Analysis including Northern Ireland
Dear support team,
I have a question regarding the different sampling design used in Northern Ireland (NI). I am using nine waves of data from wave 'L' of the BHPS to USoc waves b and c (inclusive).
Given there are no PSUs or Strata in the BHPS for the NI sample, is there a recommended strategy for UK-wide analysis which includes NI? I note that for the Understand Society waves strata and PSUs are present for the NI sample. PSU seem to be set to be constant within household, and strata for the most part are '701' (though some are different for reasons I can't work out). I assume that it is not correct to follow this strategy for previous waves in the BHPS: assigning NI cases a PSU based on the household, and set strata to '701'? If this is incorrect, is a design based approach possible at all, or should I go with a model based approach?
Updated by Peter Lynn over 8 years ago
The appropriate strategy for UK-wide analysis as you describe is to consider the NIHPS to be a single stratum (so, give all persons whose sample origin is NIHPS the same value of stratum; a value not used in the rest of the sample) and to consider each original wave n NIHPS household a separate PSU (so all persons descended from the same household inherit the same PSU indicator.
This is what we have done for the Understanding Society weights for combining samples though, as you have spotted, there was an error in production of these weights. This error should have had only a small effect and has now been corrected. The corrected version of the weight will be on the wave 4 Understanding Society data release, which will be available soon.