Poverty Measurement in India and Bangladesh: a Great Indian Rope Trick? Seminar presentation IDPM, University of Manchester, 09/10/07 Richard Palmer-Jones,

Poverty Measurement in India and Bangladesh:

a Great Indian Rope Trick?Seminar presentation IDPM,

University of Manchester, 09/10/07Richard Palmer-Jones,

School of Development Studies,University of East Anglia,

Norwich, NR4 7TJWith acknowledgements but no inculpation of Amaresh Dubey, or

Kunal Sen, sometime partners in this ….

DEVODGthe Indian rope trick is “[S]ometimes described as "the world’s greatest illusion"”. Its origins are obscure but our use of it is to suggest that and claim that current methods provide a reliable basis for poverty lines and poverty aggregates that represent a comparable standard of welfare is an illusion

Poverty Measurement in India and Bangladesh: a Great Indian Rope Trick?1

Abstract The measurement of poverty is important because poverty is a key policy variable in poor countries, although to a much lesser extent in developed ones. To be useful for policy analysis poverty must be measured by a common yardstick of welfare in different social, spatial, and temporal domains. India and Bangladesh are important cases of poverty measurement because of the prominence of the former in debates about poverty, poverty measurement and development, and in the latter because the methods of poverty measurement have played a significant role in the methodological literature, as suggested by the title of the paper in Economic Development and Cultural Change “When Method Matters”, which deals with poverty measurement in Bangladesh.

This paper asks “how comparable are these poverty lines?”. In India the official poverty lines are anchored in the expenditure at which household in the rural and urban sector consume calorie norms specified for each sector and updates them spatially and temporally using CPIs. Many authors have noted that over time calorie consumption and the updated poverty lines has fallen significantly below the original norms. Angus Deaton has recently suggested replacing these poverty lines with ones based on Unit Value Consumer Price Indexes (CPIs) derived from the National Sample Survey household expenditure surveys anchored in a single arbitrary poverty line. Several different methods have been used to produce poverty lines in Bangladesh. Those preferred by officials are based in calorie norms, but authors associated with the World Bank, criticizing the calorie based methods have argued for “Cost of Basic Needs” (CBM) methods, and these methods have become the orthodoxy in recent manuals, handbooks and trainings on poverty measurement from the World Bank and UNSTATS (among others in the aid business). Surprisingly, the requirements for the CBN method to be welfare consistent seem to be the same as those for the calorie based methods which they were developed to replace. The pure CPI methods fall down because there can be no reason to think that the procedures used to produce them can correspond to true cost of living indexes, since they ignore many variables that can reasonably be thought to affect the transformation of commodities, and calories, into well-being (as well as having many other deficiencies).

Thus, this paper argues that there are no good theoretical grounds for thinking that any of the methods used or proposed for setting poverty lines in these contexts give any grounds for believing that they correspond to the required common yardstick of well-being. The question that arises, then, is what is going on when so much effort is devoted to this sort of money-metric poverty measurement if they are based on a grand illusion? It suggests some other more useful things to do for the participants in the production and use of poverty statistics. 1 Richard Palmer Jones, School of Development Studies, University of East Anglia for Manchester University Institute of Development Policy and Analysis, 9th October, 2007

Outline• Poverty is an important policy variable• India and Bangladesh are significant case studies

– but there is controversy over trends (and patterns)• Indian Planning Commission claims poverty come down

– critics suggest hunger and poverty have increased– Apparent modest improvements in child undernutrition but lacking decentralised

recent data• In Bangladesh World Bank (and BBS) claim poverty has come down but

child undernutrition may not (by WHO method).• Standard methods of poverty assessment have dubious theoretical

bases– Methods

• DCI, FEI, CBN, CPI– Practice & Precept– Theory revisited

• What does it mean and what to do?

-0.5

-1

-1.5

-2

-2.5

ann

ual r

ate

of

HC

R p

ove

rty

decl

ine

(%

pa)

Himanshu Dev & Ravi RPJ

rural urban all rural urban all rural urban all

Annual Rates of Poverty Decline by Calculation Method

1983-1993/4 1993/4-2004/5

HCR Poverty decline in India

40

45

50

55

60

hcr

1983 1985 1988 1991 1995 2000 2005

CBN1CBN2 CBN2&UVCP1FEI HIESDCI HIESFEI PMSDCI PMSCBN3/CPI2World Bank, 2007

Rural

20

30

40

50

hcr

1983 1985 1988 1991 1995 2000 2005

CBN1CBN2 CBN2&UVCP1FEI HIESDCI HIESFEI PMSDCI PMSCBN3/CPI2hcr

Urban

Note: Upper Poverty Line (2122kcal pcpd); 1983/4-1988/9 - 2200kcal pcpd

Sources: CBN1 Ravallion & Sen, Sen & Mujeree; CBN2 World Bank, 1998; CBN2&UVCPIa World Bank, 2002; FEI & DCI - 1983/4-1991/2 BBS; CBN3/CPI2 1991/2, 1995/6 & 2000/1 Author; 2007, BBS/World Bank, 2007

Figure 3: Rural and Urban Poverty, Bangladesh, 1983/4 - 2005

0

.2

.4

.6

.8%

stu

nte

d

nchs who

1992 1996 2000 2005 1992 1996 2000 2005

nchs: calcualted by BBS using NCHS standardsWHO: calculated using igrowup.ado from WHO Anthro, 2005, www site

Author's calculation from CNS 1992,1996, 2000 & 2005

Rural and Urban Stunting in BangladeshChildren (6-71 months)

ruralurban

nfhs1/1992 nfhs2/1998 nfhs3/2004

stunted 47.3 45 38

w asted 19.7 16 19

underw eight 52.7 47 46

imr 79 68 57

Women BMI < normal 33 40.3 33

< 3 years

Child Anthropometry in India

Why be concerned about poverty?• A personal history of “trickle down”

– Irrigation, agricultural growth, wage rates of agricultural labourers and poverty in Bangladesh and India

• MDG Goal No 1 (and “headline” value)• PRSPs & assessments of progress

– (south Asia – including Afghanistan)

• Manuals from World Bank and UNSTATS– Including Sourcebook for PRSP

• Why Poverty– Outrage– policy analysis? – poverty profiles– Poverty comparisons

• Common yardstick – the same thing

Also 2005

Methods • Minimum socially acceptable standard of living

– Comparable across domains space, social group and time)

• Set a poverty line(s) and aggregate– Identity, Incidence, Intensity & Inequality

• Poverty Lines – Calorie based

• Rowntree – cost of nutrition & allowance for non-food expenditure

• Direct Calorie and Food Energy Intake (FEI & DCI)• Cost of Basic Needs (CBN)

– Cost of Living Index methods (CPI)• CoGIs or CoLIs?

Aggregation is (arguably) less important than incidence

• Robustness - stochastic dominance does not address the key problem of comparability– Compare aggregates for different relative poverty lines

0

.2

.4

.6

.8

1

cum

ulat

ive m

pce

0 200 400 600 800 1000monthly per capita expenditure

AP Coastal North Andhra Inland NorthernAP Coastal South AP Inland Southern

MPCE CDF, Andhra Pradesh,55th Round (rural)

FEI & DCI

1 1 normfeiz f g c

F and y are Food and total expenditure household per capita respectively

suppose F=f(y) is a well behaved function of y with f' >0 & f'' < 0

and that c (per capita calorie intake) is a a well behaved fnorm

unction of g(F), with g'>0, g''<0

then if c is the normative household per capita calorie requirement, then the FEI PL is:

This is usually estimated from a regression of reported (constructed) expenditure per capita on reported (constructed) per capita calorie “consumption”

DCI HCR poverty is the ratio of population with c < cnorm / total population

FEI Poverty Lines

Food

exp

end

iture

= f(

cals

pcp

d)

Tota

l exp

endi

ture

fun c

ti on

=f

(cal

spc p

d )

FEI poverty line

Food expenditure to attain normative calorie requirement

Tota

l exp

endi

ture

(m

pce

)

Calories per capita per day

Non-food expenditure when expenditure meets normative calorie requirements

food

CBN Method – recommended by World Bank (and UNSTATS)

• Food component – zfood• Non-food component - two levels (znfu & znfl)

– Upper and lower PLs (zu & zl)

• Food Component - recommended– Behavioural food bundle (households around poverty line)

• Scaled to normative calories• Priced at local prices gives zf – the cost of food bundle• Tarp et al., 2002, variant - different food bundles in different domains

• Non-food component• Inverse Engel share of households around poverty line• Estimate the following regression

2

1 2log logi i ii if f

i

f y y yd d e

y z z

Where zf is the food poverty line, yi is total expenditure, and d are demographic variablesAnd f(yi) is food expenditure

Food expenditure= f m pce(

)

45o

Spending required to purchasenutritional requirements

Upper poverty line

Lower poverty line

Total expenditure (mpce)

Lower non-food componentnon-food share of hhouseholds whosetotal expenditure = Z = Z*(1-a)

f

f j

zf

zf

Exp

end

iture

(p

er c

apita

)

Upper non-food component =non-food share of households whose food expenditure = Z= (1-f )*z

f

f

- 1

Zf

f (z )-1f

Z *(2- )f j

Upper poverty line =

Lower poverty line =

Zl

Zu

Food poverty line

CBN Poverty in Bangladesh• R&S, 1996, for 1983/4 – 1991/21

– normative food bundle (from Alamgir, 1974)• Not typical of consumption of poor

– More high calorie cost foods (pulses, milk, oils, meat, fish, sugars, fruits) (Unclear origin of food “unit values”2 – not poor relevant)

• Non-food share – “guesstimated” at 35% of cost of food in 1983/4– Updated using national Rural and Urban non-food CPIs

• Wodon & World Bank, 1998; 1983/4 – 1995/6– Same normative food bundle– UVs estimated by “regression” to be poor relevant– Non-food share from inverse Engel Curve for each HIES

• World Bank 2002– Use Wodon 1991 CBN PLs and update using “synthetic” CPIs

• “Better”• World Bank 2005

– Re-estimate CBN using same food bundle, 2005 prices & inverse Engel shares

1: updated by Sen and Mujeri; based on critique of FEI & DCI for 1995/6 & 2000/12. median “unit values” for rural and urban sectors for 11 “composite” groups of items

In its 1995-96 HES report, BBS re-estimated poverty lines for each year separately using the methodology described below for the base year. However, one of the disadvantages of this approach is that it does not guarantee that the poverty lines calculated across years represent basic-needs bundles of constant value. In particular, if living standards in a country improve over time, and even poor households spend a larger share of their income on non-food items, the allowance made for these items in the poverty line increases over time as well. The current methodology is superior in that it ensures that comparisons of poverty rates over time are based on poverty lines that are held constant in real value terms (World Bank, 2002:92, emphasis added)

CBN

FEIMpce Calspcpd

Foo

d e

xoe

ndi

ture

per

cap

itaC

als

pe

r ca

pita

pe

r d

ay

E(food_exp | calspcpd)

E(mpce | food_exp)

E(mpce | cals)

Normative calories

Estimate: mpce = f(calspcpd)

Estimate cost of calspcpdfrom food bundle scaled to normative calories

Estimate non-food share(nfs) for households whosefood expenditure = zfood& Compute mpce = zfood/(1- nfs)

zfood

Poverty lines

•Is CBN so different from FEI?

•Calorie base to food component•Estimate non-food shareby Engel regression

•Difference is constraint on cost per calorie

•Both give rising poverty

•Both are inconsistent with elementary demand theory

CBN & FEI (cost per kcal unconstrained)

CBN PL99.9 Rps pcpm

050

100

150

200

exp

. fo

od

exp

en

d.

pe

r ca

pita

pe

r m

on

th

0100200300400500

monthly per capita expenditure

normative calories2400 cals pcpd

050

100

150

200

exp

. fo

od

exp

en

d.

pe

r ca

pita

pe

r m

on

th

0 1000 2000 3000 4000 5000

kcal. per capita per day

FEI PL127.0 Rps pcpm

010

0020

0030

0040

00

exp

. ca

lori

e a

vail.

pe

r ca

pita

pe

r m

on

th

0100200300400monthly pre capita expenditure

010

0020

0030

0040

00

kcal

. per

cap

ita p

er d

ay

0 1000 2000 3000 4000 5000kcals pcpd

Bottom half: FEI UnconstrainedSource: 38th Round, AP, Rural: Fei&CBNMethods.do

top half: CBN cost per calorie < 1.4 Rpspcal

CBN & FEI (rps/kcal < 1.4)

CBN PL99.9 Rps pcpm

05

01

001

502

00

exp.

foo

d ex

pend

. pe

r ca

pita

per

mo

nth

0100200300400500

monthly per capita expenditure

normative calories2400 cals pcpd

05

01

001

502

00ex

p. f

ood

expe

nd.

per

cap

ita p

er m

ont

h

0 1000 2000 3000 4000 5000

kcal. per capita per day

FEI PL99.9 Rps pcpm

01

000

200

03

000

400

0

exp.

cal

orie

ava

il. p

er c

apita

per

mo

nth

0100200300400

monthly pre capita expenditure

01

000

200

03

000

400

0

kcal

. pe

r ca

pita

per

da

y

0 1000 2000 3000 4000 5000

kcals pcpd

Source: 38th Round, AP, Rural: Fei&CBNMethods.do

cost per calorie < 1.4 Rpspcal

Non-calories

Non-calories Calories

Xc1c

Xc0Xc

1ncXncal

1nc Xncals0Xncals

1c

Pnc0

Pnc1

H (p ,p , ....| U )x1 0 1 0

M (p , p ,..)x1 0 1

X00 X0

1c

X01

Xcals1c Xcals

1nc Xcals0

M (p , p ,..)x0 0 1

H (p ,p , ....| U )x0 0 1 0

Pnc1

Pc1 P1

1Pc

0

Pnc0

U0

U1

Calories

Pnc

Hicksian demand curves (utility compensated) show fall in demand for calories with fall in relative price of non-calories

Non-calories

Non-calories Calories

Xc1c Xc

feiXc

1ncXncal

1nc Xncals0Xncals

1c

Pnc0

Pnc1

H (p ,p , ....| U )x1 0 1 0

M (p , p ,..)x1 0 1

X00 X0

1c

X01

Xcals1c Xcals

1nc Xcals0

M (p , p ,..)x0 0 1

H (p ,p , ....| U )x0 0 1 0

Pnc1

Pc1 P1

1Pc

0

Pnc0

U0

U1

Calories

Pnc

U2

Fei poverty lineexpenditure will bemuch higher

FEI poverty line expenditure is higher than utility compensated expenditure

Calories

Non

-cal

orie

s

Freein sector 1

Income needed in sector 2 to induce C consumption of calories

0

Sector 2 consumptionat same mpce

Sector 1 consumptionat same mpce given “free” noncalories

Common utility functions

Y1* Y2

*

Y is the income in sector 1 required to achieve normative caloriesY is the income in sector 2 required to achieve U , the poverty line utility level

1

2 1

*

* *

Calories

X0

X0 X1

X10 X1

1

X11c

X00

X01cX0

1

P10

P11

M (p ,p ,..)x0 0 1

X00

X01c

X01

u0

u1

M (p ,p ,..)x0 0 1

No

n-ca

lori

es

Hicksian demand curves disappear with zero utility compensated substitution.

CPI Methods:

CoLI Poverty Lines & Utility consistency

PL * rR

PLR

PL *(1- )U u

Pl = U PL *UVPCIR

Ru

Rural f

ood Enge l c

urve

450

450

Urban fo

od Engel c

u rve

Per capita expenditure

Per

cap

ita e

xpen

ditu

re

PL *(1- )R R

OPLR43*( *UVCPI -1)r rais43OPL43 * rR

OPL43R

OPLR43*(1- )*UVPCIr rais43

DPL43su

Rural fo

od Engel curve

450

450

Deaton (and Tarrozi)’s method

PL * rR

PLR

PL *UVPCIR

R

u

Rura l food Engel curve

450

450

Urban fo

od Engel curve

Per capita expenditure

Pe

r ca

pita

exp

en

ditu

re

PL * *UVPCIR URR

A

BO

C

D

Suppose we treat Deaton’s method as calculating the urban cost of the food expenditureof rural households’ food expenditure, what should we add as an allowance for non-food?Would it be the non-food share of urban households whose food expenditure was equivalentIn real terms to the the food expenditure of rural households?

What is to be done?• Teach economists ethics – no code of practice! – and get them to practice them

– Honesty, transparency, humility?– Improve capacity for diverse groups to practice evidence based policy– Reduce dependence on powerful donors and their agendas

• Use money-metric poverty for policy analysis more carefully– Constrain domains of comparison– Encourage greater data availability and more critical use of official data (set our data free)– Encourage evidence based policy analysis (and quality data production)– Forget comparability with earlier series (all that intellectual capital!)– Adjust for household type and location– Record value of public goods and environment to comply with Canberra group concept of

income (heavy!) – Triangulate with other indicators (nutrition, health, educational attainments)

• Adopt more sophisticated procedures taking account of the value of services in kind, public goods, the environment, culture, etc.

– Improve survey concepts, methods and procedures, and resources• field survey officials feel undervalued – “kill for a data set”

– Improve Consumer Price Indexes• Don’t ask

• Alternative methods of assessing differences and progress in well-being– Longitudinal studies

• Ensure good practice – can we rely on those who brought us money-metric poverty assesment to do a better job with longitudinal studies?

– Take deliberative and participatory democracy seriously (no media stunts please)