I am compiling summary statistics on a sample of mothers (identified using the CHBORN variable) from the South African 2001 Census and 2007 Community Survey. I am unclear how I should apply weights to my variables of interest.
I understand that I should use the WTPER for all individual-level variables (e.g. age, personal income, etc.). Since I am reporting the summary statistics for mothers (not households) what weighting variable should I use for my household-level variables of interest (household income, number of persons in house, ownership of dwelling, single-parent household)? I think it should be the WTHH weight, but I’m a bit confused here.
Also, should the number of observations change when I apply the weights? I added the WTPER to the weight statement in my SAS procedure, but the number of observations is the same as if the weight statement was not there. Does the weighting variable only adjust the mean/standard deviation/frequency?
Which weight you use depends more on the unit of analysis than the particular variable. Since IPUMS places household variables on each person record they can be treated as person-level variables as well.
For instance, if you were wondering how many people live in 5-person households, you would use the variable PERSONS, which is technically a household-level variable, however, since you want to know how many people live in 5-person households you would treat it as a person-level variable and use WTPER. Now, if you wanted to know how many 5-person households there were you would use WTHH (making sure to only select one person per household, since each household member has a copy of the household record. If you do not select only one person per household 5-person households will be “counted” 5 times, once for each person with that household record copied on the end of their person record).
The number of observations should not change. As you noted, weights only effect the values of analyses and not the original data.