index
Briefing note on using the 2004 Health Survey for England weights
Background
There are two files in the 2004 HSE dataset:
(1) The General Population (GP)
(2) Ethnic Boost file
GP refers to the whole population of England, regardless of ethnic group. The GP file is representative of the general population of England so it contains ‘white’ individuals and minority ethnic individuals in the GP sample. Those in the GP file who were classified as belonging to one of the target minority groups were given the same questionnaire as the ethnic boost sample.
The Ethnic Boost file includes all those from the ethnic boost sample PLUS the ethnic minorities from the GP sample.
The following weights are in the files:
(1) GP file:
- Household level: wt_hhld : adjusts for non-contact and refusal of households. It corrects the distribution of household members to match population estimates for sex/age groups and GOR. These weights are generated using calibration weighting with household selection weights to correct for the limit of 3 households at addresses.
- Individual level, adults: wt_int : a combination of the household weight and a component which adjusts for additional non-response among individuals within households
- Individual level, children: child_wt: a combination of the household weight and a selection weight for only including a maximum of 2 kids in a household. The combined weight is then adjusted to ensure the weighted age/sex distribution matches that of all children in co-operating households.
(2) Ethnic boost file:
- Individual level: wt_int : adjusts for probability of selecting an address in the minority ethnic sample (addresses within different ethnic profiles had different chances of being selected). Then combined with selection weights for households within addresses and for individuals within households
- Nurse visit: wt_nurse : accounts for drop out at this stage
- Blood sample: wt_blood : accounts for drop out at this stage
Combining the files – how to use the weights
When combining the GP file and the Ethnic Boost files for analysis purposes you should delete the duplicate cases from the GP file. The following stata syntax can be used to combine the files and drop the duplicate ethnic minority cases from the GP file:
| *drop ethnic minority cases on the gp file
use "<insert name of GP file>"
drop if (dmethn04 >=1 & dmethn04<=7)
*append the ethnic boost file to the gp file
append using "<insert name of ethnic boost file>" |
You should then check that the combined file is in the correct proportion (e.g. the white group reflect the correct proportion of the population - based on the mid-year pop estimates for 2003). We have done this and the proportions are fine so no further adjustments are needed.
As advised by NatCen you should then create a new weighting variable (e.g. wt_intnew) which divides the combined weights by the mean weight (6.67) so that the weighted total equals the actual total (i.e. so that your output shows the correct population size rather than the weighted population size). 6.67 is the mean of your combined sample 2004 weights.
If using the nurse data you should use the variable wt_nurse as this overrides the individual weight wt-int. Similarly if using the blood data you should use the variable wt_blood as this overrides the variable wt_nurse.
The weights for children and households are different within the general population and ethnic boost files. The GP file has 3 separate weights for adults, children and households but the Ethnic Boost file only has one weight (wt_int) which combines the adult, children and household weights. So when doing analyses based purely on children or on households which weights should you use? NatCen advise to use the variable wt_int within both files when using purely child data. However the household data is more problematic – the household selection weight is very complex because it depends on the profile of household members, if you are using this weight you should contact NatCen for further advice.
In summary you should use your new weighting variable (e.g. wt_intnew) for all individual analyses (adults and children) but use wt_nurse if using nurse visit data or wt_blood if using the blood data.
Do I need to use the weights i.e. do the weights make a difference?
We ran the following crosstabs on the adults in the combined file without the weights and then with the weights applied to see if the weights make a difference to the final result. We examined the following crosstabulations separately for men and women
- BMI by ethnic group
- Portions of fruit and vegetables by ethnic group
- Physical activity by ethnic group
The conclusion was that the weights do make a difference to the final results (some are larger differences than others). For example the percentage of Black-Caribbean women classified as obese is 28% when unweighted but rises to 30% when weighted. The weighted crosstabulations we ran replicate the weighted results (and unweighted bases) shown in the published HSE report.
