# Chapter 15 Drivers of Poverty Reduction: Decomposing changes in poverty

Using poverty and socio-economic indicators observed over two time periods, the Oaxaca-Blinder decomposition method (Blinder 1973; Oaxaca 1973) decomposes the poverty decline into endowment accumulation and return to endowment. Using this approach, changes in poverty rates are decomposed into three parts; [1] endowment accumulation, like increases in the education level of household heads, [2] return to endowments like education, employment or employment in high return sector, and [3] the interatction between endowment and its return. This methodology allows to decompose poverty across the full distribution of consumption rather than simply between the poor and the non-poor households.
Table 15.1: Decomposing changes in poverty into endowments and returns by components
variable BGD BTN PAK
Year2 21.63 5.99 19.21
Year1 28.49 16.95 29.47
Difference -6.86 -10.96 -10.26
Endowments -1.66 -2.51 -0.33
Coefficients -5.62 -10.04 -10.17
Variable PAK (2011 vs 2015) BTN (2007 vs 2017) BGD (2010 vs 2016)
Year2 21.63 5.99 19.21
Year1 28.49 16.95 29.47
Difference -6.86 -10.96 -10.26
Endowments -1.66 -2.51 -0.33
Coefficients -5.62 -10.04 -10.17

Table 15.1 reports poverty rates in both periods, the difference between them, and estimates for endowment and returns from endowment for Pakistan, Bhutan, and Bangladesh. All countries have experienced a large decline in poverty rate, which ranges from 6.9 to 11 percentage points. When the changes are decomposed into three components, effects from endowment accumulation and returns from endowment drove the poverty reduction in the region.

Poverty reduction in each country was partially driven by the change in households head endowment such as gender, literacy, employment status and sector. The estimates in table 15.1 of -0.33, -2.51, and -1.66 for Pakistan, Bhutan, and Bangladesh, respectively, indicate that changes in literacy, gender, labor force status, and sector employment contributed to the decrease in poverty rates.

Estimates for coefficients demonstrate that an increase in endowment returns largely impacted on the poverty decline. In Pakistan, for example, the change in endowment triggered poverty reduction by 0.33 percentage point whereas return to endowments lowered the poverty rate by 10.17 percentage points, 5 times higher than endowment effect.

Table 15.2: Decomposing changes in poverty into endowments and returns by components
variable PAK BTN BGD
Endowment
Urban -0.33 -0.59 -0.11
Literacy 0.04 -0.70 -1.40
Male -0.01 -0.48 0.03
Employed -0.02 0.35 -0.04
Unemployed -0.05 0.03 0.12
OLF 0.00 0.10 -0.02
Agriculture -0.08 -0.89 -0.11
Industry 0.12 0.08 0.13
Service 0.00 -0.03 0.00
Other -0.02 -0.05 -0.01
Interaction 0.25 1.60 0.42
Coefficient
Urban 1.91 2.17 1.22
Literacy 1.64 1.84 3.74
Male -5.45 -4.92 -1.69
Employed -0.28 17.04 0.06
Unemployed 0.01 -0.01 -0.01
OLF -0.06 -0.76 -1.04
Agriculture 2.23 -4.58 1.41
Industry -0.74 -0.23 -1.01
Service -0.42 -0.01 -0.63
Other -0.34 -0.26 -0.06

The Oaxaca-Blinder decomposition breaks down the decline in the poverty rates points at more disaggregated level, identifying whether the decrease was driven by the changes in literacy level, gender, labor force status or sector of work of a household head. The sum of the effects does not add up to the changes in poverty rates since residual effects and some categories with missing observations are not included in the table. Following sections summarize the findings for each country.

## 15.1 Pakistan

Migration to urban area was the largest contributor to the decline in poverty rate in terms of endowment effects. Compared to the effects of the changes in the share of male, employed heads, and agriculture employment, 2 percentage point increase in urban population in the country lowered its poverty rate by -0.33 percentage point.

The estimates for returns to endowment show that being a male household head has large advantages in avoiding poverty. Through male headed households, the poverty rate declined by 5.45 percentage points, at least 8 times higher than other effects including employment status, industry and service employment. Industry and service sector workers also experienced poverty reduction whereas agriculture did not help households move out of poverty.

## 15.2 Bhutan

When the effect of endowment accumulation is considered, increases in urban population, literacy rate, share of male heads and a decrease in agriculture employment led to poverty reduction. Although the period gap is bigger due to a data availability issue, compared to Pakistan and Bangladesh, the country had gone through a large structural change which lowered agriculture employment and the population with illiteracy.

The coefficient estimates show that male headed households are more likely to escape poverty and agriculture sector has higher returns to endowment than other sectors do. Being employed, however, does not necessarily lead to poverty reduction, and other indicators including urban and literacy turns out to be minimal in lowering poverty rate.

In terms of the effects from endowment accumulation, the increasing share of household heads that can write or read and decreasing share of heads engaged in agriculture sector triggered a decrease in the poverty rate. Decreasing illiteracy rate among household heads led to a declined in poverty rate by 1.4 percentage point. As household heads depart from agriculture sector, moving to manufacturing or service industry, the effects from endowment in agriculture lowered poverty rate by 0.11 percentage point.

Although the increase in the share of male household head didn’t affect poverty reduction, being a female head raises a chance of staying poor by 3.73 percentage points. This finding implies gender discrimination in Bangladesh in poverty reduction. Also, employment in agriculture does not affect poverty reduction but the effects from returns to endowment in manufacturing or service industry indicate advantages of leaving agriculture sector.

## 15.4 Stata code example

You may access our full Stata do-file by accessing the following link. Our work consists of running the following command for each dataset and saving the results in order to export to Tableau.

* Example: Decomposing changes in poverty into endowments and returns

* Open dataset
datalibweb, country(country') year(year') type(SARMD) surveyid(surveyid') clear

* Generate explanatory variables

keep if relationharm==1

gen lfs_empl = (lstatus==1)
gen lfs_unem = (lstatus==2)
gen lfs_OLF = (lstatus==3)
gen lfs_miss = (lstatus==.)

gen industry_agri = (industry==1)
gen industry_ining = (industry==2)
gen industry_manu= (industry==3)
gen industry_public = (industry==4)
gen industry_miss = (industry==.)
drop industry
g poor_190=poor_nat*100
drop if poor_nat==.

* Blinder-Oaxaca decomposition

oaxaca poor_190 urban literacy  male lfs* industry* [aw=wgt] , by(year) swap relax  ///
categorical( lfs*, industry*)
`

### References

Blinder, Alan S. 1973. “Wage Discrimination: Reduced Form and Structural Estimates.” The Journal of Human Resources 8 (4): 436–55. https://doi.org/10.2307/144855.

Oaxaca, Ronald. 1973. “Male-Female Wage Differentials in Urban Labor Markets.” International Economic Review 14 (3): 693–709. https://doi.org/10.2307/2525981.