Single PSU per strata
I've been looking at the time 1 youth completion data. When I attempt to adjust for the design using SVYSET i'm warned that there are numerous instances of strata containing only a single PSU. I assume this is due to the youth completion data representing a subsample of the overall dataset? I understand that STATA 10 has further svyset options for dealing with singleton PSUs, but I do not have access to this version. I was wondering about ignoring strata when using svyset (e.g., assuming all psu belong to the same stata) but i'm uncertain about the degree of bias this would incur. Advice around this or other approaches to managing this singleton PSU issue would be appreciated. Thank you for your time.
#1 Updated by Olena Kaminska about 8 years ago
Thank you for your question. Indeed, stratification used in the design was very detailed in order to improve precision. This does result in the issues you describe. There are a few alternatives for you:
1. ignore stratification. This will not introduce bias (stratification influences only standard errors), and will make your estimates more conservative - confidence interval will be higher than it is by design.
2. You can create strata variable yourself such that there are at least 2 clusters per strata. Note, that because strata variable has ordered values, you can combine the adjacent values.
Hope this helps,