| Poolled logistic regression [message #29861] |
Wed, 14 August 2024 09:37  |
gwasswa
Messages: 4
Registered: March 2024
|
Member |
|
|
Hello there,
Could anyone guide me on how to do this?
I want to estimate a pooled logistic regression of three surveys appended to each other but for one country. Through reading on the existing chats i realise there could be an issue with how to weight the pooled data set. Could anyone advise on how to normalise the weights? One of the chats suggest this as a possible solution. Another question is, since there are three surveys then i would expect to include two dummy variables for two of the three survey years. Is there anything else i have to do to ensure accurate results? I will appreciate your help.
Gabriel
|
|
|
|
| Re: Poolled logistic regression [message #29870 is a reply to message #29861] |
Thu, 15 August 2024 10:02  |
Bridgette-DHS
Messages: 3201
Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS staff member, Tom Pullum:
Yes, you need a variable such as survey=1, 2, 3, which I easily constructed during the appending process. In the analysis you can treat "survey" as a categorical predictor or convert it to dummies.
We recommend that you leave the weights as they are. You only need to adjust the weights if you want to construct a pooled estimate of something, for example the median age at first marriage, and I really don't believe such an pooled estimate has any value. The main reason for appending surveys from the same country is to see whether there has been significant CHANGE between surveys, and for that purpose you should leave the weights alone. If you do that, the analysis will weight each survey in proportion to its sample size, and for efficient statistical estimation that's what you want.
|
|
|
|
|
|
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
