| Any way to use GSEM with survey data? [message #11798] |
Fri, 10 February 2017 06:50  |
nwegbus
Messages: 15
Registered: December 2015
|
Member |
|
|
Hello,
I'm using Stata/SE 13.1 and working with DHS Nigeria survey data.
I have one outcome variable which is binary (lifetime experience of intimate partner violence: Yes/No). I have 20 predictor variables (14 binary, 5 categorical and 1 latent variable)
I'm trying to prove that 2 of my predictor variables 'marital status' (.i.e. ever married versus never married) and 'age at first marriage' (10-14years, 15-17 years, 18-21 years, 22-46years) have the greatest effects on the outcome - 'lifetime experience of intimate partner violence' (Yes/No).
The Stata manual states that survey data is only supported by SEM. However, apart from this detail, everything else I need for my analysis is supported better by GSEM.
My understanding is that not using survey weights would bias my variance estimates but should not be that much of a problem for my point estimates.
Given what I'm trying to prove here, am I wrong in thinking that biased variance estimates should not be too much of an issue in this instance?
If I am wrong, and using GSEM would lead me to make erroneous conclusions, then is there any way at all to get around this?
Thanks in advance.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
