| Identifying individuals living in poverty by applying poverty ratio to wealth index score [message #31451] |
Wed, 01 October 2025 14:02 |
Yiqun Luan
Messages: 4
Registered: February 2021
|
Member |
|
|
Hello,
I am working on an analysis to estimate how many household members live in poverty in a given country-year. My procedure involves three steps:
Step 1: From the PR file, I keep the de jure population and rank individuals' wealth index scores, applying household sampling weights. The code is:
xtile wealth_rank_weight=hv271[aweight=hv012*hv005/1000000], nquantiles(total number of de jure population)
Question: Since I am working with the PR file, should I apply the weight as hv012*hv005/1000000 or just hv005/1000000?
Step 2: I obtain the country's poverty headcount ratio from the World Bank for the survey year and use this ratio as the percentile to identify the value of wealth_rank_weight at that percentile, which serves as the poverty cut-off:
centile wealth_rank_weight, centile(ratio)
gen cutpov_weight = `r(c_1)'
Step 3: I classify household members with wealth_rank_weight <= cutpov_weight as living in poverty:
gen pov_weight=0
replace pov_weight=1 if wealth_rank_weight<=cutpov_weight
Experiment: I also tried skipping Step 1, and directly applying the poverty ratio to hv271 in Step 2, then proceeding to Step 3:
centile hv271, centile(ratio)
gen cutpov_weight = `r(c_1)'
gen pov_weight=0
replace pov_weight=1 if hv271<=cutpov_weight
Interestingly, the resulting pov_weight is the same as when I follow all three steps. Could you help me understand why this sameness occurs? Does this suggest that Step 1 can be skipped in the analysis? This would be very helpful, since running xtile with such a large number of de jure individuals in the nquantiles option is computationally intensive.
Thank you very much for your guidance!
|
|
|
|
Members
Search
Help
Register
Login
Home
