* Example of individual-level and cluster-level analysis with the same variables
* Kenya 2022 DHS survey

use "C:\Users\26216\ICF\Analysis - Shared Resources\Data\DHSdata\KEIR81FL.DTA" , clear

* construct a binary outcome variable for 4+ children
gen     nch4plus=0
replace nch4plus=1 if v201>=4

* construct dummies for wealth quintiles
xi, noomit i.v190
rename _I* *

* Individual-level analysis
svyset v001 [pweight=v005], strata(v023) singleunit(centered)
svy: glm nch4plus v190_* , family(binomial) link(logit) eform


* Cluster-level analysis; first switch to clusters as units
gen cases=1
collapse (first) v005 v023 (sum) nch4plus cases (mean) v190_*, by(v001)

* Calculate to observed proportions with outcome=1 in each cluster
* Weights not needed because all cases in the same cluster have the same weight

gen nch4plus_prop=nch4plus/cases

svyset [pweight=v005], strata(v023) singleunit(centered)

* Run with eform to get odds ratios
svy: glm nch4plus v190_* , family(binomial cases) link(logit) eform

* Run without eform to get fitted frequencies and fitted proportions
svy: glm nch4plus v190_* , family(binomial cases) link(logit)

predict nch4plus_hat
gen nch4plus_prop_hat=nch4plus_hat/cases

scatter nch4plus_prop_hat nch4plus_prop, graphregion(color(white)) xlabel(0 .25 .5 .75 1) ylabel(0 .25 .5 .75 1)

* Note: there are clusters in which all households are in the bottom wealth quintile
*   or all households are in the top wealth quintile