What are weights correcting for? [message #22137] |
Fri, 05 February 2021 18:44 ![Go to next message Go to next message](/theme/default/images/down.png) |
sylvan
Messages: 11 Registered: August 2020
|
Member |
|
|
Hello,
I am working on a project that is using weights provided in the individual-level surveys. In reading the documentation, our understanding is that the individual weights essentially upweight individuals interviewed in strata where strata, household, and individual response rates are low.
We have observed (and this is mentioned as well in the documentation) that response rates seem to differ for different groups such as age and education. While these weights adjust for average response rates of all people in a given strata, it does not appear to us that they adjust differentially for people of different age or education levels. Is our understanding correct?
Thank you for the clarification and help.
Regards,
Sylvan
|
|
|
Re: What are weights correcting for? [message #22171 is a reply to message #22137] |
Tue, 09 February 2021 08:08 ![Go to previous message Go to previous message](/theme/default/images/up.png) ![Go to next message Go to next message](/theme/default/images/down.png) |
Bridgette-DHS
Messages: 3151 Registered: February 2013
|
Senior Member |
|
|
Following is a response from DHS Research & Data Analysis Director, Tom Pullum:
The main purpose of the weights is to compensate for stratum-level over-sampling or under-sampling. The goal is to have enough cases in each stratum to be able to produce relatively stable estimates. If a stratum has a relatively small population, then we need to have a higher sampling fraction in that stratum in order to get an adequate sample size. But then for the purpose of national-level estimates those cases need to be weighted down. As you say, the weights also adjust for non-response, but that is usually secondary in magnitude. The largest adjustments for nonresponse are usually made for the DV and HIV weights (dv005 and hiv05).
We do not do post-stratification weighting on the basis of age, education, etc. to force the weighted distribution in the sample to match the known distribution in the population. However, as a user you could choose to make such adjustments and re-weight.
I'll add that in surveys restricted to ever-married women (EMW surveys) we provide multipliers to the weights called "all-women factors". These factors are specific to single year of age and to covariates in the household file, such as region, place of residence, level of education, and wealth quintile. Those factors adjust for ever-married status, not for non-response, but are similar to what you are thinking of in terms of how they are calculated and what they accomplish.
|
|
|
|