Documente Academic
Documente Profesional
Documente Cultură
=
N
n
,
Where N is the total number of districts, and n is the number of districts selected.
The Second Stage (household) Weight
o Since the subsampling of households occurred within sampled districts, then the basic second stage
weight is defined as:
Ew
=
k
,
Where k
i
is the number of households in the selected district and s
i
is the number of
sampled households within district i.
The Final household Weight
o Therefore the final household weight is defined as:
w
= w
Ew
=
N k
n s
,
21. As a welfare variable, the adult equivalent (AE) expenditure used was measured by the total expenditure of
the household divided by the household specific adult equivalence scale. The total household expenditure is defined
as the sum of cash expenditures on: (a) food and the estimated value of in-kind receipts or own production
consumption in the past 7 days; (b) fuel for cars/motorcycles and on public transportation in the past 7 days; (c)
prepared meals and drinks outside the home, housing, tobacco, utility bills, medicine, health insurance, clothing,
shoes and hygiene products in the past month; (d) large medical expenses (e.g., hospitals, etc.), car repairs, and
recreational/cultural activities in the past six months; and cash expenses on (e) education and school fees, licensing,
property tax and other general expenses in the past year. For the adult equivalence scale the PCBS formula was
used:
AE = (#oJult + u.46 #cbilJrcn)
0.89
where #adult is the total number of adults age 18+ living in the household; and
#children is the total number of children age 0-17 years old living in the
household.
22. The interaction between the last payment and data collection was explored to evaluate whether a
pre-transfer household expenditure was more appropriate to be used in the targeting assessment. Almost all
households declared receiving their last payment from the CTP in December 2011, and a large majority
immediately spent the transfer on food. Therefore, it was decided to use the current household
consumption/expenditure without adjustment, that is, the total household expenditure was used without subtracting
the transfer amount received by beneficiaries, given that the survey data were collected in March-April 2012
(almost three months after the last payment).
13
23. Classification of a household as extremely poor or poor was done in two ways. First, each household has its
household expenditure compared to the extreme poverty line (NIS 1,898.72 a month for a family with 2 adults and 4
children) and to the poverty line (NIS 2,278.47 a month for the same type of family). These two values of the
poverty lines are expressed in 2009 prices. Therefore, since the data were collected in March-April 2012, the PCBS
consumption price index (CPI) was applied to have the lines presented in 2012 values. Households with total
expenditure below the extreme poverty line were classified as extremely poor, households with total expenditure in
between the extreme poverty line and the poverty line were classified as poor; and households with total
expenditure above the poverty line are classified as non-poor. Second, each household was previously classified by
MoSA according to their PMTF score. A PMTF score below 6.39 classifies a household as extremely poor; a PMTF
score in between 6.39 and 6.57 classifies a household as poor; and PMTF score above 6.57 classifies a household as
non-poor.
24. The next sections will focus on the PMTF MoSA classification of extremely poor households and compare
it with the consumption poverty levels.
I V. Assessment tools
25. Before beginning the assessment of the formula, it is useful to review targeting. Targeting is a tool to
channel resources into the hands of those in need to achieve greatest impact for a given budget, since it helps reduce
a waste of funds. Targeting helps to concentrate programs on the desired population and even in budget-constrained
settings it allows the Government to use the available, but limited, resources most efficiently.
26. But targeting incurs costs. There are, for example: administrative costs to manage the program, gather
information of potential beneficiaries, monitor distribution of resources, etc.; private costs as some applicants may
have financial, time or mobility constraints in going to a place and applying for the program; incentive costs since
some applicants may change their behavior in an attempt to become beneficiaries; and social costs because
beneficiaries can be stigmatized or politicians can decide to use the program in their own interest affecting the
program administration.
27. Targeting also incurs errors. A limited budget affects the ability of the Government to ensure outreach to all
individuals who constitute the desired population leaving a significant share of this population without access to the
program. Gathering applicants is costly and time-consuming and is vulnerable to information errors if enumerators
or social workers are not well-trained. Regardless of the choice of targeting method, two types of errors will always
occur:
Error of Inclusion: Not-targeted population selected to participate in the program; and
Error of Exclusion: Targeted population not selected to participate in the program.
14
28. In summary, targeting is an important tool that needs to be evaluated since it has significant cost and errors.
Usually more attention is paid to errors forgetting that the errors can come from different sources and not only from
the selection method. Main source of errors can be seen in Figure 2 below.
Figure 2: Main source of targeting errors
Budget
Outreach
Method
Private costs
29. To address the quality of the targeting different indicators will be looked at such as:
Beneficiary incidence
o Percentage of program beneficiaries relative to the total number of beneficiaries by
decile/quintile of the post-transfer welfare distribution
Benefit incidence
o Percentage of program benefits relative to the total budget allocated by decile/quintile
of the post-transfer welfare distribution
Cost-benefit ratio
o A NIS 1 reduction in the poverty gap for each NIS 1 spent in transfers.
30. Additionally, a sensitivity-specific analysis adapted from Hoddinott et al. (2002) will be performed, which
systematically examines how a model works with reference to the benchmark variable based on a 2 by 2 cross-table
computed as follows:
Cross-tabulation for sensitivity-specificity analysis
Current Household Consumption
Extreme Poor (Y Extreme poverty line) Non-Extreme Poor (Y > Extreme poverty line)
PMTF groups
Selected
PMTF 6.39
True positives False positives
a
Not selected
PMTF > 6.39
False negatives
b
True negatives
31. From this table, the following indicators are computed:
Sensitivity (Coverage) the proportion of Extremely Poor households selected by the PMTF, that is the
number of true positives divided by the number of Extremely Poor households.
15
Specificity the proportion of non-extremely poor households also selected by the PMTF, that is the
number of true negatives divided by the number of all non-poor.
Exclusion (Undercoverage)- the proportion of Extremely Poor households not selected by the PMTF, that is
the number of false negatives divided by the number of Extremely Poor households (100% - Coverage)
Inclusion - the proportion of non-extremely poor households not selected by the PMTF, that is the number
of false positives divided by the number of non-extremely poor households selected by the proxy.
Positive predictive value the proportion of households selected by the PMTF that are Extremely Poor, that
is the number of true positives divided by the number of all those classified as such by the PMTF.
Misclassification the proportion of cases that do not match, that is, is the sum of false positives and false
negatives divided by the entire number of households.
32. Based on these indicators a good proxy indicator strongly associated with welfare levels will have: high
sensitivity (coverage) that implies low errors of exclusion; high specificity that implies that non-poor households
are not classified as poor which implies low errors of inclusion meaning that few selected cases are non poor, high
positive predictive value, and low misclassification. There is gold no standard by which to compare the
achievements of this targeting mechanism but this set of indicators will highlight good features of the model.
V. Findings
33. Based on the sample and using expansion factors (weighted estimates to represent the full set of 110,000
applicants) extreme poverty levels is estimated at 68.7%, meaning that a large majority of MoSA applicants are
extremely poor. Around 10.3% has consumption levels in between the extreme poverty and poverty lines and only
21% of applicants had an adult equivalent household expenditure above the poverty line. Breaking down these
estimates by region (West Bank and Gaza) the study found that 83% of Gaza applicants are extremely poor
compared to 56% in West Bank, reinforcing the message that the prevalence of poverty is higher in Gaza and that it
is more difficult to identify an extremely poor household in the West Bank. In other words, larger errors are
expected in the West Bank.
V.a. I mpact on poverty
34. The overall budget needed to eradicate extreme poverty of applicants is estimated at NIS 87.7 million a
month, while NIS 61.5 million of this budget should be allocated to Gaza given that it has a larger poverty gap,
42.7% compared to 17.78% in the West Bank (the term larger poverty gap implies that people are, on average,
farther away from the poverty line). However, MoSA had transferred, NIS 29.9 million a month (one-third of the
budget needed) in the region, from which NIS 16.6 million of this amount went to Gaza, for the CTP and Vulnerable
population program. Therefore, MoSA allocations are far from optimum for two reasons: first, the program
transferred one-third of the budget needed to eradicate extreme poverty in Gaza and the West Bank; second, despite
budget allocation in the region, targeting errors were expected in both regions.
35. Then, in order to measure the CTPs efficiency in reducing extreme poverty, the poverty reduction and the
cost-benefit ratio were estimated. For the pool of applicants, transfers can immediately reduce the extreme poverty
headcount and extreme poverty gap ratio. In Gaza, extreme poverty would be reduced by 9% and the poverty gap by
21%, compared to a 20% reduction of extreme poverty in the West Bank and a 33% reduction in the poverty gap.
16
Comparing the CTPs impact on poverty with international experience, it can be concluded on the basis of these
data that the CTP is performing quite well.
36. As an international benchmark, the ASPIRE
3
data can be used, which compiles information for more than
60 countries. ASPIRE shows that poverty reduction for the poorest 20% population in flagship programs such as the
Brazilian Bolsa Famlia is 6% with a decline in the poverty gap of 14%. In terms of the impact among
beneficiaries only, Skoufias and Di Maro (2010) showed that the Mexican PROGRESA (former Oportunidades
program) reduced poverty among beneficiaries by 16.5% and the poverty gap by 24.3%.
37. These reductions in poverty and the poverty gap are translated into a cost-benefit ratio where the current
impact of the program can be measured in terms of the poverty gap reduction for each NIS 1 spent in transfers by
MoSA. For the CTP, this analysis found that NIS 0.66 out of each NIS 1 spent in benefits goes to extreme poverty
gap reduction. Breaking it down by region (West Bank and Gaza), in Gaza the cost-benefit ratio is NIS 0.77 while in
the West Bank it is NIS 0.52. The fact that the impact on poverty is higher in the West Bank but the cost-benefit is
lower indicates that selected beneficiaries in the West Bank are closer to the extreme poverty line than those in
Gaza.
38. Comparing these numbers with the ASPIRE data, which estimate the cost-benefit ratio of cash transfer
programs (both conditional and unconditional) for the poorest 20% population, it is evident that the program
indicator is among the best practices (see Figure 4).
Source: ASPIRE 2011, authors calculation
3
ASPIRE: The Atlas of Social Protection Indicators of Resilience and Equity. Downloadable at www.worldbank.org/spatlas
17
V.b. Sensitivity-specific analysis
39. The findings on poverty are a good indication that the overall targeting approach of MoSA that combines
careful selection and revision of cases by social workers and a PMTF to rank applicants by a predicted welfare
levels is quite efficient. To better understand how this process works, it is important to spend some time on the
analysis of the PMTF model alone using the sensitive specific analysis to discuss the magnitude of errors and
predictive quality of the current PMTF model
4
.
40. In Table 2 the analysis for the full set of applicants is presented. It is a good matching (true positives (TP)
and true negatives (TN)) model identifying correctly almost 70% of the cases (sum of main diagonal divided by
total number of cases or 100% minus misclassification). As a result of such a high level of correct matching, the
current PMTF model has significant sensitivity, 76%, and 53% of specificity that indicates that about one-half of the
non-extremely poor are correctly classified as non-extremely poor. In addition, exclusion and inclusion errors are
around 20% and positive predictive value is estimated at 78%. All of these indicators imply that the PMTF model is
fairly predicting the extremely poor population and that is strongly associated with actual consumption.
Table 2: Sensitivity analysis
Source: Authors calculation
41. Breaking down the indicators by region and selecting only refugees, the main findings hold (see Table 3).
However, it is important to note the low sensitivity in Gaza and larger inclusion and exclusion errors in the West
Bank.
42. Low specificity means that a large majority of the non-poor households in Gaza are classified as poor.
However, since the number of non-poor households in Gaza is small, the low specificity does not translate into large
inclusion errors.
4
Current PMTF model was designed on the basis of the PECS 2007 data and it has a unique set of weights for both the West
Bank and Gaza.
18
43. Larger inclusion and exclusion errors in the West Bank were expected. In the West Bank, the prevalence of
extremely poor is estimated at 8% according to the World Bank Poverty Assessment (2011), meaning that it is hard
to identify the extremely poor among the population. However, the errors, 39% and 31%, are just a reflection of the
difficulty in finding the extremely poor population. Notice on the other hand that sensitivity is estimated at 61% and
specificity at 65%, meaning that the extremely poor (sensitivity) are likely to be classified as extremely poor and
that the non-extremely poor (specificity) are likely to be classified as non-extremely poor. In other words, the
models ability to correctly classify is quite significant.
44. As a strategy to reduce errors, MoSA has a verification process which explores PMTF findings and
provides an opportunity to bring some households previously excluded by the PMTF into the CTP case load. Some
cases are classified as vulnerable and for those a concept of poor, not extremely poor, is considered. By adding
this verification element and using the poverty line instead, Table 4 shows that sensitivity (coverage) increases
while inclusion and exclusion errors decrease significantly, mainly in the West Bank.
Table 3: Summary of sensitivity analysis
Source: Authors calculation
Table 4: Summary of sensitivity effect MoSA verification
Source: Authors calculation
45. In summary, the main findings of this sensitivity analysis are very encouraging because international
benchmarks of key and flagship programs around the world have sensitivity (coverage) levels of around 50% for
cash transfer (conditional or unconditional) programs and even family allowances (see Table 5). Only in Chile
which also provides in-kind transfers does the coverage for the poorest 20% reach about 90%. In the next section
other findings are presented that corroborate the main study results presented here.
19
Table 5: Sensitivity (coverage) of flagship programs for the poorest 20% population
Source: ASPIRE 2011
V.c. Beneficiary (Benefit) I ncidence Analysis
46. Another question that is often asked about a program like the CTP is: who received the benefits from this
program or where are the beneficiaries of this program? This question was somehow answered previously when
findings on the Programs impact on poverty were presented. But in this section the overall welfare distribution is
explored to measure the seriousness of exclusion and inclusion errors in targeting. That is to say, does a 31%
inclusion error in the West Bank imply that these households are wealthy or are they around the extreme poverty
line? In other words, while this study is interested in determining whether the extremely poor or non-extremely
poor households are covered by a program, it is necessary to look at the benefit incidence of the program across the
welfare distribution, as a finer analysis. By calculating the share of benefits going to deciles of the welfare
distribution, which gives a finer grid than the poverty analysis, a benefit (beneficiary) incidence analysis is
provided.
47. It should be noted in Figure 4 that among the applicants 70% of those classified as extremely poor by the
PMTF are among the poorest 10%, and 84% among the poorest 20% of the population. Looking separately at Gaza
and West Bank, there is a better benefit incidence in Gaza as already mentioned but even in the West Bank around
60% of applicants classified as extremely poor are indeed among the poorest 10% and 78% are among the poorest
20% of the population. Notice that beyond the poorest 20% the percentages are quite small and basically the
wealthier deciles are not eligible. In Figure 5, the benefit incidence analysis is presented to explore the amount of
benefit received in each group of the welfare distribution. For the West Bank, more than 50% of resources allocated
to those classified as extremely poor by the PMTF are going to the poorest 10% and 71% to the poorest 20%. These
incidence indicators are quite large because the international figures for the poorest 20% population incidence
analysis summarized in Figures 6 and 7 are on average between 30-40%. Therefore, in terms of benefit incidence,
the CTP is among the worlds best targeted programs.
20
Source: Authors calculation source: Authors calculation
Source: ASPIRE 2011; Authors calculation Source: ASPIRE 2011; Authors calculation
VI . Conclusion
48. This assessment evaluated the CTP which is the flagship program operated by MoSA to mitigate poverty in
WBG. The CTP was launched in mid-2010 as a result of the merger of the two main cash assistance programs in
place at that time: the SHC Program financed by the EU and the SSNRP financed by the World Bank. The CTP
selects as beneficiaries extremely poor households according to a PMTF model that predicts the welfare of each
applicant household. Based on a sample of applicants for the CTP model, the actual consumption level of applicants
was compared with the PMTF score to determine the effectiveness and accuracy of the PMTF model.
49. The main finding of the assessment is that the large majority of applicants classified as extremely poor by
the PMTF are indeed extremely poor. Results differ across the two regions, Gaza and West Bank, but in general the
21
model is generating indicators that are among the best in the world. The CTP targeting approach reflects its
objectives of reducing the poverty gap with its emphasis on assisting extremely poor households.
50. Both the poverty rates and poverty gap can be reduced significantly if transfers are made regularly to
families. Poverty among current applicants would drop by 10-20% and the poverty gap by 20-30%. In addition, the
efficiency of the program measured by the cost-benefit ratio shows that the CTP is quite effective because for each
NIS 1 spent in transfers by MoSA, NIS 0.66 goes to reduce the extreme poverty gap (NIS 0.77 in Gaza and NIS 0.52
in the West Bank).
51. In terms of model accuracy, the PMTF correctly identified almost 70% of the cases. As a result of such a
high level of correct matches, the current PMTF model has significant sensitivity (coverage) (76%) for the
extremely poor. Exclusion and inclusion errors are around 20% which are below those errors found in analyses of
important programs that are widely considered successful such as Mexicos Oportunidades and Brazils Bolsa
Familia. Given the lower levels of extreme poverty in West Bank, errors are larger but they do not affect the CTPs
efficacy. The current MoSA strategy of verification that brings some households that are not classified as extremely
poor by the PMTF into the Program under the vulnerable classification helps MoSA to reduce errors.
52. Errors are indeed minor and do not affect the Programs effectiveness. The beneficiary (benefit) incidence
analysis findings shows that more than 70% of those classified as extreme poor (67% of benefits of current
beneficiaries) goes to the poorest 10% of the population, which is quite a high level when compared to international
experience.
53. Therefore, the current PMTF is efficiently identifying the extremely poor households but as with any model
there is room for improvement. Despite good overall findings, the current model can be improved to reduce errors
but more importantly than redefining the current model, a more in-depth analysis of this current dataset can provide
more information to guide such revisions. MoSAs verification strategy is efficient and it could be that improving
this process would lead to even better results.
54. Given that the CTP targeting mechanism is largely effective and has targeting accuracy similar to the best
programs in the world, extending coverage to poor households would likely yield large efficiency gains. First, by
extending outreach to ensure inclusion of extremely poor in the poverty-targeting database, information on such
households can be shared with other relief and development organizations to further coordinate efforts avoiding
duplication and providing support to the poorest populations. Second, a larger portion of total social assistance can
be directed to the poorest of the poor. For example, prior to the National CTP it was found that 40% of the poor
were not receiving any formal assistance (Palestinian Household Expenditure and Consumption Survey (PECS),
2007). Further, analysis of leakages to the non-poor in the distribution of amounts of benefits shows that these
amounts were not insignificant. Half of the monetary value of all social assistance accrued to the non-poor, and
almost four in every ten dollars were captured by the richest 40% of the population. For MoSA, at the time this
figure was about 35 percent, providing some indication that there is potential for efficiency gains in extending CTP
coverage as for every two dollars distributed through MoSAs assistance only one dollar went to the poor (World
Bank Poverty Assessment, 2010).
55. These study data also indicated that the CTP processes are effective, though there is some room for
improvement in terms of communication to ensure information about the Program reaches the poorest of the poor
22
and to ensure that adequate and accurate information is provided to CTP applicants and beneficiaries (e.g., on
targeting, procedures, complaint mechanisms, etc.). Further, while the study found that timely payments are
received in a majority of cases, some delay was experienced (for example, 15% of respondents in the West Bank
reported delays) in receiving the cash transfer from the CTP. It was emphasized, moreover, that the time lag
between applying and receiving assistance is significant and that this might be an area where the process might be
modified to reduce this time lag. Finally, in terms of priority spending, it was noted that the majority of households
identified food as the top priority expenditure for the cash transfer they received and did not save or invest the cash
transfer.
VI I . Recommendations
56. The following policy recommendations are made on the basis of the study findings:
Given that this targeting assessment demonstrated the effectiveness and efficiency of the PMTF, no
modifications should be made to the PMTF until further analysis is carried out in the follow on technical
assistance (TA) which will look at issues related to the energy affordability, food consumption, etc. It is still
necessary to:
o Explore further the target population (extremely poor households) using information obtained from
EMCC and Al Markaz for Development and Marketing Consultancies through this survey.
o Verify whether such characteristics are presented in the PMTF and whether the current model
identifies this population.
o Explore current data to understand the cost of having a person with disability in the household
(requires use of PECS data). Data from Gaza indicates that approximately 30% of the households
sampled for this study had one household member who was disabled; and
o Explore characteristics of households that were erroneously excluded or included.
Explore special price differences between the West Bank and Gaza to evaluate whether cut-off points
should be different.
Explore the role of the PMTF in targeting for other social protection programs. For example: Do those
households identified as extremely poor by the PMTF also need energy subsidies? Are these households
also food insecure?
Explore further the use of the Food Consumption Score indicator to complement the PMTF given the high
correlation coefficient between PMTF and Food Consumption Score in comparison to the coefficients
between PMTF and other indicators.
Design training modules for MoSA muderia staff on the CTP targeting approach.
Improve the appeal/grievance mechanisms to deal with mismatch cases.
Use, always, a cut-off point for assessing eligibility, even for the cases classified as vulnerable.
Explore the predictive power of any changes to the PMTF model on the targeting database.
Explore the possibility of applying different models for the West Bank and Gaza in determining the PMTF
for CTP eligibility
Address and explore in further detail other key issues relevant to modifying the CTP approach and
appropriately targeting and effectively assisting the poor, specifically looking at:
o The impact of the energy tariff on poor households from the data collected through this survey to
understand characteristics of these households, financial burden and consider how this might be
addressed to support poor households.
23
Address the time lag between applying and joining the CTP and improve outreach and communication
related to the CTP.
57. In light of study findings the Bank will continue to provide technical support to identify potential
approaches to further enhance/improve the CTP's targeting mechanism that should only be geared towards the
chronic poor while looking into effective ways of diverting the vulnerable and less poor to non-cash assistance
programs. Further, in order to promote the development of a comprehensive and collaborative approach to social
assistance in WBG, the Bank will continue to foster dialogue and collaboration among and between interested
partners in technical discussions related to formula revisions and by identifying actions that can be taken to ensure
that all partners can benefit from this important research.