Project

General

Profile

Support #1890

Extracting PSU and Individual-Level Weights for a Multilevel Model

Added by Laurence Rowley-Abel 11 months ago. Updated 4 months ago.

Status:
In Progress
Priority:
Normal
Category:
Weights
Start date:
04/11/2023
% Done:

60%


Description

Dear Understanding Society team,
I am running a multilevel model using individuals nested within census areas (LSOAs) in Waves 9, 10, 11 and 12. To account for clustering I am using the following levels in my multilevel model: individuals at the first level, PSUs at the second level and LSOAs at the third level. Therefore, from the provided weights, I need to extract separate weights for individuals and for PSUs. Having read your response here [[https://iserredex.essex.ac.uk/support/issues/1572]], I am wondering if the below would be the correct approach:

- For the individual level, I would divide l_psnenus_xw by a_psnenus_xd (from the l_indall.dta and the a_indall.dta files respectively)
- For the PSU level, I would use a_psnenus_xd (from the a_indall.dta file)
- For the LSOA level, I would not be able to calculate a weight as it is not part of the sampling design. I would set this weight to 1 for all respondents.

Would this be correct? Additionally, would this mean I could only include respondents who were included at Wave 1, since I need to use the design weight (a_psnenus_xd) from Wave 1?

Many thanks for your help.

Best wishes,
Laurence

#1

Updated by Olena Kaminska 11 months ago

Laurence,

Thank you for your question. Yes, your consideration of the weights at each level is correct.
You can include all respondents in our datafile. If you use only OSMs (e.g. if you use longitudinal weights), there is a ub design weight you could use: psnenub_xd. Find it in xwavedat.dta from the previous (not current) release in xwavedat.dta. For OSM newborns in the current wave give mother's design weight for them (if not using newborns, you don't need to do anything).

If you are using xw weights, you could weight-share design weights to TSMs. This needs to be done in each wave separately, and is easy to do: you just average the sum of weights for all OSMs in a household to everyone OSMs and TSMs. If a household doesn't have a TSM, you don't change the values. You don't have to do this - cross-sectional analysis using only OSMs is still correct.

Hope this helps,
Olena

#2

Updated by Laurence Rowley-Abel 11 months ago

Dear Olena,
Thank you very much for your reply. How do I access previous releases of the data?

Additionally, I tried to do this using the a_psnenus_xd design weight, but a_psnenus_xd is not the same for all individuals within the same PSU, meaning that Stata will not run the model (as it requires the PSU-level weight to be the same for all observations within a PSU). Do you know why some individuals have a different value for a_psnenus_xd than others in the same PSU and how I could fix this (eg: by exlcuding certain individuals)?

Many thanks.

Best wishes,
Laurencew

#3

Updated by Olena Kaminska 11 months ago

Laurence,

Thank you for coming back to me. Please ignore my earlier recommendations. You actually don't need design weights for each individual, but a weight for each PSU. There are multiple reasons why different people would have different design weights within one PSU, including different selection probabilities for different ethnic minorities in EMB boost, and IEMB boost, and selection of multiple households / dwellings where there are more than 3 per address. You don't need to know the details though for your purposes. I suggest that as a PSU weight use the lowest individual design weight in that PSU. Share this value to all people from the same PSU, including TSMs. Wave 11 release will have ub design weight.
To get an individual weight, divide their weight by the above PSU weight.

Hope this helps,
Olena

#4

Updated by Laurence Rowley-Abel 11 months ago

Dear Olena,
Thanks for getting back to me so quickly. Just one further question - for the individual weight, you stated that I should divide their weight by the PSU weight. Should the numerator be the normal weight one would use for an analysis (eg: l_indinui_lw) or the enumeration weight (eg: l_psnenus_lw)?

Many thanks.

Best wishes,
Laurence

#5

Updated by Olena Kaminska 11 months ago

Laurence,

The individual weight should be the appropriate weight for your analysis.

Best of luck,
Olena

#6

Updated by Understanding Society User Support Team 11 months ago

  • Status changed from New to Feedback
  • % Done changed from 0 to 80
  • Private changed from Yes to No
#7

Updated by Laurence Rowley-Abel 10 months ago

Dear Olena,
Above you stated that "Wave 11 release will have ub design weight", however I am unable to find this weight. Would you be able to let me know what the name of this weight is and in which file I can find it?

Many thanks again for your help!

Best wishes,
Laurence

#8

Updated by Olena Kaminska 10 months ago

Laurence,

Look in xwavedat.dta. But please for this look in wave 11 release (which is not the most recent). You should be able to find it on UKDA website. If you have problems let me know.

Best,
Olena

#9

Updated by Laurence Rowley-Abel 10 months ago

Dear Olena,
Thanks for your reply. I wrote to UKDA about accessing previous releases of Understanding Society, as they are not publically available it seems. UKDA have replied asking for the name of the specific variable I require from the Wave 11 release (I think so that they can send me it individually). Since I don't have the file I am guessing what the design weight in the Wave 11 release would be called - would it be "k_psnenub_ld"?

Many thanks.

Best,
Laurence

#10

Updated by Olena Kaminska 10 months ago

  • Assignee changed from Olena Kaminska to Understanding Society User Support Team
#11

Updated by Understanding Society User Support Team 10 months ago

Hello Laurence,

The variable you are looking for is psnenub_xd. You can locate it in the xwavedat file.

I hope this information helps you.

Best wishes,
Roberto Cavazos
Understanding Society User Support Team

#12

Updated by Olena Kaminska 10 months ago

Laurence,

It is possible for you to create psnenub_xd yourself. But first, could you tell me if you are using any children, or only adults in your analysis?

Thank you,
Olena

#13

Updated by Laurence Rowley-Abel 10 months ago

Hi Olena,
That's good to know - thanks! I am just using indresp, so adults.

Thank you so much for your on-going help! It is greatly appreciated!

Best,
Laurence

#14

Updated by Laurence Rowley-Abel 7 months ago

Hi Olena,
I just wanted to follow up on the above, about creating a psnenub_xd weight.

Many thanks.

Best wishes,
Laurence Rowley-Abel

#15

Updated by Understanding Society User Support Team 4 months ago

  • Status changed from Feedback to In Progress
  • Assignee changed from Understanding Society User Support Team to Olena Kaminska
  • % Done changed from 80 to 60

Also available in: Atom PDF