Strange poverty gap in China at $5.50 in the '80s

[danielmahler]

Hi all, I'm seeing some strange patterns with China's poverty gap in the 1980s at $5.50. Strange meaning that it differs quite a bit between PovcalNet and PIP even though the headcount rates are identical. Here is some code to understand what I find.

// Prepare PovcalNet data
povcalnet, country(CHN) year(all) povline(5.5) fillgaps clear
decode coveragetype, gen(reporting_level)
replace reporting_level = lower(reporting_level)
replace reporting_level = "national" if reporting_level=="national (aggregate)"
keep year reporting_level headcount povgap
order year report
rename headcount hc_pcn
rename povgap pg_pcn
tempfile pcn
save    `pcn'

// Prepare PIP data
pip, country(CHN) year(all) povline(5.5) fillgaps clear
keep reporting_year reporting_level poverty_gap headcount
rename reporting_year year
rename headcount hc_pip
rename poverty_gap pg_pip

// Merge and calculate differences
merge 1:1 year reporting_level using `pcn', nogen
gen dif_hc = abs(hc_pip-hc_pcn)
gen dif_pg = abs(pg_pip-pg_pcn)

// Plot
format dif* %3.2f
twoway line pg_pcn year || line pg_pip year, by(reporting_level)

[Tefera19]

Hi @randrescastaneda , @danielmahler,

I looked at this issue and I noticed that the 1982 and 1983 (rural and national) data points are not similar for PovCalNet and PIP. Would it be possible to check the pip data source for those two years?

Thanks,
Tefera

[danielmahler]

Thanks for looking into this Tefera!! That's super weird. We don't have data for China for 1982 and 1983 - we are interpolating data between 1981 and 1984. Is the 1981 data for China similar between PovcalNet and PIP?

[Tefera19]

Hi @randrescastaneda , @danielmahler,

The 1981 data points are similar.

[danielmahler]

Thanks Tefera. I think the issue is with the 1981 rural data. It has a mean consumption of $1, but a poverty gap of 1. Those are inconsistent -- a poverty gap of 1 means that everyone in society has 0$. Perhaps the formula for how the poverty_gap and poverty_severity is calculated for group data is off? Best, Daniel

[randrescastaneda]

Hi @danielmahler,

I can't replicate the results you mention. Could you please send me the code where the poverty gap is 1? The table below is what I got.

Thanks.

[danielmahler]

Sorry about that, here you go :)
pip, country(CHN) year(1981) povline(3.65) clear

[randrescastaneda]

Yes, you're right. I replicated the issue. I think the problem is with the fitting of the lorenz at the tail of the distribution. Perhaps it changes from one model to the other (I don't really have an easy way to check that). It is clear that there is a discontinuity around $3.3.

Code to replicate

library(data.table)
library(ggplot2)

pl <- seq(.5, 3.7, length.out = 100)

chn <- purrr::map_df(pl,
                     ~{
                        x <- pipr::get_stats("CHN", 1981, povline = .x)
                        setDT(x)
                        x[, .
                          (reporting_level,
                            mean,
                            headcount,
                            poverty_gap,
                            poverty_severity, 
                            poverty_line)]
                     }
                     )


chnm <- melt(chn, 
             id.vars = c("reporting_level", "poverty_line"), 
             variable.name = "measure", 
             value.name = "value")



ggplot(chnm[reporting_level == "national" & measure != "mean"], 
       aes(x = poverty_line, 
           y = value, 
           color = measure) ) +
  geom_line()