Informational Constraints on Antipoverty Programs: Evidence for Africa

Informational Constraints on Antipoverty Programs:

Evidence for Africa

Martin RavallionGeorgetown University

Presentation at the Economic Research Forum, Cairo, December 2017

New hope for redistributive interventions using transfers in developing countries

• World’s aggregate poverty gap “ …is roughly what Americans spend on lottery tickets every year, and it is about half of what the world spends on foreign aid.” (Annie Lowrie, NYT, Feb 2017). (Based on an estimate of $80 billion for global PG at MERs; at PPP it is $160 billion.)

• The implication drawn is that it should be relatively easy to eliminate global poverty using transfers. Is that right?

9/23/16 Brown, Ravallion & van de Walle 2

Huge expansion in “social safety nets” (SSN) in the developing world

• SSN: Direct non-contributory income transfers targeted to poor families

• In last 15 years many developing countries have introduced new SSN programs.

• Today almost every developing country has at least one SSN program.

• Roughly one billion people currently receive assistance.

• Using the World Bank’s ASPIRE database I estimate that population coverage of SSN programs (% receiving any help) is growing at 9% per annum (3.5% points).

3

http://datatopics.worldbank.org/aspire/indicator_glance

Cruel irony: Poorer countries are less effective in reaching their poor

0

20

40

60

80

100

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 22000

GDP per capita at PPP for year of survey

Safety net coverage for poorest quintile (%)Safety net coverage for whole population (%)

Poorest quintile

Population

One billion poor; one billion SSN recipients

Living in poverty Receiving help from SSN

5

But mostly not the same people in poor countries!

How much does imperfect information constrain antipoverty policies in poor countries?

9/23/16

• The “poverty gap” calculation assumes that we can accurately identify poor people and tell how poor they are. This is a strong assumption.• Limitations of even the best household surveys

• Policies in practice use a smaller set of poverty proxies

• Reaching poor households does not mean we will reach all poor individuals

• Many questions left begging:• How well can we do against poverty with such data?

• What methods of aggregating these data into a score do best?

• How much do implementation lags restrict performance against poverty?

Brown, Ravallion & van de Walle 6

Questions for this presentation

1: How well can we identify poor households with the type of data routinely used by policy makers in poor countries?

2: How well can we identify poor individuals using such data?

In both, focus on Sub-Saharan Africa; the region with highest poverty measures today + large expansion in social protection programs.


How well can we identify poor households with the data routinely used in poor countries?


Based on Caitlin Brown, Martin Ravallion and Dominique van de Walle, “Poor Means Test? Econometric Targeting in Africa,” NBER Working Paper

Targeting with imperfect information

• Standard measures of economic welfare are not fully observed in the

population

• Instead, governments use proxies for targeting (location, family size, housing

conditions,…)

• Households receive a “score” = Proxy Means Testing (PMT)

• Main challenge is setting the weights of that score


Econometric targeting

• Regression for (log) consumption; h’hold 𝑖 in country 𝑗, date 𝑡, covariates 𝑥𝑖𝑗𝑡:

𝑦𝑖𝑗𝑡 = 𝛼𝑗𝑡 + 𝛽𝑗𝑡𝑥𝑖𝑗𝑡 + 𝜀𝑖𝑗𝑡 (𝑖 = 1, … , 𝑁𝑗𝑡)

• The PMT score:

𝑦𝑖𝑗𝑡 = 𝛼𝑗𝑡 + 𝛽𝑗𝑡𝑥𝑖𝑗𝑡

• H’holds in population with scores below some cutoff point are eligible.

• This is used by World Bank, other donors, and many governments. Supporters:

“it produces the best targeting outcome, esp., in excluding the non-poor”

(Grosh & Baker, 1995; Grosh, 1994; del Ninno & Mills, 2015).


Critics of econometric targeting

• Econometric targeting has been criticized for its “considerable inaccuracy at low levels of coverage.” Kidd & Wylde (2011)

• Transparency concerns (Adato & Rooppnaraine, 2004).

“...the targeting process as a whole is poorly understood… One informant from a geographically-targeted community noted: ‘Well, some people wonder why they weren’t targeted even though they live in this same area. So we tell them that the Bible says that many are called but few are chosen.’” Adato & Rooppnaraine (2004)

• Local social unrest generated by PMT (Cameron and Shah, 2004).

• Erosion of local social capital; distrust of local administrators.

• Can’t address all these concerns; but at the core: how well does it work?


Econometrics 101: Understanding regressions

• By design, a standard regression line passes through the means of the data.

• The residuals will be positively correlated with the dependent variable, and

more so the higher the variance of the residuals:

• Econometric targeting will tend to work less well near the extremes; tends to

overestimate living standards for the poorest.

• This is intrinsic to the use of regression.


If xy with 0),( xCov then 0)ˆ()ˆ,( VaryCov

In this paper we…

… provide a systematic assessment of econometric targeting as a tool for social policies aimed at reducing poverty. We:

• Study the most common form of “Basic PMT”

• Consider alternative models using additional covariates: “Extended PMT”

• Consider alternative estimation methods more appropriate to poverty reduction

• Characterize optimally targeted transfers with imperfect information

• Introduce plausible lags in implementation

• Compare econometric targeting to other targeting methods


How do we assess performance?


The literature on targeting

• “Targeting” tries to attain greater poverty impacts from transfers, or reduce the cost of a given impact.

• One strand of literature: Choose transfers across types of households to minimize a measure of poverty subject to a budget constraint (Kanbur, 1987: Ravallion and Chao, 1989; Glewwe, 1992).

• Others emphasize “targeting efficiency,” defined in terms of reducing certain “targeting errors” (Grosh and Baker, 1995; Coady et al., 2004).

• Here we shall study both types of measures.


What we don’t do

• We don’t question that household consumption is an adequate welfare

indicator. PMT makes that assumption, and we accept it for our purpose.

• However, we do allow for measurement errors (using panel data)

• We do not consider alternatives such as self-targeting using work

requirements (“workfare”) or community-based targeting in which local

communities are engaged directly in deciding who is poor and who is not.

• We do not consider the (economic, social and political) costs of targeting,

which have received some attention in the literature.


Measures of targeting


2. Exclusion Error Rate (EER): Proportion of poor who are identified as non-

poor: 𝐸𝐸𝑅 𝑗𝑡 =

𝑖=1

𝑁𝑗𝑡𝑤𝑖𝑗𝑡1( 𝑦𝑖𝑗𝑡>𝑧𝑗𝑡|𝑦𝑖𝑗𝑡≤𝑧𝑗𝑡)

𝑖=1

𝑁𝑗𝑡𝑤𝑖𝑗𝑡1(𝑦𝑖𝑗𝑡≤𝑧𝑗𝑡)

Measures cont.

3. Normalized Targeting Differential (NTD)

Difference between proportion of poor predicted as poor and proportion of the non-poor predicted as poor

𝑁𝑇𝐷𝑗𝑡 = 1 − 𝐸𝐸𝑅 𝑗𝑡 −

𝑖=1

𝑁𝑗𝑡𝑤𝑖𝑗𝑡1( 𝑦𝑖𝑗𝑡 ≤ 𝑧𝑗𝑡|𝑦𝑖𝑗𝑡 > 𝑧𝑗𝑡)

𝑖=1

𝑁𝑗𝑡𝑤𝑖𝑗𝑡1(𝑦𝑖𝑗𝑡 > 𝑧𝑗𝑡)

• When all and only the poor receive the transfer, 𝑁𝑇𝐷𝑗𝑡 = 1

• Under a uniform transfer scheme, 𝑁𝑇𝐷 𝑗𝑡 = 0

• When all and only the non-poor receive the transfer, 𝑁𝑇𝐷𝑗𝑡 = -1


Poverty reduction is the objective, not targeting per se

• EER and NTD better predict impact on poverty than IER (Ravallion, 2009)

• However, with the same data we can measure impacts on poverty directly, bypassing the targeting measures.

• Three poverty measures =>

1. Headcount index: 𝐻𝑗𝑡 = 𝑖=1

𝑁𝑗𝑡𝑤𝑖𝑗𝑡1(𝑦𝑖𝑗𝑡 ≤ 𝑧𝑗𝑡)

2. Poverty Gap Index: 𝑃𝐺𝑗𝑡 = 𝑦𝑖𝑗𝑡≤𝑧𝑗𝑡𝑤𝑖𝑗𝑡(1 − 𝑦𝑖𝑗𝑡/𝑧𝑗𝑡)

3. Watts Index: 𝑊𝑗𝑡 = 𝑦𝑖𝑗𝑡≤𝑧𝑗𝑡𝑤𝑖𝑗𝑡 ln(𝑧𝑗𝑡/𝑦𝑖𝑗𝑡)


Can we do better?


Can we do better by adding extra covariates?

• “Basic PMT”: Basic consumer durables and assets, demographic variables, attributes of the head

• “Extended PMT”: HH water source; housing materials; separate room for cooking; electricity; more details on the characteristics of the household

• But maybe the more important margin for improving PMT is the method of aggregating information =>


Can we do better by tailoring the estimator to the goal of poverty reduction?

• Regression gives equal weight to all error variances, whether for poor or rich: two options considered:

• Quantile regression setting the quantile at the overall poverty rate

• Poverty-weighted least-squares (PLS)

• Calibrate the PMT using OLS with poor households only

• Also a model adding some h’holds above the poverty line (“poor + 20”)

• Many other options (such as lasso regression). Instead we characterize the optimally differentiated transfers for the same information set.


How much better with optimal transfers?


Think of the information-feasible transfer as a function of the observed x’s.

The problem is to choose the parameters of a score for assigning the real-

world transfers based on the x’s:

m

k

k

ijt

k

jtijt x0

0)(

The choice of the score parameters k

jt is made to minimize the Watts index:

jtN

i

ijtijtjtijtjt yzwW1

))]/(ln(,0max[

The choice is constrained by the budget:

jtN

iijtijtjt wB

1

We solve this problem numerically (using Matlab’s “fmincon” program).

Data and measurement assumptions


Data from the LSMS

Data from the World Bank’s Living Standards and Measurement Surveys

• Burkina Faso, Ethiopia, Ghana, Malawi, Mali, Niger, Nigeria, Tanzania,

Uganda

• Survey years range from 2009 to 2014

• Between 2,600 and 10,000 observations per country


Measurement assumptions

• Log per capita household consumption as dep. var. (spatially deflated)

• Three poverty rates: 𝐻𝑗𝑡 = 0.2 and 0.4 and national poverty rate

• 40% approximates average poverty rate in SSA

• 20% allows for focus on the poorest households

• Cut-off points for PMT scores calculated in two ways:

1. Fix poverty line: 𝑦𝑗𝑡 ≤ 𝑧𝑗𝑡 , 𝑧𝑗𝑡 ≡ 𝐹𝑗𝑡−1 0.2 or 𝐹𝑗𝑡

−1 0.4

2. Fix poverty rate: assign lowest 20 or 40% of 𝑦𝑗𝑡 as poor

• Under method 2, 𝐼𝐸𝑅 𝑗𝑡 = 𝐸𝐸𝑅 𝑗𝑡

• Each time a poor household is incorrectly identified as non-poor, a non-poor household is identified as poor (since the total rate of poor is kept constant)

• We call this the targeting error rate, 𝑇𝐸𝑅 𝑗𝑡


Results


Basic PMT

Variables in 𝑥𝑖𝑗𝑡 include:

• Type of toilet a household has

• Floor, wall and roofing material

• Type of fuel most commonly used for cooking

• Characteristics of the household head, including gender, education and occupation

• Household’s religion and demographic composition

• Dummy variables for categories of household size, age of head, month of survey and region of residence

Average of 𝑅2 is 0.54, with a range from 0.32 (for Ethiopia) to 0.65 (Malawi)

• Similar to what is found in existing literature


Covariates (x)


Basic Extras in Extended Toilet: flush Water: piped

Toilet: pit Water: well or outside pipe

Floor: finished Persons per room

Wall: finished Separate room for kitchen

Roof: finished Has radio

Fuel: electricity gas or kerosene Has television

Fuel: charcoal Has fridge or freezer

Urban area Has bicycle

Female head Has motorbike

Head has at least primary education Has car or truck

Head has at least secondary education Has telephone (owns)

Divorced or separated head Has mobile phone (owns)

Widowed head Has computer

Head worked in a paid job (last week) Has video player

Head worked in own non-farm enterprise Has any type of stove

Muslim Has sewing machine

Christian Has air conditioner or fan

HH share: girls 5 and under Has iron

HH share: boys 5 and under Has satellite dish

HH share: girls 6 to 14 Has electric generator

HH share: boys 6 to 14 Household owns dwelling

HH share: women 65 and over HH share: widows

HH share: disabled men 15+

HH share: disabled women 15+

HH share: orphan boys less than 15

HH share: orphan girls less than 15

Summary stats on regressions


Burkina

Faso Ethiopia Ghana Malawi Mali Niger Nigeria Tanzania Uganda

Basic PMT

R2 0.644 0.319 0.561 0.573 0.418 0.634 0.581 0.585 0.498

N 10265 5017 4224 3931 3212 3833 3720 4753 2650

Extended PMT

R2 0.724 0.381 0.587 0.674 0.520 0.718 0.666 0.647 0.596

N 10265 5017 4224 3931 3212 3833 3720 4753 2650

Actual and predicted values for Basic PMT


Exclusion errors

Inclu-sionerrors

Results for Basic PMT: Targeting errors

32

Inclusion error

rate

Exclusion error

rate

Inclusion error

rate

Exclusion error

rate

Targeting

Error

Targeting

Error

(IER) (EER) (IER) (EER) (TER) (TER)

Fixed poverty line Fixed poverty rate

z = F -1 (0.2) z = F-1(0.4) H = 0.2 H = 0.4

Burkina

Faso 0.401 0.751 0.304 0.375 0.522 0.329

Ethiopia 0.515 0.945 0.396 0.562 0.621 0.413

Ghana 0.354 0.628 0.257 0.350 0.428 0.288

Malawi 0.431 0.880 0.333 0.451 0.553 0.373

Mali 1.000 1.000 0.348 0.485 0.553 0.375

Niger 0.539 0.875 0.384 0.340 0.584 0.362

Nigeria 0.332 0.548 0.247 0.243 0.392 0.244

Tanzania 0.396 0.822 0.323 0.291 0.513 0.314

Uganda 0.357 0.663 0.350 0.294 0.455 0.335

Mean 0.481 0.807 0.309 0.359 0.505 0.319 About half those identified

as poor are notAbout 80% of the poor are not identified as poor

Basic PMT reduces inclusion errors, at a cost to exclusion errors

• For a poverty rate of 20% and a fixed line, the PMT method has nearly halved the IER that would be obtained with a uniform transfer payment.

• However, this has come at the expense of exclusion errors. The average EER is sizeable, with 81% of those who are in the poorest 20% in terms of survey-based consumption being incorrectly identified as non-poor by PMT.

• There is considerable variation across countries, with IER ranging from 33% to 100%, and EER from 55% to 100%.


Residuals for Basic PMT against log consumption


Figure 2: Residuals for Basic PMT and log consumption per capita

Econometric targeting using OLS substantially overestimates consumption by the poorest

• As expected, the residuals tend to be lower (more negative) for poor people.

• What is notable is how much the PMT regression is over-estimating the living standards of the poorest.

• For the poorest 20% in terms of actual consumption, the mean residual ranges from -0.73 to -0.37, implying that the PMT regressions yield predicted consumptions for the poor between 50% and 100% above their actual consumption!


Results for poverty-focused methods

• Two “poverty-focused” options to OLS:

• Quantile regression using the poverty rate as the quantile

• “Poverty-weighted least squares” (PLS)

• Both => reduction in exclusion error rate using a fixed poverty line

• This comes at the cost of higher inclusion errors, especially for the lower

poverty line

• The PLS method is better at covering the poor but has very high inclusion

errors

• The “poor + 20” PLS regression reduces the inclusion error relative to the PLS,

but also increases the exclusion error


Results for Extended PMT

• Adding variables increases 𝑅2 and improves targeting. But by how much?

• Additional variables in 𝑥𝑖𝑗𝑡 include:

• Household’s water source

• More details on housing conditions, e.g. number of household members per room; separate room for cooking

• Whether the household has electricity

• Household assets

• As expected, values of 𝑅2 are higher

• Simple average is 0.61 with a range from 0.72 (Burkina) to 0.38 (Ethiopia)

• In most cases the gains are relatively small, though the number of explanatory variables has almost doubled


Targeting errors for Extended PMT


Inclusion

error rate

Exclusion

error rate

Inclusion error

rate

Exclusion

error rate

Targeting

Error

Targeting

Error

(IER) (EER) (IER) (EER) (TER) (TER)

Fixed poverty line Fixed poverty rate

z = F -1 (0.2) z = F-1(0.4) H = 0.2 H = 0.4

Burkina Faso 0.334 0.626 0.257 0.336 0.449 0.282

Ethiopia 0.419 0.857 0.373 0.455 0.541 0.405

Ghana 0.349 0.591 0.256 0.331 0.421 0.267

Malawi 0.439 0.698 0.295 0.374 0.470 0.315

Mali 0.444 0.951 0.344 0.444 0.572 0.341

Niger 0.458 0.751 0.328 0.323 0.539 0.327

Nigeria 0.330 0.496 0.228 0.217 0.384 0.223

Tanzania 0.403 0.670 0.283 0.279 0.481 0.281

Uganda 0.379 0.552 0.313 0.279 0.467 0.307

Mean 0.362 0.639 0.283 0.308 0.456 0292

Basic PMT

Mean 0.481 0.807 0.309 0.359 0.505 0.319

Further results (see paper for details)

• Using panel data for a subset of countries to measure time-mean consumption.

• We also estimated the model separately for urban and rural households

• Performs slightly worse than when model is estimated with full sample

• Consider additional models with more variables than the Extended PMT

• Food security variables

• Community-level variables

• Provide only modest improvements on targeting performance

• Compare performance with “pruned” models

• Stepwise regression and Extended PMT variables

• Similar results to the Basic PMT


Targeting differentials

• The mean NTD for Basic PMT is 0.21, meaning that if program participation was based on the PMT scores the participation rate for the poor would be 21% points higher than that for the non-poor.

• The poverty-quantile method yields the highest NTD, at around 0.49 on average for Basic PMT, rising to 0.53 for the Extended PMT.

• For all nine countries, the poverty-quantile regression method comes out best by this measure.


Mean

Basic PMT covariates

Basic PMT 0.214

Using means from panel data 0.366

Poverty quantile regression 0.485

Poverty weighted: Poor only 0.107

Poverty weighted: Poor + 20 0.421

PMT with Urban/Rural 0242

Extended PMT covariates Extended PMT 0.309





Stepwise (p=0.01) 0.249

HH Shocks + Food Security 0.333

Shocks, Food Security + Community

Variables 0.331

Impacts on poverty for stylized transfer programs

Setting the budget at each country’s aggregate poverty gap, we compare how

well the different PMT specifications do at reducing the three poverty measures

considered

• We assume a poverty line corresponding to H = 0.2

• A uniform per capita transfer is given to all households who are predicted to be

below the poverty line according to the PMT

• Aggregate poverty gap divided by total number of people predicted to be poor

• This transfer amount is distributed to households according to their size


Comparisons with other targeting methods

We also compare the poverty-reducing performance of PMT with other targeting methods:

• Uniform transfer (“basic income”)

• Every household receives a transfer adjusted for household size

• Categorical targeting

• Use several different categories, for example elderly, widows, households with children

• If a household has two members who fit the category, or one person fits two categories, the transfer amount is doubled (and so on for three or more categories)


Impacts on poverty rate (H=20%)

• Most methods bring H down to 15-16%; more than 3/4 of the poor remain poor.

• Basic PMT does only slightly better than UBI with the same budget,

• Basic PMT does not do as well as the universal transfer in one third of countries.

• Quantile regression does better, bringing the poverty rate down to about one % point below that for Basic PMT.

• Categorical targeting does well. Targeting to h’holds with elderly, widows, disabled and children does as well as Basic PMT.

• The poverty quantile regression method typically has the greatest impact on poverty.


Mean

Universal (basic income) 0.171

Basic PMT covariates

Basic PMT 0.163





PMT with Urban/Rural 0.159

Extended PMT covariates

Extended PMT 0.154





Allowing for lags in implementation

• Use panel data; between 1 and 2 years apart

• Calibrate PMT using Round 1; two different lags

1. Use Round 2 covariates to determine eligibility – lag in estimation only

2. Use Round 1 covariates – lag in implementation

• In general lags increase targeting errors, and make impact of program

more uncertain

• On average, poverty reduction is hampered by one percentage point on

average due to lags


Allowing optimally differentiated transfers

• So far: standard practice of giving the same transfer payment to all those predicted to be poor using PMT.

• Differentiating the transfers could be expected to work better if the predicted poverty gaps are quite accurate.

• However, we have already seen that this is not the case—that PMT works poorly in predicting the levels of living of the poorest.

• So it is unclear on a priori grounds whether differentiated transfers will have larger impacts on poverty.

• How much does the constraint of relying on uniform transfers to the “predicted poor” limit the effectiveness of PMT?



Optimal transfers

PMT Linear Non-linear

Burkina Faso 0.044 0.038 0.036

Ethiopia 0.074 0.056 0.053

Ghana 0.066 0.048 0.044

Malawi 0.060 0.051 0.044

Mali 0.064 0.057 0.053

Niger 0.048 0.042 0.039

Nigeria 0.062 0.049 0.043

Tanzania 0.069 0.060 0.051

Uganda 0.083 0.061 0.053

Mean 0.066 0.054 0.049

Watts index post transfer using differentiated transfers

Modest gain only using differentiated transfers

• Recall: optimal allocation based on the set of PMT covariates. Watts index is the objective function.

• Objective function tends to be quite flat in the sub-set of the parameter space corresponding to the various solutions found.

• Overall, filling the predicted gaps brings only a small further reduction in the poverty measures.


How well can we identify poor individuals using standard household level data?


Based on Caitlin Brown, Martin Ravallion and Dominique van de Walle, “Are Poor Individuals Mainly Found in Poor Households? Evidence Using Nutrition Data for Africa.” NBER Working Paper

Poverty is an individual deprivation, but individual consumption is rarely observed

• Even if we can reach poor households, that does not guarantee that we reach poor individuals. And poverty is ultimately an individual state.

• Unitary model of h’hold (single utility fn) finds little support empirically; various alternatives (Chiappori and Mazzocco; Baland and Ziparo). • These alternative models permit new sources of inequality within

households, such as in reservation utility levels.

• An extensive literature details intra-household inequalities in resource allocations and outcomes (reviewed in World Bank 2012).

• Targeting poor h’holds may miss may disadvantaged individuals and targeted h’holds may not allocate the benefits to the neediest.


One clue to individual poverty: nutritional status

• Anthropometric data at individual level

• Nutritionally-vulnerable women and children are a high priority for social policies• Malnutrition has both immediate and long-term social and economic costs

• However, individual level interventions are often intrusive and expensive• Difficult to observe undernourishment in large populations

• Key question: is household-level targeting effective at reaching poor individuals as indicated by nutritional status?


Evidence on wealth and nutrition

• ”Wealth effect” suggests poorer households are more likely to include undernourished individuals

• Behrman and Deolalikar (1987); Ravallion (1990, 1992); Filmer and Pritchett (1999, 2000); Filmer and Scott (2012)

• However, even strong correlations between wealth and nutrition may not mean wealth is a satisfactory predictor of individual nutrition

• Other factors may mean non-poor households also include the undernourished

• Intra-household inequality; local health environment; heterogeneous responses to shocks

Intra-household inequality constrains performance, but how much?

• Our theoretical model formalizes the intuition that heterogeneity due to intra-household inequality or the local health environment diminishes the scope for reaching poor individuals (nutritionally vulnerable women and children) by targeting poor households.

• But how much does this matter empirically? Is the wealth effect on individual nutritional status strong enough to allow satisfactory targeting of vulnerable women and children?


Data sources

Demographic and Health Surveys (DHS) for 30 countries in SSA • 350,000 women and children

• Women 15 to 49; children under 5

• Nutritional outcomes: BMI, height-for-age (stunting) and weight-for-height (wasting)

• DHS wealth index

• Additional household- and individual-level variables

Wealth index constructed using variables related to a household’s welfare • Household assets, materials used in housing construction, access to water and

sanitation etc.

• Country specific, mean zero and standard deviation of one

Data cont.

• Household consumption may be a better indicator of individual nutritional status than wealth index • DHS does not collect consumption data

• Also draw from Living Standards and Measurement (LSMS) surveys

• 7 countries with nutritional data

• Focus on bottom 20% and 40% of households based on wealth index• 40% roughly coincides with overall poverty rate in SSA

• Include other indicators relevant to nutritional outcomes • GDP per capita, poverty rate, female literacy rate, access to improved water

and sanitation.

Nutritional outcomes and household wealth

Regression coefficients of individual nutritional outcomes on the DHS household wealth index

BMI Height-for-age Weight-for-height

Benin 0.194*** 0.189*** 0.080***

Burkina Faso 0.279*** 0.281*** 0.097***

Burundi 0.242*** 0.505*** 0.115***

Cameroon 0.285*** 0.451*** 0.257***

Congo 0.265*** 0.292*** 0.051**

Cote D'Ivoire 0.203*** 0.279*** 0.043

DRC 0.276*** 0.378*** 0.073***

Ethiopia 0.352*** 0.374*** 0.229***

Gabon 0.182*** 0.397*** 0.056**

Proportion of undernourished individuals in the poorest 20% and 40% of households

Poorest 20% of households Poorest 40% of households

Underweight

women

Stunted

children

Wasted

children

Underweight

women

Stunted

children

Wasted

children

Benin 0.248 0.233 0.223 0.444 0.446 0.464

Burkina Faso 0.307 0.242 0.224 0.551 0.458 0.433

Burundi 0.276 0.249 0.281 0.464 0.451 0.506

Cameroon 0.396 0.326 0.364 0.637 0.594 0.630

Congo 0.221 0.310 0.232 0.460 0.534 0.465

Cote D'Ivoire 0.226 0.289 0.240 0.414 0.516 0.447

DRC 0.252 0.247 0.209 0.521 0.482 0.442

Ethiopia 0.235 0.218 0.259 0.461 0.445 0.534

Gabon 0.246 0.434 0.206 0.422 0.634 0.388

Joint probabilities of being undernourished and wealth poor

Poorest 20% of Households Poorest 40% of Households

Underweight

women

Stunted

children

Wasted

children

Underweight

women

Stunted

children

Wasted

children

Benin 0.016 0.095 0.032 0.028 0.181 0.067

Burkina Faso 0.047 0.072 0.031 0.085 0.136 0.060

Burundi 0.044 0.129 0.014 0.074 0.233 0.026

Cameroon 0.027 0.091 0.018 0.043 0.166 0.031

Congo 0.032 0.058 0.012 0.066 0.100 0.023

Cote D’Ivoire 0.018 0.069 0.017 0.032 0.123 0.032

DRC 0.036 0.090 0.015 0.075 0.177 0.032

Ethiopia 0.063 0.085 0.022 0.123 0.173 0.046

Gabon 0.018 0.057 0.007 0.031 0.083 0.012

Most of Africa’s nutritionally vulnerable women and children are not found in poor households

• On average, 75% of undernourished individuals are not found in the poorest 20% of households using household wealth.

• Similar results using consumption (slightly better for women)

• About half of undernourished women and children are not found in the bottom 40%.

• Joint probabilities for underweight women and wasted children close to zero.

• Results are robust to:• Adjusting for demographic imbalance of wealth fractiles• Reweighting the wealth index • Adding additional household-level variables• Adding individual-level variables

How do these results vary with the rate of undernutrition?

• Are a higher proportion of undernourished found in poor households in countries with a high rate of undernourishment?

• Look at relationship between Pr 𝑤 < 1|𝑛 < 1 and Pr 𝑛 < 1 =>


Countries with fewer underweight women have a higher proportion of those women in wealth-poor households

.0

.1

.2

.3

.4

.5

.6

.7

.00 .05 .10 .15 .20 .25 .30

Poorest 20%

(r=-0.31)

Poorest 40%

(r=-0.22)

Share of women who are underweight

Pro

po

rtio

n o

f u

nd

erw

eig

ht w

om

en

fo

un

d

in p

oo

r h

ou

se

ho

lds

Countries with fewer stunted children have a higher proportion of those children in wealth-poor households

.0

.1

.2

.3

.4

.5

.6

.7

.0 .1 .2 .3 .4 .5

Incidence of stunted children

Pro

po

rtio

n o

f stu

nte

d c

hild

ren

fo

un

d

in w

ea

lth

-po

or

ho

use

ho

lds

Poorest 20%

(r=-0.47)

Poorest 40%

(r=-0.56)

Conclusions


64

Exclusionerrors

InclusionerrorsTargeting trade off

• Strengths and weaknesses of econometric

targeting

• It substantially reduces inclusion errors in an

antipoverty program

• But this comes at the cost of a large rise in

exclusion errors

Uniform-untargeted (basic income)

Basic PMT

New PMT methods

• The choice of method ultimately rests on the weight policy makers give

to these two sources of errors

• High weight on poverty => higher weight on reducing exclusion errors

Our findings on econometric targeting

• This filters out the nonpoor but excludes many poor people, diminishing impact on poverty (relative to untargeted transfers).

• More data and better methods (geared towards poverty reduction) do better, although the gains to the poor are typically modest.

• In some cases, much simpler targeting methods dominate.

• However, no method we consider performs especially well when poverty reduction is the goal.

• Even with sufficient budget to eliminate poverty with full information, none of the methods bring the poverty rate below 75% of its initial value.

• Prevailing methods do not reliably reach the poorest.


Household data are weak in identifying poor individuals• The bulk of individually poor women and children—as indicated by

nutritional status—are not found in observationally poor households• On average, about 75% of undernourished women and children are not found

in the poorest 20% using the wealth index

• When the rate of undernourishment is high, these individuals are more spread out across the wealth distribution

• Adding variables e.g. household-level, individual-level improves targeting performance, but still misses many.

• Results are robust to: • Reweighting the wealth index

• Adding additional household-level variables

• Adding individual-level variables


Overall conclusion: Eliminating poverty using transfers is much harder than the total poverty gap suggests

• With budget = aggregate poverty gap we will not be able to eliminate household poverty with the data routinely used by antipoverty policies.

• And even when we have correctly identified poor households, the majority of poor individuals (as indicated by nutritional status) will be missed.

• Much better data and/or much better targeting methods will be needed.


Time to reconsider basic income

• Push for fine targeting has dominated social policy discussions in developing countries.


0

20

40

60

80

100

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 22000

GDP per capita at PPP for year of survey

Safety net coverage for poorest quintile (%)Safety net coverage for whole population (%)

Poorest quintile

Population

• However, the information constraints are severe. Weak administrative capabilities.

• (Other constraints: incentive effects and political economy.)

• Basic (full) income should be on the menu of options in developing countries. • India example: Basic income with same budget as NREGA has more impact on

poverty!

• Promotes social cohesion and sustainability of social policies.

• So too should the development of administrative capabilities over longer term.

Further reading:

Martin Ravallion, The Economics of Poverty: History, Measurement and Policy

Oxford University Press, 2016

economicsandpoverty.com

Thank you for your attention!

Government & Nonprofit

Informational Constraints on Antipoverty Programs: Evidence for Africa