| MULTILEVEL WEIGHT FOR POISSON [message #31930] |
Mon, 29 June 2026 09:07 |
Khawar
Messages: 22
Registered: April 2023
|
Member |
|
|
I am working on multilevel Poisson. I faced a problem with the svy command while running mepoisson. As I am working on the DHS in Pakistan, the average household size per cluster is also not available. I have calculated this from the following formula: Empirically derived from the 2017 Census: 32,205,111 total households / 168,943 total EBs. The first question is, 'Is it correct?' This table is my coding scheme for 2-level calculations according to DHS report 27.
************************************************************ **********
* Stage A *** Compile parameters/inputs for Level-weights calculations
************************************************************ **********
* 1. a_c_h completed clusters by strata
gen a_c_h=.
quietly levelsof v023, local(lstrata)
quietly foreach ls of local lstrata {
tab v021 if v023==`ls', matrow(T)
scalar stemp=rowsof(T)
replace a_c_h=stemp if v023==`ls'
}
* 2. A_h total number of census clusters by strata
gen A_h = 0
* Punjab
replace A_h = 26958 if v023 == 1 // Punjab Urban
replace A_h = 60048 if v023 == 2 // Punjab Rural
* Sindh
replace A_h = 21916 if v023 == 3 // Sindh Urban
replace A_h = 17223 if v023 == 4 // Sindh Rural
* KPK
replace A_h = 3221 if v023 == 5 // KPK Urban
replace A_h = 18777 if v023 == 6 // KPK Rural
* Balochistan
replace A_h = 1826 if v023 == 7 // Baluchistan Urban
replace A_h = 8386 if v023 == 8 // Baluchistan Rural
* Islamabad
replace A_h = 727 if v023 == 9 // Islamabad Urban
replace A_h = 787 if v023 == 10 // Islamabad Rural
* FATA
replace A_h = 43 if v023 == 11 // FATA Urban
replace A_h = 3763 if v023 == 12 // FATA Rural
*** 3. Average household size per cluster (M_h) ***
* Empirically derived from 2017 Census: 32,205,111 total households / 168,943 total EBs
gen M_h = 190.63
*** 4. Define Population Denominator (M) ***
* Based on Census 2017 total minus AJK/GB households
* (Calculation: 32,205,111 total - 1,004,241 AJK/GB)
gen M = 31200870
*** 5. Total number of completed households (m_c) ***
* verified this is 14,540 from the Household (HR) file
gen m_c = 11869
*** 6. Households selected per cluster (S_h) ***
gen S_h = 28
*** 7 ***************************
gen DHSwt = d005 / 1000000
************************************************************ **********
* Stage B *** Approximate Level-weights ***
************************************************************ **********
* Steps to approximate Level-1 and Level-2 weights from Household or Individual Weights
*Step 1. De-normalize the final weight, using approximated normalization factor
gen d_HH = DHSwt * (M/m_c)
*Step 2. Approximate the Level-2 weight
* f the variation factor
gen f = d_HH / ((A_h/a_c_h) * (M_h/S_h))
scalar alpha = 0.5
gen wt2_05 = (A_h/a_c_h) * (f^alpha)
gen wt1_05 = d_HH / wt2_05
I am working for Alpha 0.5
* Create a constant cluster-level weight
bysort v001: egen wt2_05_constant = mean(wt2_05)
mepoisson children_born i.dv_any_violence i.region v012 i.education ///
i.marriage_age_group i.terminated_preg i.ethnicity_final ///
i.wealth_3cat i.residence i.ever_used_binary ///
lom_prof_community ///
[pweight=wt1_05] || v001:, pweight(wt2_05_constant) irr vce(robust)
My code is correct for multilevel Poisson, as the svy command does not support the mepoisson command. I have also followed the Stata manual for multilevel Poisson modelling to create the above command. Your kind response will be very appreciated.
Regards,
KA
[Updated on: Mon, 29 June 2026 09:25] Report message to a moderator |
|
|
|
Members
Search
Help
Register
Login
Home
