| Re: Analyse multiniveau [message #25122 is a reply to message #24983] |
Thu, 01 September 2022 16:26  |
Janet-DHS
Messages: 893
Registered: April 2022
|
Senior Member |
|
|
Following is a response from DHS Research & Data Analysis Director, Tom Pullum:
Your question is about the interpretation of multi-level models, rather than about DHS data. Such questions are outside the scope of the forum. However, in general, when the variances are statistically significant, the interpretation is that the differences between clusters are systematic rather than random. The usual strategy is to add cluster-level covariates until the variances lose statistical significance. These covariates can be aggregations of the individual-level data for the cluster, for example the proportion of the households in the cluster that are in the bottom two quartiles or the proportion of women who have no schooling. Alternatively, cluster-level covariates can be drawn from the files of spatial covariates that are linked to the cluster ID codes. For Nigeria, for example, the files NGGE7BFL.dta and NGGC7BFL.dta contain geographically structured data. The goal is to explain (statistically) the differences between clusters with cluster-level covariates.
In the future please post in English per forum instructions.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
