| Re: accounting clustering effects of women's data when using baby-based analysis [message #799 is a reply to message #798] |
Sun, 29 September 2013 10:21   |
rkinoshita
Messages: 6
Registered: September 2013
Location: Nicaragua
|
Member |
|
|
Hi. Thank you so much for all of useful inputs- they will be fruits for thoughts for my paper when I discuss about limitations and interpretations of the significant (or non-significant) results.
As for the number of clusters, please see below, women by area but limiting to those women who have ever married or in union because they also responded to my two violence variables (main exposure and outcome variables). Number of strata is 17 and PSU is 731. I am guessing that this would probably make less justifiable to use cluster model (which makes more precise estimates of SE than Robust model)- then again, my question is for which model do I need to use now???
I was discussing with my supervisor and we agreed that we will stick to birth-based analysis for associations between intimate partner violence and infant/child outcomes and discuss possible biases arising from repeated data of women with more than one birth.
My supervisor also found that svy + logistic is the same as using just logistic with vce(cluster) as an option, but with added value of weighting too. she suggested me to stick to svy + logistic because in order to pass this thesis and get my degree out, I don't need to learn new methods. Later on, I would need to look at this though, as I am hoping to publish this.
anyway, thanks a lot, again!
Rinko
. svy: tab Area if QW814F==1, obs cell count format(%10.4f)
(running tabulate on estimation sample)
Number of strata = 17 Number of obs = 12065
Number of PSUs = 731 Population size = 16178.63
Design df = 714
-------------------------------------------------
Area de |
Residenci |
a | count proportions obs
----------+--------------------------------------
Urbano | 9419.3941 0.5822 5935.0000
Rural | 6759.2356 0.4178 6130.0000
|
Total | 16178.6297 1.0000 12065.0000
-------------------------------------------------
Key: count = weighted counts
propor~s = cell proportions
obs = number of observations
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
