ld-archer
commented
2 years ago Taken from the weighting faq's document:
6. Which weight should I use for my analysis?
There are a number of weights reflecting the complex structure of the data. The weight name has the following structure: w_xxxyyzz_aa. To select a weight please answer the following questions:
- _aa part: Is your analysis longitudinal or cross-sectional?
- Longitudinal _lw
- Cross-sectional _xw
- w_ part:
- if your analysis is cross-sectional – which wave you are using? e.g. wave 8: h_
- if your analysis is longitudinal – which is the last wave in your analysis? e.g. you are looking at wave 1-9: i_
Wave: 1 2 3 4 5 6 7 8 9 10 11 Prefix: A b c d e f g h i j k
- xxxyy part: Is your analysis household level or individual level?
- If it is household level: _hhden
- If it is individual level see below
- xxxyy part: Is your analysis for all persons aged 0+, for youth (10-15) or for
adults (16+)?
- 0+ population: _psnen
- Youth (10-15): _ythsc
- Adults (16+): see below
- xxxyy part: you are studying adults aged 16+. Where does your data come from?
- Just one survey instrument (e.g. individual questionnaire): use the weight indicated on the appropriate row of the table below
- A combination of instruments: use the weights from the lowest level in the table below.
For example, if you are using information from the household grid and self- completion questionnaire, the levels are respectively 5 and 2 with 2 being lower – hence the weight will be for self-completion data (_indsc ). Similarly, if you are combining information from household grid, adult main interview and nurse visit, your lowest level is 2 so the weight will be _indns.
There will be situations when you combine information from different instruments at the same level: an example would be adult self-completion interview and nurse visit. In this situation we do not have an optimal weight for you and you could use either a suboptimal weight (see Q14) or you can create a weight adjustment tailored to your analysis (see Q15).
- zz_ part: what is the timeline of your research?
- Starting at wave 6 (2014-15) onwards: ui_
- Starting between wave 2 (2010-11) and wave 5 (2013-14): ub_
- Starting at wave 1 (2009-10): us_
- Starting at any point between 2001 and 2008: 01_
- Starting at any point between 1991 and 2000: 91_
Translated
Because we are combining data from
We have 2 variables from the self-completion sample at present (scghqi: depression_change, and sclonely: loneliness). These are not key variables but will most likely be used, so for a first attempt I'm using the self-completion weight variable from the table in question 5.
After going through the weighting faqs doc I think we need to use the cross-sectional weights, as we are only ever estimating models with a single wave of data (somebody correct me if that's wrong!). Also using the longitudinal weight would reduce our effective sample down to 30-50% of the original size (including 36% of people with at least 1 wave of positive weight), whereas the cross-sectional ranges from 70-80% (including 66% of people).
The weight variable is indscub_xw.
Using the Weight
Need to account for the weights in all the transition estimation functions.
- [x] income_ols
- [x] SF-12_ols
- [x] labour_nnet
- [x] housing_clm
Making this an issue in and of itself just in case we need to generate our own analysis weights. See weighting faqs for more information, or a video on weighting guidance.