Home » Data » Weighting data » When weights are not supported
| Re: When weights are not supported [message #3245 is a reply to message #3244] |
Tue, 11 November 2014 20:19   |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
I almost pitched the same weighting work-around idea - just make your dataset sample size proportional to population by expanding the number of observations (and within-survey, you expand relative to the probability weights). I would just add that, if you use the "cluster" command and cluster at a level above the individual, then you should get proper inference because the model will "know" that there is no added variation from the extra observations.*
*Note: a great way to convince yourself of this is to generate some fake data, add an id number, run a regression, get the standard errors, and then replicate each observation 100 times (expand) and "cluster" on the id variable. You'll get back the original standard errors (while OLS on the expanded sample will produce SEs that are far too small). You could also do this using an original DHS dataset.
|
|
|
|
Current Time: Fri Apr 26 00:09:04 Coordinated Universal Time 2024
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
