which weight to use [message #10649] |
Thu, 25 August 2016 09:09 |
claudia-vdb
Messages: 1 Registered: August 2016 Location: Maastricht
|
Member |
|
|
Hi,
I have appended men and women DHS for Uganda and also the AIS survey. I am confused now on which weights to use (I am using STATA).
I read in several forums that for sampling weights I need to divide the variable v005 by 100,000 and to set svyset [pweight = weight], psu(v021) strata(v023). For simple descriptive statistics do I also need to use the pweights as I defined them above (using the command svy: ....)?
Thanks!
Claudia
|
|
|
Re: which weight to use [message #10656 is a reply to message #10649] |
Fri, 26 August 2016 07:42 |
Bridgette-DHS
Messages: 3216 Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS Stata Specialist, Tom Pullum:
DHS weights include a factor of 1000000, not 100000 (that is, 6 zeroes, not 5). If you use iweight, then you definitely need to divide by 1000000. If you use pweight you can do that if you want, but you do not need to, because it will make absolutely no difference with pweights. (Try it both ways and you will see that the results are the same.) Yes, you should use the weight even for descriptive purposes. Otherwise your results will be biased toward the subpopulations that were over-sampled.
Estimates of coefficients, means, proportions, etc. will be affected by the weight adjustment but not by the psu and stratum adjustments. Those two adjustments affect the standard errors and confidence intervals, but not the estimates of coefficients, means, proportions, etc. It is common for people to use the full svyset and svy commands when they really only need to use the "[pweight=v005]" part of svyset. That's ok--but for descriptive statistics all you really need is the weight adjustment.
You may have additional questions about the weights. Feel free to ask again if you cannot find answers on the forum.
|
|
|