Understanding Society User Support

Dear Frida,

UKHLS weights make any subgroup representative, including Northern Ireland. So, just use the weights provided. If you are interested in the 2001+ analysis for example you will need the W_indall01_lw weight.

Best wishes,
Olena

Dear Olena,

Thank you for your help. Unfortunately I am still struggling to figure out what weights to use, with the harmonised BHPS/UKHLS from 2001 to 2017 for Northern Ireland, as the indin01_lw is only available from 2010 (ukhls w2) and onwards, and wlrwght from BHPS wave 12, not wave 11. What are the appropriate weights to use for the harmonised BHPS 11 -18 and UKHLS 2 -8? Is it w_lrwtuk1 for BHPS 12-18 and then w_indin01_lw from ukhls 2-8? What weight should I then use for BHPS w11?

Thank you in advance for your help.

All the best,
Frida

Frida,

Can you tell me which type of analysis you are using? If you are looking at people longitudinally you will need W_indall01_lw from the last wave in your analysis. But if you are pooling the data from different years (and don't need a person to be present in all waves) you could use cross-sectional weights for each year.

Olena

Hi Olena,

I am sorry, slightly confused as can’t find the w_indall01_lw variable in any of the waves. However, I do find w_indin01_lw, described as the longitudinal main interview weight. Is this the one I should use, or where do I found w_indall01_lw?

Frida

Frida,

Yes sorry, w_indin01_lw is good if you are using information from the main questionnaire.

Olena

Hi Olena,

I am still slightly confused whether the w_indin01_lw weights are the correct weights for analysing the UKHLS/BHPS longitudinally for Northern Ireland only. Examining the variable indicates that of the 2550 observations in UKHLS wave 8, 1839 of them equal 0. Moreover, the mean is 0.04877 and the third quartile 0.08262. Is this a problem when I am restricting the sample to Northern Ireland only?

In addition, if I only use the weights in the last wave, does this not mean that all the respondents that have left the survey between 2001 and 2017 are no longer included? Is there a way to use weights, that will allow you to include all the respondents available at an individual level between 2001 and 2017?

Thank you for all the time and help!

All the best,
Frida

Frida,

Yes, this is correct. Since 2001 we had 3 boosts: GPS+EMBOOST in 2007-2008, and IEMB in 2013-2014. For these people longitudinal information is not available since 2001. So, it depends on your analysis. If you are looking at people longitudinally since 2001 you can't use the above boosts and you should use '01' weight. If you are pooling the data, and are not looking at people over time, but looking at events, then you can use cross-sectional weights like 'us', 'ub' and 'ui'.

Hope this helps,
Olena

Hi Olena,

Thank you for this. However, I am still wondering whether I can use the weights to represent Northern Ireland as an individual country within the UK. The weights appear to be designed for the UK as a whole, adjusting for the higher sampling probability in Northern Ireland. For my analysis, I would only want to adjust for unequal selection probabilities, differential nonresponse, and potential sampling error of the Northern Irish population, not the UK population.

Best,
Frida

Frida,

Yes, I can assure you that you can represent Northern Ireland as an individual country using our weights.
The weights are not designed to represent specific countries only - they are designed to represent any subpopulation. From weighting perspective Northern Ireland is one subgroup of the population that the sample represents. Using weights you can also represent females in Northern Ireland, elderly people in Northern Ireland, or any other substantive subgroup.

Hope this helps,
Olena