that contain a person over 60, the counts should tally.