Reaching Poor and Vulnerable Households in Indonesia€¦ · 0 625 1250 1875 2500 000 000 000 000 000 Indonesia per-capita consumption distribution, 2011) Per-capita expenditure (Rp)

Reaching Poor and Vulnerable Households in Indonesia

1Friday, April 13, 12

0

625

1250

1875

2500

Rp

100,

000

Rp

200,

000

Rp

300,

000

Rp

400,

000

Rp

800,

000

Indonesia per-capita consumption distribution, 2011#

of I

ndiv

idua

ls (

thou

sand

s)

Per-capita expenditure (Rp) 2011

Poverty Line

1.5x Poverty

Line

Many households are vulnerable to poverty...


...and half (or more) of the poor in a given year are new poor

0

25

50

75

100

Times poor TImes near-poor

Exposure to Poverty, 2008-2010%

sha

re o

f all

indi

vidu

als

Never poor/near-poorPoor/near-poor oncePoor/near-poor twicePoor/near-poor every year


• Crop failure

• Job loss

• Loss of income due to health problems

• Global crises

• Natural disasters

Disasters are poverty factories:


school

work

family

old age

birth

The ladder of life....

...is filled with opportunities

and risks


0

10

20

30

40

Brazil China Indonesia

Malnutrition (height for age)

% o

f 0-5

yea

r ol

ds

birth

Indonesia’s child malnutrition rate is 5 times higher than Brazil, and 3 times higher than China


0

25

50

75

100

1 2 3 4 5 6 7 8 9 10 11 12 >12

Poorest quintile

Richest quintile

school

Over 80% of the poorest students drop out before

reaching grade 10

Final Education Attainment by quintile, 2010

(26-28 year olds in 2010)


work

92% of all workers are informal or work without a contract

Employers 2%

Permanent Contract

Employees6%

Employees with No Contract

38%

Informal Workers

54%

Profile of Workforce by Job Status


Indonesia Brazil China Malaysia

family

Mothers in Indonesia are

almost 10 times more likely to die

after childbirth than in Malaysia

0

50

100

150

200

250

Maternal Mortality Ratio

per

10,0

00 li

ve b

irth

s


0%

25%

50%

75%

100%

Pension Coverage

With pension

12%

Without pension

88%

old age

Most workers in Indonesia have no pension to rely on when they retire


Why safety nets?

• Help people who can’t help themselves

• Avoid negative coping mechanisms

• Reducing inequality

• Compensation for reforms


PNPM

Safety NetsLabor

SJSN

Reform

Infrastructure Agriculture

Education

SME Health

Safety nets are just one part of the overall

poverty reduction strategy


2012 Extremely Poor Poor Vulnerable

Healthy Families

Continuous Education

Income Protection

Some building blocks are in place...

BLT

BSM

Raskin

PKSA

JSLU JSPACA

Jamkesmas

PKH


...but better engineering and construction are required...

• Filling in the gaps by expanding and extending

• Integrating programs into a single system

• Reforming current programs


... to get from here...


... to here






1. Providing the right benefits at the right time...

• Increasing scholarship levels to match actual costs

• Defining a financially-sustainable Jamkesmas benefit package with facilitation

• Delivering scholarships and PKH in-line with school year

Reforms for all programs


0

240,000

480,000

720,000

960,000

1,200,000

Q1 Q4 BSM-SD Q1 Q4 BSM-SMP Q1 BSM-SMU

Education Expenditure (w/ transport) and BSM amounts

Rup

iah

per

quar

ter

SDSMPSMUBSM (quarterly basis)

1st grade 6th grade 7th grade 9th grade 10th grade

Cash Aid for Poor Students (BSM)

• Provides too little

• Delivers benefits too late

• Does not help


2. ... reaching the right people

• Very little Raskin actually reaches the poor

• BSM scholarships are not prioritized for poor scholars

• Jamkesmas is used more by the non-poor


Subsidized Rice for the Poor (Raskin)

0

10

20

30

40

50

Kilograms (Kg)

Raskin beneficiaries

receive less than 1/3 of what they are promised.

ProcuredRaskin

Delivered Raskin

HouseholdRice

Needs


3. ... delivered in the right way

• Jamkesmas users need comprehensive facilitation

• Households can not apply for BSM

• Raskin rice is not safeguarded and is spread too thinly


Pre-natal care

Blood work

Medicines

Dental services

0 25 50 75 100

% answering “covered” % answering “not covered” % answering “do not know”

Jamkesmas Knowledge: covered Hospital outpatient services

• Most cardholders do not know benefits or how to access.

Health Fee Waivers for the Poor (Jamkesmas)






Filling in the gaps

• New programs cover the remaining serious risks:

Standing shock response system, which may include BLT and a new generation Padat Karya (public works)

Unemployment and old-age addressed through combination of public works and cash transfers for vulnerable elderly

• Expand programs that have promise: PKH, BSM, Jamkesmas, Kemensos initiatives for the marginalized

• Vulnerable households have the highest risk of falling into poverty year after year and need reliable coverage.



Healthy Families


Income Protection

BLT

BSM

Raskin

PKSA

JSLU JSPACA

Jamkesmas

PKH

Filling in the gaps



Healthy Families


Income Protection

BLT

BSM

Raskin

PKSA

JSPACA

Jamkesmas

PKH

Padat KaryaJSLU

Filling in the gaps


Should Indonesia’s SSN include a crisis-response mechanism?


Should Indonesia’s SSN include a crisis-response mechanism?


Indonesia during the post-AFC years...

Shocks develop, escalate, and spread quickly

Food, Fuel Price Shock (BLT I)

-20

-6

8

21

35

Janua

ry 20

03

Janua

ry 20

04

Janua

ry 20

05

Janua

ry 20

06

Janua

ry 20

07

Janua

ry 20

08

Janua

ry 20

09

Janua

ry 20

10

Janua

ry 20

11

Perc

ent

chan

ge, y

ear-

on-y

ear

Food, Fuel Price Shock (BLT II)

Rice Price Shock

Food CPI

Retail Rice PricePoverty

Basket CPI

Price Shocks- Food and Fuel, 2005/6- Food and Fuel, 2008/9- Food 2010/11- Fuel 2012???

Economic Shocks- AFC 1997/8- Global Economic Crisis 2008/9

Natural Disasters- Aceh 2004- Yogya 2006- Padang 2009- Merapi 2010


Crisis response social assistance packages require special characteristics

Adequate

‣ Sufficient protection for households to continue functioning regularly but not too large for labor market disincentives

Feasible & Flexible

‣ Rapid response? Delivered everywhere? Can vulnerable households be prioritized?

‣ Total costs outweigh crisis impact?‣ Can it be short-term?

Politically acceptable


Temporary large-scale cash transfers have more of these characteristics

Flexible duration?

Quickly disbursed?

Sufficient protection?

Well-targeted?

Cost-effective?

Politically acceptable?

Raskin yes not usuallynot

currently no no yes

Jamkesmas noyes, on demand no n/a no yes

BSM no no nonot

currently n/a yes

PKH no no yes yes no yes

BLT yes yes yes can be yes ???

Public Works

yesyes, if on demand maybe

yes, if self-targeted n/a unknown


Was BLT effective at protecting the poor and vulnerable?


BLT 2005/6 and 2008/9• Bantuan Langsung Tunai (BLT) was delivered to approximately 1/3rd

of Indonesian households in both 2005 and 2008. Eligible households in every subdistrict received funds and BLT was by most measures the largest cash transfer in the developing world.

• Initial questions and apprehensions regarding BLT effectiveness remain controversial. Objective, evidence-based assessments have not been well-publicized.

• Nationally representative survey data and qualitative studies can address:

• Did BLT households have sufficient consumption protection?• Did BLT households become “dependent” on handouts?

• Did BLT have labor market disincentives?• Did BLT encourage “unproductive” consumption?

• Did BLT disrupt social capital or create disharmony?


Quick note: methodologies

Fundamental Problem of Evaluation: estimating the Counterfactual - how to determine the most likely outcome had there been no program?

900

950

1000

1050

1100

T=-1 (2004) T=0 (2005, pre-BLT) T=1 (2007, post-BLT)

Before and After Estimates of BLT on agricultural productivity

rice

yie

lds

(kg

per

ha)

Observed change

Counterfactual C?

Counterfactual B?

Counterfactual D?

BLT impact 100 kg/ha?

BLT impact 150 kg/ha?


Quick note: methodologies

• The causal impact of a progam is equal to the value of outcomes when a unit experiences the program minus the value of outcomes when the same unit does not experience the program.

• But only one (of the two states) is observed: In the social sciences, if we observe outcomes after treatment, we cannot simultaneously observe the same units’ outcomes after not getting the treatment.

• Selection bias becomes a serious concern: potential counterfactual outcomes (e.g., from groups not receiving the program) may differ systematically from other units whether or not there had been a program.

• The counterfactual is therefore crucial: simply put, without a valid estimate of the counterfactual, the impact of a program cannot be established.


Effective Methodologies for BLT

0.38

0.55

0.73

0.90

T=0 T=1 T=2

outc

ome

• Difference-in-Differences (DID): compare changes in outcomes (over time) between units in a program (treatment group) and units not in a program (comparison group)

C = 0.78

D = 0.81

A = 0.58

B = 0.74

Impact = 0.13Treatment Group

Comparison group

comparison group trend

• DID advantages: 1st diff. nets out time-invariant features that may condition response. 2nd diff. nets out changes due to features that produced changes even in untreated units.


Effective Methodologies for BLT

• DID disadvantages: any time variant characteristics cannot be addressed with DID alone. BLT recipients will differ from those without BLT in both invariant and variant characteristics.

• Matching procedures identify the set of untreated units that are most comparable, based on observables, to treated units.

• Propensity scores, which are an estimate of the latent likelihood that units would have received the program conditional on all observed characteristics that affect that probability, provide a way to matched treated to untreated households.

• Propensity scoring and matching plus DID try to mimic random assignment to treatment and control groups. Propensity scoring, matching, and DID can recover approximate balance (on pre-program characteristics, both variant and invariant) after program implementation.


Cash Transfers during Crisis (BLT)

Was BLT enough to protect households?

keluarga

3. Especially where overall economic growth was weak or negative, BLT caused higher expenditures and greater equality

1. BLT provided more than enough for families to continue consuming/saving as before the price increases.

2. BLT provided support for long enough for households to adjust to new prices.


Methodologies - Was BLT enough to protect households?

1. “Naive” estimate of household fuel consumption using Susenas data on quantities, (pre-BLT, pre-subsidy reductions) and price data (post-subsidy reductions) to determine cost of continued fuel consumption with no behavior change.

In fact, poor BLT households reduced fuel quantities less drastically - by 4% in 2006 and 34% in 2009 for gasoline - than non-poor households, who saw a17% reduction in 2006 and 57% reduction in 2009 for gasoline quantities.

Despite quantity reductions, total fuel expenditures increased for all households in 2006.

2. From Susenas, calculate fuel shares (in total consumption) just before the fuel-subsidy reductions (in 2005 and 2008) in the cross-section of poor households who received BLT.

As pre-price change fuel shares are approximately constant across time (at just under 9%), BLT did not finance or encourage a shift in equilibrium fuel consumption (relative to total consumption) either up or down.

3. Triple-difference estimate of average changes in ln(consumption) over (a) pre- to post-BLT periods; (b) in BLT and non-BLT households; and (c) in macro-economically weak versus macro-economically strong districts.



3a. LATE - local average treatment effect measured in weak districts relative to strong districts. Weak (strong) districts refers to districts with per-capita expenditure growth equal to the 25th %ile or below (75th %ile or above) of the Indonesia-wide distribution of district expenditure growth.

Inequality in consumption is reduced in those areas where growth was weakest:

BLT Consumption Impact Summary BLT Consumption Impact Summary BLT Consumption Impact Summary BLT Consumption Impact Summary

Average relative % growth of p.c. exp, BLT HH relative to

non-BLT HH

Triple difference (%), weak relative to strong districts

Triple-difference t-stat

2008, poor and non-poor households2008, poor and non-poor households2008, poor and non-poor households2008, poor and non-poor households

weak districts 816 3.3

strong districts -816 3.3

2005, poor households2005, poor households2005, poor households2005, poor households

weak districts 438 2.3

strong districts -3338 2.3



3b. Bazzi, Sumarto, and Suryahadi (2010, hereafter BSS), using reweighting estimators, independently confirm that in the overall ATT sense BLT “did not yield growth among recipients at the same pace as comparable non-recipients.” (BSS measures impacts during the 2005/6 BLT only).

3c. Positive LATE effects in weak-growth areas are encouraging, but lack of overall ATT impacts remains puzzling. Possible explanations include:

• Susenas timing combined with high propensity to consume (out of BLT) for immediate, basic necessities

• use of BLT for debt/asset rebalancing (not captured in Susenas)• local BLT targeting identified those households for whom earnings potential was

lowest• BLT was given to many wealthy households with smaller MPC (out of positive

income shock)• Households smoothed a temporary (ie, not permanent) positive income shock

(BLT) over many future periods.



Did BLT create dependence on handouts? Were BLT households lazy?

3. BLT created more spending and income for the communities at large

1. BLT households found jobs faster and BLT was responsible for net unemployment reductions

2. Households with BLT worked the same number of hours

kerja


Baseline unemployment rates (%, heads of household)Baseline unemployment rates (%, heads of household)Baseline unemployment rates (%, heads of household)

BLT non-BLT

2008, all 12 10

2008, poor & near-poor 12 8

Finding employment by follow-up, relative probability (BLT vs. non-BLT) Finding employment by follow-up, relative probability (BLT vs. non-BLT) Finding employment by follow-up, relative probability (BLT vs. non-BLT) Finding employment by follow-up, relative probability (BLT vs. non-BLT) Finding employment by follow-up, relative probability (BLT vs. non-BLT)

2008, poor & near-poor 2008, poor & near-poor % standard error t-stat

2008, poor & near-poor 2008, poor & near-poor 10.1 5.8 1.8

Becoming unemployed by follow-up, relative probability (BLT vs. non-BLT)Becoming unemployed by follow-up, relative probability (BLT vs. non-BLT)Becoming unemployed by follow-up, relative probability (BLT vs. non-BLT)Becoming unemployed by follow-up, relative probability (BLT vs. non-BLT)Becoming unemployed by follow-up, relative probability (BLT vs. non-BLT)

2008, poor & near-poor2008, poor & near-poor% standard error t-stat

2008, poor & near-poor2008, poor & near-poor-0.1 0.5 0.22

Methodologies - Did BLT create dependency or laziness?

1. Matched DID (ATT estimates) of probability of finding work (losing work) for those unemployed (employed) at baseline (pre-BLT).


Methodologies - Did BLT create dependency or laziness?

Propensity scoring (and matching) done one-to-one, one-to-many, and with kernel densities. No significant changes in point estimates; large common support especially when restricted to poor and near-poor households.

2. Calculate change in total labor hours pre- to post-BLT and determine whether there is a statistically significant difference between BLT and non-BLT households.

BSS, with reweighting estimators (ATT estimates) independently confirm that over 2005 to 2007 the relative difference in the change in total labor hours - based on between 1 and 2 hours fewer work hours for both BLT and non-BLT households - is not distinguishable from zero.

Independent confirmation in interviews with beneficiaries and non-beneficiaries (SMERU, 2006 and 2009) who both indicated that BLT, at 10-15% of regular expenditures, was not enough to live on had to be supplemented with income from production.

3. Regular OLS regression (with spatial heteroskedasticity-robust s.e.) of change in ln(p.c. exp) in all non-BLT households on share of BLT recipients in district population and other district- and province-level controls. Reverse causation (from higher anticipated consumption growth to more BLT recipients) logically prevented by pro-poor targeting and overall lack of BLT impact on consumption growth for the treated.



3. BLT households sought more healthcare services (especially when paired with health insurance, including Jamkesmas)

1. Children from BLT households exited the labor market faster

2. When BLT transfers were timed well (2008), they were used for school fees. Otherwise BLT went immediately to necessities: food, clothing, shelter (not tobacco)

sekolah, kesehatan

Did households with BLT spend productively?


Baseline child labor rates (% of 6-18 yr olds)Baseline child labor rates (% of 6-18 yr olds)Baseline child labor rates (% of 6-18 yr olds)

BLT non-BLT

2008, all 13 9

2008, poor & near-poor 13 10

Child Labor incidence, relative change (BLT vs. non-BLT households) Child Labor incidence, relative change (BLT vs. non-BLT households) Child Labor incidence, relative change (BLT vs. non-BLT households) Child Labor incidence, relative change (BLT vs. non-BLT households) Child Labor incidence, relative change (BLT vs. non-BLT households)

% standard error t-stat

2008, all2008, all -1.0 0.5 2.0

2008, poor & near-poor2008, poor & near-poor -23.0 0.8 2.9

Methodologies - Did BLT households behave responsibly?

1. Matched DID (ATT estimates) of changes in incidence of child labor, measured relative to baseline (pre-BLT).

BSS confirm (for 2005/6) with reweighting estimators that reductions in incidence of child labor (for 7-18 yr olds) are larger for BLT households.



2. Calculate changes in consumption shares, pre- to post-BLT, for BLT versus non-BLT households.

BLT households re-oriented expenditure shares (in the face of new relative prices) similarly to non-BLT households for food and non-food goods. For example, shares of total consumption for education, health, alcohol, and tobacco shares increased by approximately 0.8, 0.3, and 1.1% for BLT and non-BLT households alike between 2008 and 2009.

Independent research confirms that 2005/6 BLT also did not produce a significant increase in education or health expenditure shares (BSS).

Food items - rice, other staples, proteins, dairy, fruits and vegetables, oil, spices, sugar - which all together comprise 2/3 to 3/4 of total poor household expenditure, were substituted at similar rates in poor BLT and poor non-BLT households during the 2008/9 BLT period. Poor households generally switched out of meat and non-rice staples and into more fish, dairy, vegetables, and rice.

Independent confirmation in interviews with beneficiaries and non-beneficiaries (SMERU, 2006 and 2009) who indicated that BLT was spent rapidly on basic necessities - food, clothing, shelter, and transport, and festivities.



3. BSS (with reweighting ATT estimates) and Sparrow, Suryahadi, and Widyanti (with DID methods and baseline controls) confirm a larger increase in outpatient healthcare utilization, by about 0.04 to 0.05 visits per person per month, for BLT relative to non-BLT households during the 2005 BLT. BSS note that the longer-lasting impact of BLT is on outpatient care at public facilities.



3. BLT (and cash transfers generally) did not decrease participation in community- based collective activities, (gotong royong, credit and savings groups, etc)

community social health

Did BLT promote corruption, create disharmony or erode social capital?

2. Complaints and protest activity came from non-beneficiaries; they focused almost exclusively on targeting and benefit distribution

1. Later BLT tranches saw increases in deductions, informal levies, and redistribution: fewer households received correct amounts.


Methodologies - Corruption, Disharmony and Social Capital

BLT Deduction SummaryBLT Deduction SummaryBLT Deduction SummaryBLT Deduction Summary

2005/62005/62008/9

1st tranche 2nd tranche2008/9

Frequency of any deduction (%) < 10 ~ 10 ~ 50*

Mean amount deducted (Rp ‘000) 53 72 67

Median amount deducted (Rp ‘000) 20 60 50

Mode amount deducted (Rp ‘000) 10 100 100* 2008 deduction frequency is calculated in two ways: 1) as the percent of BLT beneficiaries receiving less than the stipulated amount (46%) based

on which target amount - one tranche or two tranches - their “amount received” answers are close to; 2) one minus the percent of beneficiaries who answer “No” to all questions asking about deductions made by different actors, including a catch-all “other” actor category (54 %).

* 2008 deduction frequency is calculated in two ways: 1) as the percent of BLT beneficiaries receiving less than the stipulated amount (46%) based on which target amount - one tranche or two tranches - their “amount received” answers are close to; 2) one minus the percent of beneficiaries who answer “No” to all questions asking about deductions made by different actors, including a catch-all “other” actor category (54 %).



1. Use Susenas panel and ask recipients how much they received; compare answers to Rp 300,000/quarter stipulated benefit

Most commonly deducted by village (or lower) administration. Most who experienced them (40-60%) claim deductions were made to redistribute funds equally.


Methodologies - Corruption, Disharmony and Social Capital

2. Use IFLS social assistance module to tabulate IFLS-wide mean.

81% of all complaints were from non-beneficiaries, and over 86% of complaints were for one of the following related reasons: “Eligible beneficiary listing and selection was not transparent”, “Unfair distribution of benefits” and “Assistance distributed to those ineligible”, and “Nepotism in the distribution of funds”.

3. Use administrative data and other second-hand data sources to match changes in community- or social-activity participation rates with coverage/extent/concentration of BLT coverage at both household and regional levels (DD estimation).

Social and Community Participation Rates and cash transfersSocial and Community Participation Rates and cash transfersSocial and Community Participation Rates and cash transfersSocial and Community Participation Rates and cash transfers

data source:

Activities with observable participation rates

Participation by poor/non-poor status available?

Change in participation correlated with cash transfer coverage?

IFLS community planning mtgs, voluntary labor et alia yes* no BLT correlations

PKH IE gotong royong (labor and cash/in-kind); religious groups; credit groups; social service groups et alia

no, but PKH/no-PKH status observed

Cash transfers increased participation in credit and

social service groups

PNPM MIS PNPM meetings yes* no BLT correlations* IFLS data contains per-capita consumption and relatively poor (not officially poor) individuals from the IFLS-wide consumption distribution

can be distinguished. In PNPM MIS data individuals are given a poverty status according to PNPM practices; as for IFLS, an individual’s PNPM-determined status does not necessarily match her official GOI poverty status.

* IFLS data contains per-capita consumption and relatively poor (not officially poor) individuals from the IFLS-wide consumption distribution can be distinguished. In PNPM MIS data individuals are given a poverty status according to PNPM practices; as for IFLS, an individual’s PNPM-determined status does not necessarily match her official GOI poverty status.





3. New targeting tools with improvements in data collection, methodology, and appeals are already available

1. BLT targeting was better than in other Indonesian social assistance transfers

addressing vulnerability

Did BLT reach the poor and vulnerable?

2. Even so, targeting and other support operations show room for improvement on both exclusion and inclusion errors


Methodologies - Did BLT reach the poor and vulnerable?

1. Use Susenas to determine which households received which social assistance programs.

Calculate %age of benefits received by officially-targeted households (usually the poorest 30%) and compare to outcomes if targeting had been random (represented by zero) or perfect (represented by 100)

0

10

20

30

40

50

BLT 2005/6 BLT 2008/9 Jamkesmas 2010 Raskin 2010

Actual Program Targeting

% im

prov

emen

t ov

er r

ando

m t

arge

ting

(0%

)


Methodologies - Did BLT reach the poor and vulnerable?

2. Use Susenas to calculate %age of target households (poorest 30%) that were excluded from BLT (exclusion error) and the %age of non-target households that were included (inclusion error)

0

20

40

60

80

BLT 2005/6 BLT 2008/9

Exclusion and Inclusion errors in BLT

Perc

enta

ge o

f HH

exc

lude

d (in

clud

ed)

Exclusion Error among poorest 30%Inclusion Error among richest 70%


• Support operations and procedures - from targeting through to complaint resolution - show much room for improvement.

• Monitoring and evaluation or a complaints and grievance system were not initiated - the frequency and size of deductions increased from 2005 to 2008.

• Targeting objectives were not well understood - communities were frustrated by a lack of transparency, irregularities, and final BLT allocations that seemed unfair. BLT-related complaints and protest activity focused overwhelmingly on targeting and benefit distribution procedures and outcomes.

• BLT was ad hoc, exposed to politics - Opposition politicians resist ad hoc transfers if they might be used to “buy votes”. Without automatic procedures for turning BLT on and off (when a shock hits and dissipates), this charge is always plausible.

Nonetheless, BLT implementation needs strengthening


Reaching Poor and Vulnerable Households in Indonesia€¦ · 0 625 1250 1875 2500 000 000 000 000 000 Indonesia per-capita consumption distribution, 2011) Per-capita expenditure (Rp)

Documents