Project

General

Profile

Support #1362

Deriving own weights

Added by Karen Mak about 2 months ago. Updated about 2 months ago.

Status:
Feedback
Priority:
Normal
Category:
-
Target version:
-
Start date:
06/16/2020
Due date:
06/24/2020
% Done:

90%

Estimated time:

Description

Hope you are well.

My research focuses on the relationship between arts engagement (Wave 2) and wellbeing (Wave 5) using OLS regression. I understand that if I am using more than one wave, a longitudinal weight is more appropriate. But using that would lead to a significant drop in my sample size, therefore I would like to derive my own weight based on the the guidelines stated in "Understanding Society: Weighting and Sample Representation FAQ 2019". I have prepared the weighting codes and I would be extremely grateful if you could let me whether the coding is correct:

gen responseW5=1 if e_hidp!=. & b_hidp!=.
replace responseW5=0 if e_hidp==. & b_hidp!=.

logit responseW5 eventfqW2_v2 marstatW2 child16W2 ageW2
predict p

gen weightW25 = (1/p)*b_indscus_xw

Thank you.

History

#1 Updated by Alita Nandi about 2 months ago

  • Private changed from Yes to No
  • % Done changed from 0 to 10
  • Assignee set to Olena Kaminska
  • Status changed from New to In Progress

Hello,

Thank you for your query. We have assigned this issue to our weighting expert who will get back to you.

Best wishes,
Alita

#2 Updated by Olena Kaminska about 2 months ago

Karen,

Thank you for your question. A few comments:
1) as a base weight you should use a longitudinal weight b_indscus_lw, not cross-sectional weight;
2) please exclude those who died and left the country in a meantime - they should not be considered as nonrespondents;
3) condition your logit model on non-zero b_indscus_lw;

Hope this helps,
Olena

#3 Updated by Karen Mak about 2 months ago

Dear Olena,

Thank you so much for your prompt response. This is really helpful!
May I ask, for point 3, does it mean fitting the model like this: logit responseW25 b_indscus_lw ?

Best wishes,
Karen

#4 Updated by Karen Mak about 2 months ago

I am sorry - I meant a model like this: logit responseW5 ageW2 if b_indscus_lw>0 & b_indscus_lw!=. ?
Would it matter if I included more W2 predictors in the logit model? Are there any specific W2 predictors that need to be included?

With appreciation,
Karen

#5 Updated by Olena Kaminska about 2 months ago

Karen,

Yes, I would recommend more predictors. Choose predictors to be related to both nonresponse and your own model of interest. But I would err on higher number of predictors if you are uncertain. Note, predictors need to be from wave 2 and should not have any missing values for non-zero b_indscus_lw.

Hope this helps,
Olena

#6 Updated by Karen Mak about 2 months ago

Thank you so much for your helo Olena! Hugely grateful.

#7 Updated by Karen Mak about 2 months ago

Karen Mak wrote:

Thank you so much for your help Olena! Hugely grateful.

#8 Updated by Alita Nandi about 2 months ago

  • % Done changed from 10 to 90
  • Status changed from In Progress to Feedback

Also available in: Atom PDF