Variance estimation using CLUSTER and STRATA

USA
1

Dear IPUMS team,

I am descriptively examining changes in totals and proportion of social work occupations (defined by using social worker and counselor occupation codes) over the last 40 years. I applied the Strata and Cluster codes to adjust variance estimation and the person weight (PERWT) for the point estimates. I use Stata and applied the following code for the survey design: svyset cluster [pweight=perwt], strata(strata)

However, when I run the analysis, such as “svy: total” or “svy:mean,” it fails to estimate the variance (picture below). It says, " Missing standard error because of stratum with single sampling unit." For your information, I used the full dataset, not selecting or deleting certain cases.

image
image902×580 81.6 KB

Is the problem related to the fact that the Strata and Cluster variables are constructed at the household level while my analysis focuses on the person level?

Thank you very much for your help!

Warm regards,
Hyojin

2

There shouldn’t be any single unit strata if you are using the full sample. I have a few suggestions for troubleshooting:

  1. For each sample year, calculate the number of individuals in each value of STRATA. I checked in 1950 and all had more than one observation.

  2. Run a separate -total- calculation for each year.

  3. It is possible that the way you are constructing sw_emp is giving some missing values, which could lead to single-unit strata. For example, maybe you are excluding those who are NIU for OCC1950.

Note you should also be able to get an estimate of standard errors in Stata by setting a different value for the singelunit() option in -svyset-. But I think it is better to figure out why you are getting these single-unit strata in the first place, since these are unexpected.

3

Thank you for your suggestions!