| Re: Convert DHS (SPSS?) missing value codes to Stata codes in Stata dataset [message #12197 is a reply to message #12193] |
Fri, 07 April 2017 10:26  |
Bridgette-DHS
Messages: 3167
Registered: February 2013
|
Senior Member |
|
|
Another response from Senior DHS Stata Specialist, Tom Pullum:
There is a large literature on imputation, by statisticians such as Roderick J.A. Little and Donald Rubin. "Multiple imputation" is perhaps the best approach (in Stata the relevant command is "mi impute"). You can use Heckman selection models ("heckman" in Stata) to counteract bias. I don't know the procedures in SAS. I personally would not consider either approach unless the level of missing was well over 10%, say, and I thought there was a systematic difference between the people with a response and the people without a response. You could construct a variable "missing" that is coded "." for NA, 0 if not missing, 1 if missing, and do logit regressions on the usual covariates, such as wealth, age, education, place of residence, etc., to see if there was a significant and systematic difference. (Basically a "selection model".) DHS cannot provide further guidance on the topic. You may need to talk with a statistician who knows the topic.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
