Understanding Society User Support

The number of observations for estimations using weights and without are different as the number of cases with 0 weights are not included in the number of observations for weighted estimation. You can check this by excluding cases with zero weights from weighted and unweighted estimations - you will see that the number of observations remain the same.

But in your case, there is an additional issue - the weight variable you have specified, w_indinui_xw, is only available from onwards Wave 6. So, Stata drops all data from Waves 1-5 in weighted estimations using this weight. What you will need to do is create a new variable which will be the cross-sectional weights that are relevant for that wave, and use that weight variable in svyset. For example,
generat weight=indinus_xw if wave==1
replace weight=indinub_xw if wave>=2 & wave<=6
replace weight=indinui_xw if wave>=7 & wave<=13

I ran an example model of job satisfaction for those who are in paid employment using OLS with and without weights, and with svyset:
glo vlist jbsat i.sex_dv c.age_dv##c.age_dv i.ethn_dv i.mastat_dv i.hiqual_dv c.paygu_dv c.jbhrs
regress $vlist
regress $vlist [pw=weight]
svyset psu [pw=weight], strata(strata) singleunit(centered)
svy: $vlist

The number of observations were 237311, 232939, 253933. So, even though the observations did drop the extent was not as big as you found.

Please check and let us know if this resolves your issue.

Best wishes,
Understanding Society User Support Team

Hi Yuliya,

It's hard to say what could have gone wrong (maybe you forgot to delete wave prefixes from the name of these variables?) In any case, you can always link psu and stata to your long file from the xwavedat file.

Best wishes,
Piotr Marzec
UKHLS User Support