Is there a way to identify households/ persons across the samples in Guinea in 1983 and 1996, i.e. to exploit some panel data potential? As far as I understand the variable “serial” is used as a unique identifier in a given sample, but would be given to another household in the subsequent sample?!
Unfortunately, this is not possible. In each year, the Guinea census contains a random sampling of 10% of the households in the country. As a result, one would expect a minimal amount of overlap to exist, even if households/persons could be linked across samples. You are correct that “serial” is unique within a sample, but not across samples.