Project

General

Profile

Support #1666

Clarification of weights when joining datasets and comparing COVID and main survey waves

Added by William Shufflebottom 7 months ago. Updated 4 months ago.

Status:
Resolved
Priority:
Normal
Category:
Data management
Start date:
03/07/2022
% Done:

100%


Description

Hi,

My name is William and I’m working on a social capital publication. I’m a little unsure of which weights we should be using in the “svydesign” in RStudio.
We are interested in the impacts of COVID-19 lockdowns on measures of social capital and are looking at variables in the COVID-19 wave 8 datasets primarily. However, we are also doing some comparisons between COVID-19 wave 8 and the Main Survey waves 6, and 10 where variables are the same. We have also left-joined domains by PIDP to our COVID-19 wave 8 from the xsample and Main Survey wave 11 to include missing domains.
I’ve looked at the guidance on which weights to use and I’m still a little unsure. We are comparing COVID-19 wave 8 (with joined wave 11 domains) to Main Survey Wave 6 and 10. So, we have the weights:
• Wave 6 - indinub_lw, and indinui_xw (no ui version for the longitudinal lw)
• Wave10 - indinui_lw, and indinui_xw
• COVID-19 Wave 8 - betaindin_xw, and betaindin_lw
We are only producing the mean, standard errors, confidence intervals, coefficient of variation. I think it’s just the xw weights we need but I’d like to clarify which weights we should be using for our analysis if that’s alright?

Kind regards

Will

#1

Updated by Understanding Society User Support Team 7 months ago

  • Category set to Data management
  • Status changed from New to Feedback
  • Assignee set to Understanding Society User Support Team
  • % Done changed from 0 to 50
  • Private changed from Yes to No

Hi William,

Are you analysing the data cross-sectionally, that is, for example, a mean from Wave 8 Covid-19 survey with a mean from the main survey Wave 6, or longitudinally (in other words, analysing within person change), for instance, analysing whether an employed person became unemployed or a married person divorced. As a general rule, for a cross-sectional analysis you need to use _xw weights, for a longitudinal analysis _lw weights. Please let us know if this answers your question.

Best wishes,
Understanding Society User Support Team

#2

Updated by William Shufflebottom 7 months ago

Hi,

Thanks for your quick response. We are comparing means so that is the xw then. Thanks. For our variables where we are making comparisons between the means for Wave 8 of the COVID study and with waves 6, and 10 of the Main survey, am I right in saying then that we should use the 'betaindin_xw' weight for running our estimates for the COVID-19 Wave 8 variable list and indinui_xw for waves 6 and 10 of the main survey (as the guidance states that we should use _ui post wave 6)? Thanks again for your clarifying this.

Best

Will

#3

Updated by William Shufflebottom 7 months ago

update: or, do we use the xw for indicators with no comparison and then have to make our own weights for comparisons between the main survey and a covid survey?

#4

Updated by Understanding Society User Support Team 7 months ago

Hi William,

If you are analysing the main survey data separately from the Covid-19 survey data, so first generating the means and them comparing them, then yes, you should use the indinui_xw for the main survey and the betaindin_xw for the Covid-19 survey. Using ui_xw weights is recommended because this way you will use the whole sample available, so including the IEMBS subsample, other _xw weights available exclude this subsample, in other words, their value for the respondents from the IEMBS subsample equals 0.

Best wishes,
Understanding Society User Support Team

#5

Updated by Understanding Society User Support Team 4 months ago

  • Status changed from Feedback to Resolved
  • % Done changed from 50 to 100

Also available in: Atom PDF