Understanding Society User Support

Meg,

(1) For pooled cross-sectional (CS) analysis, you should use the CS weight from each respective wave. Each set of weights is independently scaled to a mean of 1.0, so each wave/year will be equally represented in your analysis if you do this. So, you can use a BHPS CS weight for the BHPS observations and a UKHLS weight for each UKHLS observation. Similarly, you can use the youth CS weight for each observation based on youth questionnaire data and an adult interview weight for each observation based on adult interview data. You will therefore have to generate your own weight variable which takes a value from the relevant original weight variable for each observation.

You will find some other notes on weighting for CS analysis here: https://www.understandingsociety.ac.uk/support/issues/494.

(2) pidp remains constant as sample members age so it should be simple to identify the same respondents in the youth and indresp data files. You would broadly expect 15 year-olds in the youth file at one wave to appear in indresp at the following wave, but you may find a few cases who go from 14-year-old youth to 16-year-old adult or 15-year-old youth to 17 year-old-adult due to having a birthday soon after/before the interview date, or indeed some who are the same age at successive waves, for the same reason. And of course, there is non-response, so not all 15-year-old youth respondents in wave t will participate at wave t+1, etc.

Hope that helps,

Peter

Hi Peter

Thanks for your reply and for the link to the weighting document.
I am a bit confused about the way that the waves are pooled in the pdf you linked to. I was following the UKHLS/BHPS coding that was provided in the course run by the Understanding Society team, file:///C:/Users/User/Downloads/Worksheet_ex_9_Stata_version%20(5).pdf
This uses append, instead of merge, and does not rescale by calendar year. Has this guidance now been updated or does it depend on the analysis I am doing? Do I need to redo my merge to adjust the periods?

Thanks for your help

Meg

Meg,

merge and append should achieve the same thing.

Whether you want to scale depends a bit on what, conceptually, you think you are doing when pooling. If you are trying to represent all occurences of an event over a time period, then each sub-time-period (i.e. wave/year) should be equally represented, hence my suggestion to rescale. Otherwise, your weighted data will be heavily skewed towards the Understanding Society years, when the sample has been much larger than in the BHPS-only years.

Peter