| Creating panel - using weights [message #10171] |
Sat, 02 July 2016 13:06  |
Nora
Messages: 14
Registered: May 2016
|
Member |
|
|
Hello,
1-I'd like to collapse the data to create a panel of observations across several years. So in this case, I should collapse across which variable to ensure that my units are constant over the years? Do I need to use the GPS data-set?
2-How do I use the weights in stata on the individual and household level data? Is it only run along with the regression command? If I create summary statistics or graphical figures do I need to specify the weight option? Is there a way to weight the whole data-set like in SPSS?
3-If I am collapsing my data-set, do I use the weight along with the collapse option? If so, would I still need to put it again as an option in the regression command or summary statistics tabulation?
Thanks.
|
|
|
|
| Re: Creating panel - using weights [message #10176 is a reply to message #10171] |
Mon, 04 July 2016 14:17  |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
Hi Nora,
Responding to your points:
1. a "panel" of what? Countries (with the time dimension being survey year)? If this is the case, you want to use the original weights given in the dataset to "collapse" each survey round down into one observation - that is, use the weights to generate the averages that become the values of your panel. After that, in the regression, it is up to you whether you want to weight those by, say, relative country population or something like that. But you would use the DHS weights in the creation of the aggregates for your panel, and then some other weight (or no weight) to adjust for population sizes in the regression (which is run on the weighted means...if I'm understanding you correctly).
2. See above.
3. Yes, use the weight when you collapse.
Happy to follow-up if I'm not understanding your data setup correctly (as in, don't understand what "panel" you are trying to create).
|
|
|
|
|
|
|
|
| Re: Creating panel - using weights [message #10180 is a reply to message #10178] |
Mon, 04 July 2016 17:41 |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
The problem isn't that household's change (that is true in any "repeated cross-section" that gets turned into a "panel" of aggregates), the problem is that the clusters change from year to year. That is, they don't represent the same areas at all. Even the "regions" sometimes change from survey to survey. On top of that, clusters are not large enough to be representative samples, and so even if they were constant across time, you'd have problems.
So basically, at a unit smaller than the "region", it is impossible to build the panel you want. At the "region" level (which as you say has too few units), it MAY be possible, but would depend on the country (whether regions stay constant across time).
I know some people have done some GPS stuff about imputing/estimating narrow-area values in repeated surveys to generate a panel like you describe, but it is not really advised by DHS*, and I think often times they are stretching the data far too thin for their own purposes (at least in a panel context, in a cross-section there are times when this might be OK).
|
|
|
|
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
