Ispoke briefly on the phone to one of your IESR colleagues about getting some information about weights in the BHPS. I was then directed to this page.
The issue I am having, is that having inspected the longitudinal weights (wLRWGHT), I am a little confused over how to use it properly!
So I notice in volume A of the documentation, it states the means of these weights as having a value of 1.00 – except for wave R, where the individual respondent longitudinal weight has a value of 2. When I tried to replicate these means, I found that the weights had a mean value of ~0.85 at wave B, down to ~0.35 at wave R – however when excluding values with a 0 weight, the mean for the weights aligned themselves to the value stated in Volume A – So, 1.00.
The documentation also states that “The longitudinal respondent weights (wLRWGHT) selects out cases who gave a full interview at all waves in the BHPS files.” And then “At each wave these cases are re-weighted to take account of previous wave respondents lost through refusal at the current wave or through some other form of sample attrition.”
The issue the weight values pose is that when analysing later waves, e.g. merging wave Q to wave R (and only merging those who have responded in both waves), and doing some longitudinal analysis, because the weight at wave R is close to 0.35, we only get roughly 35% of the sample as usable. This seemed strange to me and hence wanted to know why this was the case.
I mean I understand if one is engaging in longitudinal analysis from Wave A to Wave R, where you indeed might expect 35% of original respondents at Wave A, to still be on the survey at Wave R. However when doing two wave analysis and merging just those who have stayed on in both waves, I don’t see why my sample is still being restricted.
Hope my confusion is clear enough, but don’t hesitate to contact me for clarification.
#1 Updated by Stephanie Auty over 2 years ago
- Status changed from New to In Progress
- % Done changed from 0 to 10
- Private changed from Yes to No
Many thanks for your enquiry. The Understanding Society team is looking into it and we will get back to you as soon as we can.
Stephanie Auty - Understanding Society User Support Officer
#2 Updated by Peter Lynn over 2 years ago
- % Done changed from 10 to 50
Your question is clear. The BHPS longitudinal weights are indeed designed for analysis of sample members who have participated in /all/ waves up to and including the current one. So, this means that rlrwght is only defined for people who have participated in all 18 waves. Your analysis group (people who participated in waves 17 and 18) is of course potentially (much) larger than this - particularly because the new samples that entered at waves 9 and 11 are excluded from this weight.
A more appropriate weight for your analysis would be rlrwtuk1. It looks to me as though this is non-zero for around 88% of the OSMs who participated in waves 17 and 18 (compared to 43% for rlrwght).
An alternative would be to use the wave 18 cross-sectional weight (rxrwtuk1). This is non-zero for over 98% of your potential analysis cases, but it does not correct for the fact that not all wave 18 respondents also responded at wave 17, so in principle your estimates would be slightly more biased if you use this weight (but would most likely have smaller variance).