The selection criteria for men's anemia testing in surveys is often a subset for all households. Thus for example every second or third household is selected for men's anemia testing. In this case, the selection is a random sample of every second or third household in the cluster (randomly selected before fieldwork in the cluster starts), which causes the low overall response.

When we produce couple tables, we always use the man's weight.

You should always use weights, and your models should also make the svyset and svy adjustment for clusters and strata (svyset and svy are described in other postings). In the CR file we recommend that you use mv005. You should get a big difference between using no weights and using mv005. The difference between using mv005 and v005 should be small, but we still recommend using mv005 rather than v005 because nonresponse is usually worse for men than for women and the weights include adjustments for nonresponse.

The IR and MR files are representative of women and men, respectively (subject to the criteria for eligibility, i.e. age and de facto residency status). The CR file is representative of couples but not of all women or all men. You can compare the anemia for women in the IR file vs women in the CR file to get at the selection bias for women. Similarly for men. You can also compare with the anemia values in the PR file.