In 2000, one in six households filled out the LONG form of the census.
Does IPuMS-USA have information about WHICH one-in-six of households filled out the LONG form? That variability must be controlled for under aggregation.
The unweighted standard deviation is a biased measure of error margin and will be an inconsistent estimate of confidence if the variability that went into the process of choosing the households is not accounted for.
Are there any measures of sampling error in IPuMS-USA samples?
The weights might SCALE the error, but do not determine its densities.