| Haz weighted mean [message #6991] |
Sat, 08 August 2015 07:16  |
Espoir
Messages: 3
Registered: August 2015
Location: FRANCE
|
Member |
|
|
Hi, I need help in calculating Haz (hw5) weighted (v005) mean at the cluster level. I have tried it but the result is not different from the unweighted one. Is it the right sample weight to use in this context? Why there is no difference between the two means?
Development Economist
|
|
|
|
| Re: Haz weighted mean [message #6992 is a reply to message #6991] |
Sat, 08 August 2015 13:16  |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
The weighted and un-weighted results are the same because all of the observations within a cluster have the same weight.
Also: the DHS is not designed to be representative at the cluster level and estimates at that level are likely to be very noisy and, in some cases, almost totally uninformative due to small sample size (low power, high variance of estimates).
|
|
|
|
| Re: Haz weighted mean [message #7009 is a reply to message #6992] |
Mon, 10 August 2015 11:55 |
Espoir
Messages: 3
Registered: August 2015
Location: FRANCE
|
Member |
|
|
Thanks for your answer! If so, how to manage when we want to use GPS dataset in order to map some summary statistics (mean, median, min, max) on nutrition and anthropometry, given that GPS data are collected at the cluster level? Do you think that multiplying sample weight by the number of households within the cluster and using it as additional map symbols is a solution?
Development Economist
|
|
|
|
| Re: Haz weighted mean [message #7012 is a reply to message #7009] |
Mon, 10 August 2015 14:00 |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
I think you are confusing two types of weights here. The DHS weights are essentially "probability" weights that re-weight the sample to be representative. In the mapping you describe, I think you want "frequency" weights - you want the marker/circle size to be bigger if there are more observations, yes?
In that case, you want to weight by the number of observations, so all you have to do is count them up. You can do this with a "collapse" command if that is how you are calculating cluster-level averages (see the Stata help file for collapse), or you can use an "egen" command such as:
egen N_obs = count(HAZ), by(clusterid)
Is that what you wanted to do? If you wanted to show the "relative weight" that each cluster was getting as your marker size, you could do about what you described:
even map_weight = sum(old_weight), by(clusterid)
These should be very similar (as in, the ratio of the sum of weights between two clusters will be fairly close to the ratio of the number of observations).
|
|
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
