Weighting after Merging [message #5599] |
Sun, 14 June 2015 18:44 ![Go to next message Go to next message](/theme/default/images/down.png) |
JCartagena
Messages: 1 Registered: June 2015
|
Member |
|
|
Hi, I would like to merge the male and female files (using the household survey as base). So I can use individuals as unit of analysis. Which weight should I use? the household weight? or the male weight for male individuals and the female weight for female individuals?
Thank you very much in advance
|
|
|
Re: Weighting after Merging [message #6935 is a reply to message #5599] |
Wed, 05 August 2015 10:35 ![Go to previous message Go to previous message](/theme/default/images/up.png) |
Bridgette-DHS
Messages: 3151 Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior Sampling Specialist, Ruilin Ren:
The answer to your question depends on the analysis you are performing. Assuming you pooled male and female surveys data from the same survey together:
1. if your analysis is restricted to females coming from the households selected for male survey (usually male survey is conducted in a sub-sample of households), no matter if there is a male interviewed in the household or not, you can use the male sub-sample household weight HV028;
2. if your analysis is restricted to females coming from the households selected for male survey and with at least one male interviewed in the household, then you can use the male individual weight;
3. if your analysis is not restricted to households selected for male survey, you just simply pooled the male individual and female individual data together, then
a. if the male survey is conducted in all households, then you can use the male and female individual weight without any modifications (not exact but a rough solution, an exact solution is to de-normalize the individual weights as stated in the last paragraph);
b. but if male survey is conducted in a sub-sample, say in a% of the households selected, then you need to multiply the male individual weight by the inverse of that percentage, that's, 1/(a%). This is because the male was under sampled compared to the female survey (not exact but a rough solution, an exact solution is to de-normalize the individual weights as stated in the last paragraph).
If you pooled male and female survey data from different surveys (it is less meaningful), or from the same survey but want an exact solution, then you need to de-normalize both the male and female individual weights for pooled analysis. Please refer to the user forum for de-normalization since this is a frequently asked question.
|
|
|