Replicate weights to generate confidence intervals in the Current Population Survey

[khimmelstein]

I am attempting to use SAS 9.4 to generate confidence intervals for basic descriptive statistics (means, medians, relative frequencies) in the U.S. Census Bureau's Current Population Survey. I have downloaded the replicate weight files, and have been using the following code, for example, to derive a confidence interval for the variable "nchild."

proc surveymeans median data = temp varmethod = jackknife;

weight wtsupp;
repweights repwt1 -- repwt160;
var nchild;
run;

However, when I do so, I end up with very wide confidence intervals e.g. a percentage CI of 4.8-15.1% in a cell with 840 unweighted observations. Has anybody else used the replicate weights to generate confidence intervals, and have you calculated similarly wide intervals? Or am I misinterpreting how to use the replicate weights? 

Thanks!

[ballardw · Accepted solution]

I suggest the question is why you think that is an exceptionally wide confidence interval. I have seen such things many times with similar and larger samples.

Did you look at the variance on data within each replicate of the data? or within subpopulations?

I have no idea what the "nchild" variable represents but guessing that it may have something to do with children in families I might expect older respondents to have no children so including those families may increase the variability of the responses increasing confidence limits. Single male households are also less likely to have any children.