Effectiveness of cash transfer programmes for household welfare in Pakistan: the case of the Benazir Income Support Programme.
Durr-e-Nayab ; Farooq, Shujaat
Cash transfer programmes are widely considered a 'magic
bullet' for reducing poverty. Whether they actually have such an
incredible impact on poverty reduction is debatable but they surely are
gaining credibility as an effective safety net mechanism and
consequently an integral part of inclusive growth strategies in many
developing countries. As shown by Ali (2007), inclusive growth rests on
three basic premises. First, productive employment opportunities should
be created to absorb labour force. Second, capability enhancement and
skill development should be focused in order to broaden people's
access to economic opportunities. And lastly, a basic level of
well-being has to be guaranteed by providing social protection. Safety
nets are at the core of the last pillar, provided mainly through cash
transfers, which can be both conditional and unconditional.
The basic rationale behind the social safety nets is to assist the
poor to better manage risk and help them to adopt a strategy that
protects their assets. The importance of these safety nets has been
recognised not only for their social and economic value but also as a
means to improve political stability and control crime and riots. These
safety nets help people through short-term stress and insecurities,
which if properly managed can lead to long-term poverty alleviation as
well. Direct transfers by the government are a common means of providing
safety net to the poor. Such transfers include direct provision of food
or cash (conditional or unconditional) to the target population. Other
means of providing safety nets include: education and health subsidies;
energy, water and housing subsidies; and public works programmes. It is
worth mentioning here that, although usually used interchangeably, there
is a need to differentiate between the term social protection and social
safety nets [Bari, et al. (2005); Sayeed (2004)]. Conceptually,
analytically and by implications, social protection is a right that
every citizen must have while safety nets are instruments employed to
provide these rights. (1)
Pakistan is going through a rather prolonged phase of stagflation
making the provision of social safety nets all the more important. Even
during the periods of high economic growth the 'trickle-down
effect' did not essentially take place, necessitating the need to
introduce safety nets in the overall poverty alleviating strategies. A
variety of safety net programmes exist in the country but to mitigate
the situation resulting from the low economic growth and high inflation,
especially food inflation, the government of Pakistan launched the
Benazir Income Support Programme (B1SP) in 2008. The enrolled households
under BISP are paid an amount of rupees one thousand per month, without
any conditions attached to them. The findings of this study, as would be
seen latter, show that this cash transfer does provide relief to the
recipient households but the program does have issues of targeting and
exclusion.
The present paper aims to evaluate the effectiveness of the BISP in
sustaining a recipient household's welfare in the face of
prevailing tough economic conditions. In the sections to follow the
paper would look into the: safety net programmes functioning in the
country and the background to the BISP; data source and methodology
employed; the evaluation of the BISP as an effective safety net
initiative; and conclusions drawn from the discussion and policy
recommendations.
BISP AND OTHER SAFETY NET PROGRAMMES IN PAKISTAN
Pakistan is one of the very few developing countries that
guarantees social security of its citizens in the Constitution. Article
38, 'Promotion of social and economic wellbeing of the
people', in its clause c and d states, "The state shall:
provide for all persons employed in the service of Pakistan or
otherwise, social security by compulsory social insurance or other
means; and provide basic necessities of life, such as food, clothing,
housing, education and medical relief, for all such citizens,
irrespective of sex, caste, creed or race, as are permanently or
temporarily unable to earn their livelihood on account of infirmity,
sickness or unemployment'' [Constitution of Pakistan (2010)].
Whether this commitment is actually fulfilled in spirit is a
separate debate but a whole range of safety net programmes has been
initiated in the country over the years. Discussion on all these
initiatives is beyond the scope of this paper, as it focuses on the
BISP. However, a summary of the safety net programmes functioning in the
country is presented in Table 1. For a useful discussion on safety net
programmes operating in Pakistan see, Jamal (2010), World Bank (2007),
Arif (2006), Irfan (2005), Bari, et al., (2005).
The findings of the studies carried out to evaluate the functioning
of the various safety net programmes have a general consensus that these
programmes are having a positive impact but their effectiveness can be
significantly improved. These programmes are hindered by issues related
to coverage, targeting, and implementation [Bari, et al. (2005); World
Bank (2007)]. These programmes aim at improving their accessibility to
the poor devising means to encourage poor to move out of poverty
permanently, and improving social security in the larger context as
well. Other issues characterising these safety net programmes are:
duplication, overlap, lack of inter-organisational coordination and
fragmentation, which need to be tackled for a greater impact of these
social initiatives. (2)
The BISP, as stated earlier, was initiated in 2008 by the
Government of Pakistan with the immediate objective of mitigating the
impact of rampant inflation, especially food and fuel inflation, faced
by the poor. Over the years the BISP has become the main safety net
programme in the country having maximum numbers of beneficiaries among
all public initiatives. By the end of the third quarter of the financial
year 2011-12, the BISP covered over four million recipients nationwide
with over Rs 122 billion disbursed among them [Ministry of Finance, MoF,
(2012)]. The programme envisaged spreading its reach to seven million
people nationwide by the end of the financial year 2011-12.
At the start of the BISP, in the absence of data for the
identification of the underprivileged, the parliamentarians were
entrusted with the task of identifying the deserving people in their
constituencies to be provided relief. A simple application form, along
with the eligibility criteria, was given to the parliamentarians at both
provincial and national level to identify the underprivileged and needy
in their constituencies [Khan and Qutub (2010)]. With time and in the
face of criticism from opposition parliamentarians, however, a more
scientific procedure was adopted. The eligible households are now
identified through a survey and application of Proxy Means Test (PMT)
formula. The PMT procedure estimates the welfare status of a household
on a scale of 0 to 100 helping in identifying the poorest households
[MoF (2012)]. For the application of the PMT formula, a nationwide
Poverty Scorecard Survey was conducted in 2010 covering around 27
million households in the country. To increase the accuracy, objectivity
and replicability of the survey, GPS readings were also taken, which
also helped in devising coping strategies for natural disaster. After
conducting this survey the eligibility criteria for households to
receive the monthly cash transfer from the BISP was redefined and is as
follows:
(1) The Proxy Means Test (PMT) score of the household is 16.17 or
lower
(2) All the married women within a household are beneficiaries whom
PMT score is below the cut off point
(3) Woman is holder of a computerised national identification card
(CNIC) from NADRA. (3)
The BISP is being implemented in all four provinces of the country
(namely, Punjab, Sindh, Balochistan and Khyber Pakhtunkhwa), the
Federally Administered Tribal Areas (FATA), Azad Jammu and Kashmir (AJK)
and the Islamabad Capital Territory (1CT). The eligible households,
through their females, receive a monthly cash transfer of Rs 1000, which
for a poor family with a monthly income of Rs 5000 is an increase of 20
percent, which equals to 12 percent of the minimum wage in Pakistan. It
is worth mentioning here that the BISP cash assistance amount is
equivalent to 60 percent of the 2010 official poverty line in Pakistan.
Initially the payments to the BISP selected households were made through
the Pakistan Post, which paid the money to the recipients at their
doorstep. To increase the transparency of the programme, and reduce any
possible pilferage, the BISP is adopting more technology based solutions
such as: Benazir Debit Cards, which can be used as ATM cards by the
recipients withdrawing the cash payment every month; Smart Cards,
authorized by a commercial bank; and Phone to Phone Banking, by
providing free mobile phones and SIMs to beneficiaries for the transfer
of monthly cash assistance [MoF (2012)].
What comes as a relief regarding the design of the programme is the
building in of various graduation initiatives helping the recipient
households to exit from the poverty trap. Starting as a solely cash
transfer programme, the BISP has been redesigned in 2011-12 to launch
various initiatives in order to add a sense of permanence to the
benefits gained by the recipient households [BISP (2012)]. Each of these
new programmes has been initially launched in a few selected districts
of the country with the aim to spread them nationwide. Some of these
initiatives include the: Waseela-e-Haq micro-finance programme,
providing soft loans up to Rs 300,000 for setting up small businesses,
to households randomly selected by computers on monthly basis;
Waseela-e-Rozgar programme under which one member of the selected
household is provided technical and vocational training to sustain his
livelihood; Waseela-e-Sehat programme providing life insurance cover of
Rs 100,000 to the breadwinner of the selected households; and
Waseela-e-Taleem in which primary education is imparted to the children
of the recipient households [MoF (2012); B1SP (2012)].
It may be mentioned here that this paper restricts itself to the
cash transfer programme carried out under the BISP initiative on the
whole. Literature voices a strong concern about creating a dependency
among households receiving such cash transfers [Kunnemann and Leonhard
(2008); IBRD (2009)]. Dependency, as expressed by Samson (2009: 46)
implies that, "the choice by a social cash transfer recipient to
forego a more sustaining livelihood due to the receipt of the cash
transfer". Worldwide evidence, however, suggests otherwise. Studies
conducted in a vast number of developing countries including Brazil,
Mexico, Kenya and Zambia, analysing the impact of the BISP-like cash
transfers have found that workers in households receiving such cash
transfers look for employment more intently than comparable poor
households not receiving any such cash assistance [Samson (2009); Posel
(2006); Kunnemann and Leonhard (2008); Samson and Williams (2007);
Barrientos (2006); and Kidd (2006)].
Another factor, which needs our attention regarding the BISP design
is the unconditionality of the cash transfer under the Programme to the
recipient households. Conditionalities are basically behavioural
requirements expected from the recipients in order to remain eligible to
receive the cash transfer. These conditions are considered an effective
tool for poverty alleviation, helping to break the inter-generational
transmission of poverty by increasing the human capital of individuals.
Examples of such conditions can be found in a number of successful
programmes being carried out in different countries like the
Oportunidades/PROGRESSA in Mexico, Bolsa Escola and Bolsa Familia in
Brazil, Food for Education in Bangladesh and Programme of Advancement
through Health and Education in Jamaica [Son (2008)]. The conditions
laid down under these programmes are usually linked to education,
especially girls' education, and health, generally women and child
health. The idea behind these conditions is that handing over cash to
families is not enough to deal with poverty in the long run and such
conditions will obligate the recipient households to empower themselves
by investing in human capital and, hence, improve their chances of
decent employability and moving out of poverty on a permanent basis.
Along with achieving the socially optimal targets of human capital,
conditional cash transfers have some other advantages as well including
those mentioned by Adato and Hoddinott (2007):
(i) Lessening the possible stigma associated with cash transfer by
considering it a part of a social contract between the recipient
household and the state.
(ii) Preferred for political economy reasons, and making it
politically and economically more acceptable in the larger context.
Improvement in education and health indicators helps increase the
credibility of a programme which otherwise might be seen with suspicion,
especially by those not receiving it.
Contrary to this view there are those who believe that
conditionalities compromise the very objective of poverty reduction,
especially in the short run, by reducing the benefits of a cash transfer
to a poor household by constraining its welfare choices. These imposed
conditions can be, "expensive, inflexible, and inefficient- in the
worst cases screening out the poorest and the most vulnerable. Often the
burden of complying with conditionalities falls disproportionately on
women" [Samson (2006: 51)]. Some of the most common concerns raised
for conditionalities for a cash transfer include [as observed by Handa
and Davis (2006); Samson (2009); and Basett (2008); Son (2006); and
Regalia (2006)]:
(i) The high administrative cost of handling conditional cash
transfer might outweigh its positive impact.
(ii) Lack of access to educational and health facilities in the
poorer areas can make the condition redundant for the poor and hence
making them ineligible for the cash transfer.
(iii) The preferences of the poor people may differ from the
conditions imposed on them, thus, reducing the welfare gains.
(iv) Cultural and social exclusion and discrimination may leave the
neediest out of the welfare circle.
Those opposing conditionalities on cash transfers also consider it
demeaning to the poor as such conditions imply that the poor do not
themselves know what is good for them. As argued by Basett (2008),
following the traditional economic theory, cash transfers should ideally
be unconditional. Individuals, as rational beings, make decisions to
maximise their well-being, opting for choices where the perceived
benefits outweigh the perceived costs. Going by this logic a cash
transfer would be most effective with no conditions attached to it as
the poor, being rational economic beings, will maximise the benefits to
them. If a cash transfer reduces the opportunity cost of sending a poor
household's child to school instead of work, making the perceived
benefits of educating outweigh its cost, decision would be taken by the
household to send the child to school even without any compulsory
conditions. In a scenario where beneficiaries are informed and rational
economic beings, the state is caring and markets are efficient, IBRD
(2008: 48-49) believes that, "The 'theoretical
default'...... should be to favour unconditional cash
transfer".
As Samson (2009) observes, in some countries poverty levels are
high due to structural factors and not just because of the behaviour and
preferences of the poor. This would be true for any society, that has
yet to overcome its structural inequalities, which may discriminate
against certain people, restricting them not to avail the opportunities
that might be available to them otherwise, keeping them stuck in the
poverty trap. The need for a BISP-like programme, thus, becomes
important in the presence of vulnerable population in the country, which
is becoming more susceptible to poverty due to inflationary trends and
the structural inequalities characterising the societal makeup.
DATA AND METHODOLOGY
To evaluate the B1SP, the present study uses the Pakistan Panel
Household Survey (PPHS) carried out by the Pakistan Institute of
Development Economics (PIDE) in the year 2010. To link the cash
assistance with poverty dynamics the panel information of the survey is
used. It is worth mentioning here that the PPHS is a panel dataset,
comprising three waves. The Round-1 of the PPHS, named Pakistan Rural
Household Survey (PRHS), was conducted in 2001 in all four provinces of
the country, covering 2721 rural households. The Round-11 of the PRHS
was carried out in the year 2004 covering 1907 households in rural Sindh
and Punjab. The survey was not carried out in two provinces, Balochistan
and Khyber Pakhtunkhwa (K.P), due to the security conditions prevailing
there at that time. The third round of the panel survey was conducted in
2010, again in all four provinces, adding an urban sample to the survey
as well. Inclusion of the urban sample led to the renaming of the survey
as the Pakistan Panel Household Survey (PPHS). The urban sample of the
PPHS 2010 was selected from the 16 districts that were included in the
PRHS-2001. The PPHS-2010, thus, covers 4142 households in all four
provinces of the country, in both rural and urban areas. These over four
thousand households comprise 2198 panel households in the rural areas
(coming from PRHS2001), along with 602 split households from original
households, making the total rural sample stand at 2800 households. The
remaining 1342 households were included from the urban areas of the
selected districts to make up the total sample (4). It may be mentioned
here that the three waves of the PPHS-PRHS panel data collection is a
joint effort of P1DE and the World Bank.
The PPHS-2010 covers wide ranging modules to meet the objectives of
this study. A detailed section of the survey questionnaire deals with
the targeting process of the various safety net programmes initiated by
the government and by individuals to protect the marginalised segments
of the society. A transfer/assistance module included in the PPHS-2010
provides information about the status of received transfer/assistance in
three categories, namely: receive assistance; attempt but not succeed;
and never attempt. The respondents are also asked about how they had
utilised the received cash. There is, however, one limitation about the
questions asked about the cash transfers. There is no question about the
duration for which a household has been receiving any cash
transfer/assistance. The survey asks a household if it has received any
cash assistance in the last 12 months, without specifying the exact
duration for which the transfers have been taking place. For a better
analysis of the impact of these transfers on household welfare the exact
duration of transfer would have been valuable.
To analyse the socio-demographic and economic characteristics of
the households along with the status of received assistance, the present
study classifies households into three categories, that is the:
receiving group; attempt group; and never attempt group. To analyse the
effect of the BISP on a household's welfare, independent of other
cash transfers, two categories of households are formed. One consists of
households that receive the BISP, and the other category comprises those
households that receive cash transfers from sources other than the BISP.
To estimate the impact of the BISP cash assistance on a
household's welfare, this study follows the Propensity Score
Matching (PSM) method. The aim of the safety net programmes is to
improve the welfare of the poor, especially the most vulnerable.
However, all those in need do not necessarily receive it. Some of these
households get assistance and some do not, referred to as
'receiver' and 'non-receiver' households,
respectively.
Though other methods like logistic regression analysis, paired
observations and double difference method can also be used to analyze
the welfare impact, the PSM method was preferred due to its various
strengths over the other methods (5). For instance, the logistic
regression analysis ignores the issue of 'selection bias' and
considers the socio-demographic and economic characteristics of the
'receiver' and 'non-receiver' households as widely
different. It is usually understood that the 'non-receiver'
group is comparatively at a better welfare level and, therefore, is less
likely to receive assistance from the safety net programmes, that is, it
is less likely that an upper middle income or rich income household in
Pakistan will receive the assistance from Zakat or Bait-ul-Maal. Taking
the mean outcome of 'non-receiver' households as an
approximation is also not advisable as the 'receiver' and
'non-receiver' households usually differ in socioeconomic
characteristics even in the absence of these safety net programmes or
some time a programme purposely selects the 'receiver'
households [Kopeinig (2008)]. The paired observation and double
difference (DD) methods require the household information before and
after the intervention, in order to analyse the welfare impact of a
programme. Paired observation technique is usually applicable to one
variable only by assuming no impact of other variables, making it too
ideal to be applied here. The DD approach is a non-experimental approach
in which the welfare changes over time are estimated relative to the
outcome observed for a pre-intervention baseline. Though the baseline
information is available in the PPHS, because it is a panel household
survey and the 2001 and 2004 waves have the baseline information, but
this information does not necessarily precede the intervention. In the
present instance, the baseline information would not be homogenous as
the assistance-receiving households must have gone through numerous
socio-demographic and economic changes during 2004 to 2010 period,
making it impossible to capture the heterogeneity over the whole
duration. The Propensity-Score Matching (PSM) method developed by
Rosenbaum and Rubin (1983) is one of the possible solutions to deal with
the issue of 'selection bias'. The rationale behind this
technique is to find a comparison group that has similar characteristics
to those of the 'receiver' group in all aspects except one,
that is the comparison group does not get any cash assistance. This
method balances the observed covariates between the 'receiver'
group and the 'non-receiver' group based on the similarity of
their predicted probabilities of receiving the assistance, called their
'propensity scores'. The difference between PSM and a pure
experiment is that the latter also ensures that the treatment and
comparison groups are identical in terms of the distribution of
unobserved characteristics [Ravallion (2003)].
As noted earlier, two groups were identified in the PPHS on the
basis of status of cash assistance: the receivers and the non-receivers.
In the PSM analysis, the former are the 'treated units' while
the later are 'non-treated units'. Treated units are matched
to the non-treated units on the basis of the propensity score. See
Appendix A for a detailed explanation on the PSM methodology.
P([X.sub.i]) = Prob ([D.sub.i] = 1|[X.sub.i]) = E(D|[X.sub.i]) ...
(1)
Where [D.sub.i] = 1 if the household has received assistance and 0
otherwise and [X.sub.i] is a vector of pre-treatment characteristics.
Before estimating the PSM, two conditions should be met to estimate the
Average Treatment on the Treated (ATT) effect based on the propensity
score (Rosenbaum and Rubin, 1983). The first condition is the balancing
of pre-treatment variables given the propensity score. If the balancing
hypothesis is satisfied, the pre-treatment characteristics must be the
same for the target and the control groups. The second condition is that
of the unconfoundedness given the propensity score. If assignment to
treatment is unconfounded conditional on the variables pre-treatment,
then assignment to treatment is unconfounded given the propensity score.
Using equation 1, first the propensity scores are calculated through
logistic regression, and then the Average Treatment on the Treated (ATT)
effect is estimated by four different methods: Nearest Neighbour
Matching; Kernel Matching; Stratification Matching; and Radius Matching;
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)
Where
[Y.sub.li] is the potential outcome if household is treated, and
[Y.sub.0i] is the potential outcome if household is not treated
The above discussed methodology of Propensity Score Matching (PSM)
method has been applied to the PPHS-2010 dataset to analyse the impact
of the BISP on a receiving household welfare. Since household welfare is
a multi-dimensional phenomenon, therefore the impact has been estimated
on five indicators which are: poverty; food expenditure per capita;
health expenditure per capita; school enrolment of children of age 5-14;
and employment status of women of age 15-64.
Following the empirical exercise, firstly the propensity scores
have been estimated on the basis of Equation 1 where the dependent
variable is whether the household is a receiver or a non-receiver. On
the right side of the equation 1, the three sets of explanatory
variables have been used which can be the major reasons for getting
assistance. These three sets of variables are: the individual
characteristics, including the head of the household's sex,
education and employment status; the household characteristics,
including female to male ratio, household size, dependency ratio, number
of persons per room, land and livestock assets, shocks and presence of a
disabled person in the household; and the regional characteristics
including region and province. Since the dependent variable is
dichotomous in nature with two outcomes: received assistance or did not
receive assistance, therefore, the Binary Logistic Regression has been
applied to estimate the determinants of receiving assistance whereas the
'not-receiver' group serves as the reference category. Using
the 'psmatch2, pscrore, attnd, attk, altr and atts' commands
in STATA, comparison has been made between the treated and non-treated
units and the welfare impact has been calculated.
After calculating the propensity scores, the Average Treatment
Effect on the Treated (ATT) has been estimated. In order to make the
working sample even more comparable, the sample has been restricted to
only those units with probabilities that lie within the region known as
the common support, which is the area where there are enough of both,
control and treatment observations, to proceed with the comparisons
[Dehejia (2005)]. This also means that those units have been excluded
where the treated and non-treated units do not have comparable values.
EVALUATION OF THE BISP
Any effective social safety net programme needs to fulfil certain
criteria, including [Pasha, el at. (2005); World Bank (2007)]:
(i) Targeting: the extent to which a programme reaches its intended
target population rather than those who do not actually need it.
(ii) Coverage: the proportion of the target population that
benefits from a programme.
(iii) Administrative cost: the proportion of the administrative
cost against that used on the benefits.
(iv) Accessibility: the ease with which an eligible household could
access the programme socially, monetarily, logistically and
administratively.
(v) Adequacy: the sufficiency of the safety net, like a cash
transfer, to have any positive effect.
(vi) Positive incentive effect: safety nets that have a positive
incentive not only help to sustain the programme but also serve to
alleviate poverty in the larger context.
(vii) Sound financing source: safety nets with well-defined,
self-reliant sources are fiscally more sustainable than those relying on
ad hoc, external sources.
(viii) Independence from other transfers: a transfer taking place
under a programme should not exclude other transfers which may have net
negative effects on household's welfare.
Before we look into the performance of the BISP using some of the
above mentioned criteria (6) let us first see how many households are
receiving cash assistance, and their sources, in the study sample. As
reported by the respondents in the PPHS-2010, and shown in Table 2, 10.7
per cent of the households are receiving cash assistance from a variety
of programmes, with no major difference in the trends between the urban
and the rural areas. Among these programmes, the BISP is the largest
programme as it covers about two-thirds of the total households
receiving any form of cash transfer, in both the rural and urban areas.
As can be seen from Table 2, the received cash assistance has two
major categories, that is, cash assistance received from government
sources and cash assistance received from the individual sources, making
a total of 10.7 per cent of the total households receiving at least some
sort of cash assistance. Table 3 shows that out of these 10.7 per cent
cash receiving households, a significant proportion of the households
(8.8%) is getting assistance only from one source, and with the rural
areas showing a slightly higher proportion of households receiving cash
transfer as compared to urban areas. There are only a few households,
which are getting assistance from two or more than two programmes, i.e.,
a household may be getting assistance from the private Zakat and also
from Bait-ul-Maal.
As shown in Table 2 and 3, there are 435 households, which have
received assistance from various programmes and some of them have also
benefitted from more than one programme. Coming to the B1SP and the
number of its beneficiaries, we see from Figure 1 that the B1SP
receiving households outnumber all other public and private funded
safety net initiatives put together. At the national level, 264
households were receiving cash transfer under the BISP, 67 households
were getting assistance from other state-run safety net programmes, 29
households were assisted by private sources, and 27 households were
those which received assistance from multiple sources, making a total of
387 net households (see Figure 1).
As can be seen from Figure 1, the BISP is the largest safety net
programme covering more than two-thirds of the households receiving any
form of assistance in the study sample. As stated earlier, one of the
key objectives of the BISP was to help the poorest of the poor
households against rising inflation by providing for their basic needs
as these people have few physical and soft assets to cope with any
shock.
As the aforementioned discussion shows, cash transfers from the
various programmes have been split into three categories, namely
assistance from: the BISP; other government programmes; and private
sources. Likewise, the sample households have also been grouped into
three categories: recipient households; never-attempt households and the
attempt households. Table 4 summarises the patterns trends for cash
transfers across the four provinces and the two regions as reported by
the PPHS sampled households. The proportion of households receiving BISP
assistance is the highest in the province of Sindh (13.6 percent),
followed by Balochistan (8.5 percent), KP (4.9 percent) and Punjab (3.1
percent). Across the regions, not much difference is found between the
proportions of households receiving the BISP assistance in the rural
(7.1 percent) and the urban (6.9 percent) areas (Table 4). Although it
is difficult to explain some uneven distribution of the BISP cash across
the provinces found in this study and more in-depth data are required to
construe whether it is a political phenomenon or is due to any other
reason. This trend can be attributed to one probable reason that is over
representation of the poorer regions, particularly from Sindh and
Balochistan, in the PPHS sample. While the poorer districts of Badin,
Larkana and Loralai, in the province of Sindh and Balochistan, are
included in the sample, the more urbanised and well-off districts of
Karachi, Hyderabad and Quetta are not represented.
Another interesting factor to be noted in Table 4 is the proportion
of households falling in the 'attempt group' category. About
16 percent of the sampled households at the national level tried to get
assistance from the BISP but had not succeeded. Across the regions more
than one-fifth of the households attempt unsuccessfully to get some cash
assistance under the BISP in urban areas, while in the rural areas this
percentage is less with 14 percent. We may infer that the urban
inhabitants might have attempted more due to better information and
accessibility available to them as compared to the rural community
(Table 4). Contrary to the BISP distribution pattern, the percentage
distribution in other government programmes and in private programmes is
much lower and smoother, in both the 'attempt group' and the
'received group', showing little variation across the
provinces and regions. (Table 4).
BISP's Targeting
For any social safety net programme to be successful, the issue of
targeting is of utmost importance. Before the Proxy Means Test (PMT)
formula was adopted to identify the eligible households, the BISP had a
set of seven criteria that a household had to fulfil to be eligible to
receive cash assistance under the programme. Since the PPHS-2010 was
conducted before the introduction of the new PMT formula, we will
evaluate the efficiency of the BISP targeting on the basis of its
initial criteria. The initial criteria regarding the eligibility of a
household to receive BISP cash transfer included:
(i) A monthly income of less than Rs 6000.
(ii) No family member in government service.
(iii) Possession of no or less than 3 acres of agricultural land or
up to 3 marlas residential property.
(iv) Possession of Computerized National Identity Card.
(v) Should not be beneficiary of other support programmes.
(vi) Should not have an account with a foreign bank.
(vii) Should not possess a passport or an overseas Pakistani
identity card.
In this study a cross-check evaluation has been made on the basis
of available information in PPHS dataset of two indicators, which are
land holding and getting assistance from other government sources. The
above-mentioned B1SP criteria show that an eligible household should
possess less than three acres of land. However, Table 5 shows that about
10.5 percent of the BISP-receiving households have land ownership
ranging from 3 to 10 acres, and another 5.6 percent have landownership
of 10 acres and above, thus making a total of 16.1 percent of the
receiving households being ineligible in case of strict application of
the stated criteria. The criteria seem to be followed most strictly in
the province of Punjab, and to be most lax in KP (Table 5). The
cross-check analysis also shows that 12 BISP receiving households (that
is approximately 4 percent) are also receiving assistance from some
other government sources, which violates the conditions set forth by the
BISP design, as can be seen from Table 5.
Along with comparison of the households receiving cash assistance
against the BISP's prescribed criteria, another way of evaluating
the programme targeting is to look into the socio-demographic and
economic characteristics of the BISP receiving and non-receiving
households. Table 6 shows that the recipient households on average have
bigger household sizes, poor education of the heads of the households
and less working heads as compared to the other two categories, that is
the never attempt and attempt groups. Regarding assets, the households
receiving cash assistance are comparatively more deprived than the
never-attempt and attempt groups as the recipient households have fewer
assets, including house, land and livestock ownership. Two broad
conclusions can be drawn from Table 6. First, the recipient households
are at a disadvantageous position as compared to the never attempt and
attempt group. And second, the attempt group, though better than the
received group, is also under-privileged, and has much lower
socio-economic characteristics than the never attempt group. Similar
results have been found by other studies done on the topic in Pakistan,
including that done by Arif (2006).
A deeper insight into B1SP recipient, attempt and never attempt
households will help us evaluate the BISP vis-a-vis its targeting. Table
7 presents the status of the households with different socio-demographic
and economic characteristics by the status of received assistance. Based
on the PPHS 2010 dataset, these characteristics have been grouped at
individual and the household levels. The individual level
characteristics are related to the heads of households; while the
household level characteristics include family size, dependency ratio,
presence of permanent disabled person in home, room availability,
ownership of land and livestock, and experience of natural shocks.
Regarding individual characteristics, sex of the head of the
household is related to the status of received assistance from the BISP,
as can be seen from Table 7. The female-headed households have a higher
rate of receiving BISP assistance as compared to the male-headed
households. Also it is worth noting that there is a much higher
percentage of those households which are attempting to get BISP cash
transfers, reflecting an overall public interest in the programme (Table
7). The education
of the head of the household has a negative association with
receiving the BISP assistance as the households headed by more educated
persons are less likely to get any assistance from the BISP. A similar
trend prevails among the attempt group as well, with fewer educated
household heads attempting to get BISP cash assistance (Table 7).
A household's demographic, health and risk characteristics are
also closely related to the households' assistance receiving status
(see Table 7). With rising dependency ratio more households are found to
be receiving BISP assistance, with an even higher proportion attempting
to receive it. Households that have presence of a permanently disabled
person, or those who have experienced a shock during the five years
preceding the survey, do not show any definitive trend in receiving the
B1SP assistance. However, these results do show large number of those
households, which are attempting to receive the cash assistance (Table
7). It would not be wrong to infer that the high 'Attempt'
rates for the B1SP reflect the general accessibility of the programme
and the expectations people have from it.
Land and livestock ownership also shows an expected trend in
receiving and attempting to receive BISP (Table 7), with households
having fewer animals and smaller landholdings more likely to benefit
from the programme. Similarly, 'persons per room' also has a
positive association with both the received and the attempt groups. In
the rural areas, the sharecropping households have a much higher
proportion of receiving and attempting to receive rates for the BISP
cash assistance than those who own land, as can be seen from Table 7.
Summing up the patterns found in Table 7, we see a clear relationship
between the household's socio-economic characteristics and its
status of received BISP assistance. Also worth noting is the similarity
in the patterns between the received group and the attempt group. This
supports the finding presented in Table 3 which also showed that the
attempt group comprises vulnerable population as well, though in a
slightly better position than the 'received group'. These
findings hint towards a generally effective design formulation and
targeting by the BISP initiative, which probably needs an even bigger
coverage to include those eligible households that are in the
"attempt" group found in this study.
BISP's Role in Household Budget
The BISP, as mentioned earlier, is the largest social safety net
programme in Pakistan at present, covering more than two-thirds of the
households, initiated to protect the poorest of the poor households from
the rising inflation. The question of adequacy of the transferred amount
to the recipient household is an important factor in evaluating the
effectiveness of the BISP initiative. Needless to say, a cash assistance
of Rs 1000 per household per month is not such a big amount that can
change the life of the recipient but it is a reasonable enough amount to
help a poor household to cover some of its vital needs. It would,
therefore, be interesting to know where did the people spend the BISP
cash transfers. The PPHS-2010 asks the households to report the top
three priorities on which they spent the received cash transfers, the
results of which are shown in Figure 2.
The figure below (Figure 2) shows that as their first priority,
about 95 per cent of the households reported that they had spent the
BISP amount to meet daily household expenses, followed by 3.5 percent of
the households which spent this amount on education, 1.4 percent on
medical and 0.35 percent on dowry. As their second priority, more than
half of the households have spent the cash assistance on medical,
followed by education with 30 percent and daily household expenditures
with 17 percent. 33 percent of the households reported that their third
priority was to spend the BISP money to meet the miscellaneous needs,
followed by 30 percent on medical and 27 percent on daily household
expenses. The first two priorities, as reported by the households,
suggest that daily household expenditures and medical expenses are the
main concerns of the poor households on which they have spent the
BISP's assistance money. With the exception of some cash
utilisation on education, it would not be wrong to infer that the BISP
cash transfer is not primarily used to build assets for the households,
be they soft assets like education and skill development, or the
physical assets like purchase of livestock or agriculture inputs.
[FIGURE 2 OMITTED]
BISP and Dynamics of Poverty
The PPHS-2010 dataset has detailed consumption modules covering all
aspects of consumption including food and non-food items and also
sufficient information to calculate the head count poverty. It is,
therefore, possible to evaluate the relation between the BISP and other
forms of assistances with households' consumption expenditures and
poverty. For a detailed analysis, the per capita total expenditure is
split into food and nonfood expenditures. As can be seen in Table 8, the
results are quite interesting. Both average per capita food and non-food
expenditures are higher among the 'never attempt' group as
compared to the 'received' and 'attempt' group. The
'never attempt' group is, thus, comparatively better off and
in no need to get assistance. However, the 'received' group
has on average more per capita food and non-food expenditures as
compared to the attempt group, which is trying to get the B1SP
assistance (Table 8). It may be inferred from these findings that the
higher expenditures in the 'received' group as compared to the
'attempt' group is the result of the safety net intervention
made to enhance the welfare level of the vulnerable population. Since
the poor households spend a major proportion of their expenditures on
essential items like food, as can be seen in Figure 2, the expenditure
of the 'received' group on these commodities is higher than
the 'attempt' group.
The quintiles' analysis in Table 8 suggests that as we move up
the quintile ladder, fewer households are found receiving, or attempting
to receive, any form of cash assistance or to have attempted to get one.
It is, however, worth noting here a substantial proportion of the richer
households receiving the B1SP cash assistance is raising doubts about
the efficiency in its targeting. Some of these initial issues in
targeting are said to be dealt within the new criteria for selection of
beneficiaries by the BISP (as given in the discussion above) and it
would be interesting to see the effect it had on ground from a dataset
post these amendments.
A somewhat similar picture emerges when we look at the figures for
absolute poverty and receiving of the BISP cash in Table 8. As expected,
poverty is at a lower level among the households, which have never
attempted to receive the BISP cash assistance. However, if we look at
the poverty levels of those who receive and those who attempt to get the
BISP cash, we see a trend that begs explanation. Poverty levels among
the BISP recipients are slightly lower than those non-BISP recipients
who attempt to obtain it (see Table 8). Is the BISP cash assistance
helping its recipients to move out of poverty in some cases? The answer
can arguably be yes as if for nothing else it has helped improve the
recipient households' food expenditure (see Figure 2), which
eventually matters for the headcount measure of poverty.
As noted earlier, three waves of the PPHS dataset (2001, 2004 and
2010) are available, however, only for rural Punjab and Sindh. On the
basis of these panel households, five categories of poverty dynamics are
made to observe the association between the households' poverty
movements and its status of received BISP cash transfers. The five
categories are: poor in all three periods (chronic poor); moving out;
falling in; and moving in and out of poverty. Table 9 presents the
association between poverty dynamics and the status of received BISP
assistance, as found in the PRHS/PPHS. As can be seen from Table 9, the
never attempt group shows two features. First, two-thirds of the chronic
poor and moving out households have never attempted to get BISP
assistance and second, a substantial proportion of never poor are also
receiving BISP cash transfers or attempting to receive it (Table 9). It
is, however, significant to note that generally a bigger proportion of
households is either receiving the BISP cash assistance or attempting to
do so who have faced poverty at least once. Looking at the trends for
poverty dynamics and the BISP in Table 9, the lower percentage of
chronic poor households receiving BISP might be due to poor targeting or
the structural exclusion of chronic poor households due to their
socio-economic status. The behaviour of the 'attempt' group
experiencing poverty especially chronic poverty, which tries hard to get
assistance, hints towards both a need to expand the programme and an
improved targeting strategy.
Impact of the BISP: The Propensity Score Matching Analysis
As noted earlier in the methodology section, the PSM method is
applied on the PPHS-2010 dataset to analyse the impact of the BISP on
household welfare. The welfare impact of the BISP is estimated on five
household indicators which are: household poverty level; per capita food
expenditure; per capita health expenditure; school enrolment of children
of age 5-14; and employment status of women of age 15-64 (7). As briefed
in the methodology section, one has to estimate the propensity scores
through logistic regression to calculate the Average Treatment on the
Treated (ATT). There are two conditions that need to be met to estimate
the ATT, which are of balancing property and of unconfoundedness
property.
Table 10 presents the results for the determinants of the BISP
programme by incorporating the correlates, which satisfy both of the
above-mentioned conditions. The dependent variable is binary in nature,
that is whether the household has received assistance or not. The small
p-value from the LR test shows that at least one of the regression
coefficients is not equal to zero. Although the Pseudo [R.sup.2] in
logistic regression does not equate to [R.sub.2] of the OLS, the model
shows a significant Pseudo [R.sub.2]. As can be seen from Table 10,
three sets of independent variables have been added to the model,
related to household head; household; and the region. The results of the
logistic regression show that the education of the head of the household
has a significant negative association with receiving BISP cash
transfer. Among the second set of characteristics, we see that higher
the female-to-male ratio, and household size, the higher are the chances
to get assistance from the BISP (Table 10).
As can be seen from Table 10, the households that faced an
unexpected shock over the five years preceding the survey are more
likely to get B1SP assistance as compared to those who did not face any
such shock. Presence of a permanently disabled person in home and the
characteristics related to loan obtained, rooms per person and assets
ownership, including that of livestock, however, show no impact on
getting cash assistance from the BISP while the land ownership has a
significant negative impact on getting cash assistance (Table 10).
Regarding the third set of the independent variables, the coefficient of
region is not significant. On the contrary, however, a significant
variation in the BISP cash transfer prevails across the provinces, with
households in Sindh, KP and Balochistan more likely to receive BISP
assistance as compared to the province of Punjab.
This brings us to the final stage of the PSM analysis, results for
which are presented in Table 11. The Table shows the estimated welfare
impact of the BISP by displaying the Average Treatment Effect on the
Treated (ATT) against the five key indicators related to the household
welfare. The bootstrapped standard error, as well' as the number of
matching cases treated and the size of the control group, are also given
in Table 11. The results show that the impact of the BISP on headcount
poverty, though statistically not significant, is negative for all the
three measures of PSM. Despite having a reasonable targeting efficiency
(as seen in the above discussion as well), the lack of statistically
significant impact on poverty is not surprising as the rationale of the
BISP initiative suggests that it has not been designed to reduce poverty
per se, and has its main objective to protect the poorest of the poor
against the inflationary shocks. Second, the criterion of the BISP
suggests that the recipient households should be among the marginalised
segments of the society and far below the poverty line. Although these
households are getting a monthly stipend of Rs. 1000, the amount is,
however, too low to pull the households out of poverty. The fact that
these poor households on average have: bigger household sizes; higher
dependency ratios; tilted female-to-male ratios; and poor possession of
liquid, soft and physical assets which make it difficult for these
households to move out of poverty through a small cash transfer, as
provided by BISP.
The impact of the BISP cash transfer on per capita food and health
expenditure is statistically significant, as can be seen from Table 11.
Under the various measures of PSM, the BISP-covered households are
likely to spend more on food and health as compared to those households
which have not received the assistance but have similar socio-economic
and demographic characteristics. The calculated welfare impact of the
BISP transfer on food is Rs 20.6 by the Kernel method, Rs 22.9 by the
Stratification method, 29.1 by Radius method and Rs 48.4 by the Nearest
Neighbour method (Table 11). The welfare impact on health expenditure
shows that the households, which have received assistance from the BISP
are likely to spend Rs 62.4 more on health under the Stratification
measure; Rs 88.2 under the Nearest Neighbour method and Rs 55.7 under
the Kernal method as compared to those households which have not
received assistance from the BISP (Table 11). These results support the
finding presented in Figure 2, which shows that majority of the
BISP-receiving households spend the cash transfer to meet daily
household and medical expenses. These findings conform to the studies
done in other parts of the world where such cash transfers have been
found to improve the nutritional and health status of the recipients
[Duflo (2003); Aguero, et al. (2007); Paxson and Shady (2007); Cunha
(2010)].
The welfare impact of the BISP cash transfer on school enrolment of
children and women's participation in the labour market is
positive, though not statistically significant (Table 11). The
households receiving BISP cash assistance are at the threshold level of
their survival and are, thus, spending the received amount to fulfil
their basic necessities, mainly food, and not investing it to better
their physical or human capital. Other supplementary programmes of the
BISP related to skill development, employment and education may have a
positive impact on indicators other than food, whose analysis as
mentioned earlier, is beyond the scope of this study.
CONCLUSIONS AND POLICY RECOMMENDATIONS
The BISP might not be the 'magic bullet' to alleviate
poverty but findings of this study show that it has been able to provide
some relief to the recipient households as far as food and health
expenditures are concerned. In the Programme's defence it could,
however, be said that the rationale behind the initiative was to provide
assistance to the poorest of the poor households in the face of rising
food and fuel prices and not alleviating poverty per se. In the four
years since its inception, the Programme has shown the ability to evolve
with time, adjusting to the changing needs and criticism. Changes in the
recipient households' selection procedure and criteria by shifting
from the parliamentarians' recommendation to PMT scores, adoption
of technology in the delivery of cash through Smart Cards and phone to
phone banking instead of manual transfer through post offices are two
examples in this regard.
For any social protection programme to be effective it should have
the ability to reach the poor and promote a permanent exit from poverty.
The present study shows that although not all poor households were being
covered by the Programme, like those which unsuccessfully attempted to
get the BISP assistance, but the ones getting it were mostly poor (with
a few exceptions where adherence to the set criteria was found wanting
and consequently leakages to richer households were indicated). The
ability of the programme to reach the poor, however, is not matched by
its capacity to encourage a household's exit from poverty. The
original BISP design, with its unconditional cash transfer, does not
demand from the household to make an effort to invest in human or
physical capital, which may help in its transition out of poverty. With
the incorporation of other schemes under the BISP banner later,
including the Wciseela-e-Haq, Waseela-e-Taleem, Waseela-e-Sehat and
Waseela-e-Rozgar, this shortcoming in the Programme design may well have
been addressed, analysis of these schemes is beyond the scope of this
paper.
Political support at high levels is a prerequisite for the success
of any such programme. As discussed earlier, reasons linked to the
political economy may or may not encourage a government to invest in
such social protection schemes. Allocation of Rs 122 billion for the
BISP cash transfer is a huge promise which the future governments from
the other side of the political divide may not be willing to make. The
political nature of the name of the Programme, linking it to a
particular political party, (8) might not be considered desirable to
those belonging to other political parties. The slightly lower rates for
the BISP beneficiaries in the opposition-ruled province of Punjab hint
towards such issues that the Programme may face in case of a political
change at the Federal level.
Despite getting a nod from the World Bank on its performance and
being even labelled as, "An island of transparency" [Tahir
(2012)], the BISP needs to take certain factors into account for the
future. Foremost among these is the one related to fostering
inter-agency/programme coordination. As we saw in Table 1, a number of
safety net programmes exist in the country catering to different
segments of the population. As noted by Heltberg and del Ninno (2006:
8), these programmes are, however, 'fragmented, duplicative and
sometimes ceremonial' and are not able to fulfil the needs of the
recipients. There is thus a need to streamline all the existing
programmes and develop synergies between them for a more effective
impact. The BISP with its extensive data gathered for the PMT scores can
share the information with other programmes for a more efficient
delivery. This would also help counter multiple payments to the same
beneficiary under different programmes. A centralised system can also be
considered to avoid duplication and ensure more stringent application of
the eligibility criteria.
Proper monitoring and supervision need to be guaranteed to maintain
credibility of the Programme. A well-defined assessment procedure should
also be in place to judge the adequacy of the BISP cash transfer. Is the
assistance amount sufficient enough to make a reasonable impact on the
recipient household's budget? A cash transfer of Rs. 1000 per month
per household may be enough in the year 2008 but would it suffice in
years to come needs to be assessed periodically. Another factor ignored
by the BISP design at present is the transitory nature of poverty. A
household above the poverty line may move below it and vice versa in the
face of changing circumstances. The BISP cash transfer should,
therefore, take into account not just the poverty status of a household
but its dynamics vis-a-vis poverty as well. A recipient household might
become ineligible due to poverty dynamics while an ineligible household
may become eligible. Such changes need to be taken into account by the
BISP design for a more rational and equitable distribution of cash
assistance. Last but not the least, the BISP needs to formally
incorporate a mechanism for graduation out of poverty. Making a
household exit from the poverty trap should be the aim of the Programme
instead of continuously handing over cash assistance. Making households
economically stable and sustainable should be any social protection
programme's aim and the BISP should be no exception.
ANNEXURE
Table A-1 shows sample size of all the three rounds of panel survey
and it also includes the split households covered in both 2004 and 2010
rounds, building on the basic sample selected in the 2001 round. The
PPHS 2010 covered 2198 panel households from all the four provinces.
With an addition of 602 split households, the rural sample comprises
2800 households and the urban sample comprises 1342 households, making a
total sample size of 4142 households.
APPENDIX A
METHODOLOGY OF PROPENSITY SCORE MATCHING (PSM)
As noted earlier, two groups were identified in the PPHS on the
basis of status of cash assistance: the receivers and the non-receivers.
In the PSM analysis, the former are the 'treated units' while
the later are 'non-treated units'. Treated units are matched
to the non-treated units on the basis of the propensity score:
P([X.sub.i]) = Prob ([D.sub.i] = 1|[X.sub.i]) = E(D| [X.sub.i]) ...
(1)
Where
P ([X.sub.i]) = F(h ([X.sub.i]))
F(h ([X.sub.i])) can be the normal or the logistic cumulative
distribution
[D.sub.i] - 1 if the household has received assistance and 0
otherwise
[X.sub.i] is a vector of pre-treatment characteristics
Before estimating the PSM, two conditions should be met to estimate
the Average Treatment on the Treated (ATT) effect based on the
propensity score [Rosenbaum and Rubin (1983)]. The first condition is
the balancing of pre-treatment variables given the propensity score. If
p(X) is the propensity score, then:
[D.sub.i] = [X.sub.i]|([X.sub.i]) ... (2)
If the balancing hypothesis is satisfied, the pre-treatment
characteristics must be the same for the target and the control groups.
In other words, for a given propensity score, exposure to treatment is a
randomised experiment and, therefore, treated and non-treated units
should be on average observationally identical. The second condition is
that
of the unconfoundedness given the propensity score. Suppose that
assignment to treatment is unconfounded, i.e.:
[Y.sub.1] [Y.sub.0] = [D.sub.i] | [X.sub.i]
= [D.sub.i] | p([X.sub.i]) ... (3)
If assignment to treatment is unconfounded conditional on the
variables pretreatment, then assignment to treatment is unconfounded
given the propensity score. Using Equation 1, first the propensity
scores are calculated through logistic regression, and then the Average
Treatment on the Treated (ATT) effect is estimated as:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)
Where
[Y.sub.|i] is the potential outcome if household is treated and
[T.sub.0i] is the potential outcome if household is not treated
In the sense that ATT parameters focus directly on actual treatment
participants, they determine the realised gross gain from the welfare
programme and can be compared with its costs, helping to decide whether
the programme is successful or not [Heckman, et al. (1999)]. However,
calculating the effect through ATT is not immediately obvious since the
propensity score is a continuous variable. To overcome this problem,
four different methods have been proposed in the literature: Nearest
Neighbour Matching; Kernel Matching; Stratification Matching; and Radius
Matching, [Becker and Ichino (2002)]. This study uses the first three
methods.
Following Becker and Ichino (2002), the most straightforward
matching method is the nearest neighbour (NN) method where initially
each treated unit is matched with the controlled unit that has the
closest propensity score. The method is usually applied with
replacements in the control units. In the second step, the difference in
each pair of the matched unit is computed, and finally the ATT is
obtained as the average of all these differences. Let T be the set of
treated units and C the set of control units, and [Y.sup.T.sub.i], and
[Y.sub.C.sub.j] the observed outcome of the treated and control units,
respectively. If C(i) is a set of treated units matched to the control
treated unit i with an estimated PSM value [p.sub.i] then:
C(i) = [min.sub.j] [parallel] [p.sub.i] - [p.sub.j] [parallel] ...
(5)
The NN method may face the risk of bad matches if the closest
neighbour is far away. Such risk can be avoided by imposing a tolerance
level on the maximum propensity score distance (radius). Hence, radius
matching (RM) method is one form of imposing a common support condition
where bad matches can be avoided and the matching quality rises.
However, if fewer matches can be performed, the variance of the
estimates increases [Caliendo and Kopeining (2008); Smith and Todd
(2005)]. Radius matching can be shown as:
C(i) = {[p.sub.j] [absolute value of [parllel] pi ~ pj] < r }
... (6)
where the entire control units with estimated scores fall within a
radius r from treated matched [p.sub.i]. In both AW and RM measure, the
weights w/; are defined as:
[w.sub.ij] = 1/[N.sup.C.sub.i] if ... j [member of] C
and..[W.sub.ij] = 0. otherwise
The ATT for both AW and RM methods is, thus, as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (7)
The weights [w.sub.j] here are defined by [w.sub.j] = [[summation
over].sub.i] [w.sub.ij]. Similarly, variances can be estimated by
assuming that weights are fixed and the outcome is assumed to be
independent across units.
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (8)
In the third method, that is the Kernel method, all the treated
units are matched with a weighted average of all non-treated units using
the weights which are inversely proportional to the distance between the
propensity scores of treated and non-treated units. The A TT here can be
calculated as:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (9)
Where G (x) is a kernel function and [h.sub.n] is a bandwidth
parameter. The fourth method, the Stratification Matching method,
consists of dividing the range of variation of the propensity score in a
set of intervals (strata) such that, within each interval, the treated
and non-treated units have the same propensity score on average. The
method is also known as interval matching, blocking and
sub-classification method [Rosenbaum and Rubin (1983)]. Hence, the q
index defines the blocks over intervals of the propensity score, within
each block the programme computed as:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (10)
Where I(q) is the set of units in block q while N1 q and if q are
the numbers of treated and control units in block q. The ATT in the
Stratification Matching method is, thus, as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Where the weight for each block is given by the corresponding
fraction of treated units and Q is the number of blocks.
Table-A1
Households Covered during the Three Waves
of the Panel Survey
PRHS 2004
Panel Split Total
House- House-
PRHS holds holds
2001
Pakistan 2721 1614 293 1907
Punjab 1071 933 146 1079
Sindh 808 681 147 828
KP 447 -- -- --
Balochistan 395 -- -- --
PPHS 2010
Panel Split Total Urban Total
House- House- Rural House- Sample
holds holds house- holds
holds
Pakistan 2198 602 2800 1342 4142
Punjab 893 328 1221 657 1878
Sindh 663 189 852 359 1211
KP 377 58 435 166 601
Balochistan 265 27 292 160 452
REFERENCES
Adato, M. and J. Hoddinott (2007) Conditional Cash Transfer
Programs: A "Magic Bullet" for Reducing Poverty? 2020 Vision
Briefs. International Food Policy Research Institute (IFPRI).
Aguero, J. M., M. R. Carter, and I. Woolard (2007) The Impact of
Unconditional Cash Transfers on Nutrition: The South African Child
Support Grant. International Policy Centre for Inclusive Growth.
(Working Papers).
Ali, I. (2007) Inequality and the Imperative for Inclusive Growth
in Asia. Bangkok: Asian Development Bank.
Arif, G. M. (2006) Targeting Efficiency of Poverty Reduction
Programmes in Pakistan. Islamabad: Asian Development Bank. (Pakistan
Resident Mission Working Paper Series 4.)
Bari, F., E. Hooper, S. Kardar, S. Khan, 1. Mohammed, and A. Sayeed
(2005) Conceptualizing a Social Protection Framework for Pakistan.
Islamabad: Asian Development Bank. (Pakistan Poverty Assessment Update,
Background Paper Series, Background Paper 4).
Barrientos, A. (2006) Development of a Social Protection Strategy
for Pakistan. Social Protection Projects at IDS. UK: Institute of
Development Studies.
Bassett, L. (2008) Can Conditional Cash Transfer Programmes Play a
Greater Role in Reducing Child Undemutrition? The World Bank: Social
Protection and Labour. (SP Discussion Paper No. 0835).
Becker, S. and A. Ichino (2002) Estimation of Average Treatment
Effects Based on Propensity Scores. The Stata Journal 2, 358-377.
BISP (2012) Benazir Income Support Programme. Available at
http://www.bisp.gov.pk/. Caliendo, M. and S. Kopeinig (2008) Some
Practical Guidance for the Implementation of Propensity Score Matching.
Journal of Economic Surveys 22, 31-72.
Cunha, J. (2010) Testing Paternalism: Cash vs In-kind Transfers in
Rural Mexico. Stanford University. (Technical Report).
Dehejia, R. (2005) Practical Propensity Score Matching: A Reply to
Smith and Todd. Journal of Econometrics 125, 355-364.
De-Janvry, A. and E. Sadoulet (2004) Conditional Cash Transfer
Programs: Are They Really Magic Bullets? Department of Agricultural and
Resource Economics. University of California at Berkeley.
Duflo, E. (2003) Grandmothers and Granddaughters: Old-Age Pensions
and Intrahousehold Allocation in South Africa. World Bank Economic
Review 17, 1-25.
Handa, S. and B. Davis (2006) The Experience of Conditional Cash
Transfers in Latin America and the Caribbean. Development Policy Review
24, 513-536.
Heckman, J. and J. Smith (1999) The Pre-programme Earnings Dip and
the Determinants of Participation in a Social Programme: Implications
for Simple Programme Evaluation Strategies. Economic Journal 109,
313-348.
Irfan, M. (2005) Poverty and Social Safety Nets: A Case Study of
Pakistan. Islamabad: Pakistan Institute of Development Economics. (M1MAP
Technical Paper Series No. 15).
Jamal, H. (2010) A Profile of Social Protection in Pakistan: An
Appraisal of the Empirical Literature. Karachi: Social Policy and
Development Centre. (Research Report No. 81).
Khan, S. and S. Qutub (2010) The Benazir Income Support Programme
and the Zakat Programme: A Political Economy Analysis of Gender in
Pakistan. Oxford Policy Management Briefing Notes. London: Overseas
Development Institute.
Kidd, S. (2006) Social Protection and the White Paper. Social
Protection Team, Policy Division. London: DFID.
Kunnemann, R. and R. Leonhard (2008) A Human Rights View on the
Potential of Social Cash Transfers for Achieving the Millennium
Development Goals. Germany: Brot fur die Welt and Evangelischer
Entwicklungsdienst.
Pakistan, Government of (2007) National Social Protection Strategy.
A Social Protection Strategy to Reach the Poor and the Vulnerable.
Islamabad: Centre for Poverty Reduction and Social Policy Development,
The Planning Commission.
Pakistan, Government of (2012) Economic Survey 2011-2012.
Islamabad: Ministry of Finance, Government of Pakistan.
Pasha, H., S. Jafarey, and H. Lohano (2000) Evaluation of Social
Safety Nets in Pakistan. Karachi: Social Policy and Development Centre.
(SPDC Research Report Number 32).
Paxson, C. and N. Schady (2007) Does Money Matter? The Effects of
Cash Transfers on Child Health and Development in Rural Ecuador. The
World Bank. (Policy Research Working Paper Series 4226).
Posel, D., J. A. Fairburn, and F. Lund (2006) Labour Migration and
Households: A Reconsideration of the Effects of the Social Pension on
Labour Supply in South Africa. Economic Modelling 23. 836-853.
Ravallion, M. (2003) Assessing the Poverty Impact of an Assigned
Programme. In F. Bourguignon and L. P. da-Silva (Eds.) The Impact of
Economic Policies on Poverty and Income Distribution: Evaluation
Techniques and Tools. New York: Oxford University Press.
Regalia, F. (2006) Some thoughts about "Conditionalities"
in Cash Transfer Programs: Lessons from Latin America and the Caribbean.
Learning Workshop on Orphans and Vulnerable Children (OVC) and
Conditional Cash Transfers (CCTs). Nairobi, Kenya: Washington, DC:
Inter-American Development Bank.
Rosenbaum, P. and D. Rubin (1983) The Central Role of the
Propensity Score in Observational Studies for Causal Effects. Biometrika
70, 41-55.
Samson, M. (2009) Social Cash Transfers and Pro-Poor Growth.
Promoting Pro-poor Growth: Social Protection. OECD.
Samson, M. and M. Williams (2007) A Review of Employment, Growth
and Development Impacts of South Africa's Social Transfers. Cape
Town: EPRI. (Working Paper No. 41).
Samson, M., K. MacQuene, and 1. van-Niekerk (2006) Designing and
Implementing Social Transfer Programmes. Cape Town: EPRI.
Sayeed, A. (2004) Social Protection in Pakistan: Concept, Situation
Analysis and the Way Forward. Islamabad: Centre for Research on Poverty
Reduction and Income Distribution. (Working Paper 5).
Smith, J. and P. Todd (2005) Does Matching Overcome LaLonde's
Critique of Non-experimental Estimators? Journal of Econometrics 125,
305-353.
Son, H. H. (2008) Conditional Cash Transfer Programmes: An
Effective Tool for Poverty Alleviation? Manila: Asian Development Bank.
(ERD Policy Brief Series No. 51).
Tahir, P. (2012) An Island of Transparency. The Express Tribune.
Islamabad.
World Bank (2007) Social Protection in Pakistan: Managing Household
Risks and Vulnerability. (World Bank Report No. 35472-PK).
(1) For a detailed discussion on defining social protection in the
Pakistani context, and the difference between social protection and
social safety nets see Sayeed (2004) and Bari, Hooper, Kardar, Khan,
Mohammad and Sayeed (2005).
(2) For a detailed analysis on the current safety net initiatives
by the Government of Pakistan see, Jamal (2010), World Bank (2007), Arif
(2006), Irfan (2005), Bari, et al. (2005).
(3) The previous eligibility criteria used before the conduction of
the Poverty Scorecard Survey in 2010, will be used in this study, and be
discussed in the succeeding sections.
(4) See Annex 1 for the detailed household composition of the
PRHS/PPHS sample in the three rounds of survey in 2001, 2004 and 2010.
(5) Nssah (2006). Propensity Score Matching and Policy Impact
Analysis a Demonstration in Eviews. WPS 3877. The World Bank.
Washington, D.C.
(6) Some of the stated criteria to evaluate social safety net
programmes, such as the administrative cost to carry out the programmes,
and sound financing sources, are macro level issues and, thus, beyond
the scope of this study.
(7) These five indicators were formed using the PPHS-2010 dataset.
The headcount poverty was calculated by applying the official poverty
line at Rs 1,671.89 per adult per month. Monthly per capita food and
health expenditures were calculated from the consumption and health
modules of the Survey, respectively. The education module in the PPHS
has detailed information about the enrolment status of everyone in the
household, from it the enrolment status of children aged 5-14 has been
calculated. Regarding the last indicator of the socio-economic welfare,
the working status of the sampled women has been taken from the PPHS
question, "Did you work during the last week at least for one hour
for any wage or profitable home activities?"
(8) The Benazir Income Support Programme is named after Benazir
Bhutto, the twice prime minister of Pakistan and the chairperson of die
Pakistan People's Party up until the day she was assassinated in
December, 2007.
Durr-e-Nayab <dnayab@gmail.com> is Chief of Research at the
Pakistan Institute of Development Economics, Islamabad. Shujaat Farooq
<shjt_farooq@yahoo.com> is Assistant Professor at the Pakistan
Institute of Development Economics, Islamabad.
Table 1
Current Social Safety Net Initiatives with
National Coverage in Pakistan
Programme Financed by Benefit Target group
Benazir Income Public funds Cash as Married females
Support income belonging to very
Programme support poor households
Microfinance Donor funds Cash as loan To poor for self-
for setting up employment and move
business them out of poverty
Pakistan Bait- Public funds Cash support Disabled persons,
ul-Maal for daughters' widows, orphans and
wedding, food households living
and education below poverty line
People's Work Public funds Cash for work Provision of
Programme electricity, gas,
farm to market
roads, water supply
and such facilities
to rural poor
People's Commercial Financing for Unemployed educated
Rozgar Scheme banks selected people
businesses
Subsidy on Public funds In kind Poor segments
wheat, sugar
and fertilizer
Utility Stores Public funds Subsidy in Poor segments
prices
Zakat and Levy on bank Cash Deserving/needy
Uslir deposits and among Muslims
agri. yield
Child Labour Public funds Protection Working children
and children and facing abuse and
in bondage rehabilitation exploitation
services
Employees Employers' Cash Formal sector
Old-Age contribution employees
Benefit Scheme
Social Health Individuals' Cash General population
Insurance contribution
Workers Employers' Housing, Formal sector
Welfare Fund contribution schools and employees
health
facilities
Programme Coverage Managed by
Benazir Income National Fed. Govt
Support
Programme
Microfinance National RSPs and MFIs
Pakistan Bait- National Fed. Govt
ul-Maal
People's Work National Fed. Govt
Programme
People's National NBP
Rozgar Scheme
Subsidy on National Fed. Govt
wheat, sugar
and fertilizer
Utility Stores National Fed. Govt
Zakat and National Govt., zakat
Uslir and ushr
committees
Child Labour National Fed. and prov.
and children govts, FATA
in bondage and GB
Employees National Fed. Govt
Old-Age
Benefit Scheme
Social Health National Fed. Govt
Insurance
Workers National Fed. Govt
Welfare Fund
Source: Ministry of Finance 2012:226.
Note: Abbreviations used: Fed-Federal; Govt-Government;
Prov-Provincial; NBP: National Bank of Pakistan; Agri-
Agriculture; RSPs-Rural Support Programmes; MFIs-
Microfinance institutions.
Table 2
Number of Households Receiving Cash Transfer by Type/Source
of Assistance and Region
National Urban Rural
Total Number of Households 4142 1342 2800
Households Receiving Cash Transfers from Government Programmes
Benazir Income Support Programme 285 87 198
Food Support Programme 17 5 12
Zakat 19 2 17
Bait-ul-Maal 10 3 7
Food items on subsidized rates 5 3 2
People's Rozgar Programme 7 1 6
Others 29 8 21
Households Receiving Cash Transfers from Individuals
Private Zakat 21 8 13
Private Ushr 3 1 2
Fitrana/Sadqaat 16 7 9
Assistance/Gift in kind 23 8 15
Total Number of Households Receiving Cash Transfers from Any Source
435 133 302
Percentage of Households Receiving Cash Transfers from Any Source
10.7 10.5 10.8
Source: Authors' estimation from the micro-data of PPHS 2010.
Note: Total number of households in the study sample is 4142.
Table 3
Percentage of Households with the Number of Received
Assistances by Region
Number(s) of Cash Transfers National Urban Rural
0 90.47 90.94 90.26
1 8.79 8.04 9.13
2 0.52 0.87 0.36
3 and more 0.22 0.16 0.25
All 100.00 100.00 100.00
N (4,061) (1,269) (2,792)
Source: Authors' estimation from the micro-data of PPHS 2010.
Table 4
Distribution of Household's Assistance Receiving Status by Region (%)
Received Attempted Never Attempted Total
BISP
Overall 7.0 16.2 76.8 100.0
Rural 7.1 13.8 79.1 100.0
Urban 6.9 21.4 71.8 100.0
Other Government
Overall 2.1 3.1 94.9 100.0
Rural 2.7 2.8 94.9 100.0
Urban 1.7 3.6 94.7 100.0
Private
Overall 1.2 0.0 98.8 100.0
Rural 1.0 0.1 98.9 100.0
Urban 1.5 0.0 98.5 100.0
Source: Authors' estimation from the micro-data of PPHS-2010.
Note: Due to rounding off some of the figures appear as zeros.
Table 5
BISP Targeting: Compliance with the Landholding and
Multi-source Assistance Criteria
National Punjab Sindh KP Balochistan
Eligibility Criteria 1: Land Ownership
No land 73.0 79.2 72.2 61.3 73.7
Small landholding
(< 3 acre) 10.9 18.9 10.8 22.6 10.5
Medium landholding
(3 to < 10 acres) 10.5 1.9 11.3 12.9 2.6
Large landholding
(> 10 acres) 5.6 0.0 5.7 3.2 13.2
Total 100 100 100 100 100
N 285 55 161 31 38
Eligibility Criteria 2: Not Getting Cash from other Government Sources
Number of Households 12 4 5 2 1
Source: Authors' estimation from the micro-data of PPHS 2010.
Table 6
BISP Targeting: Socio-Economic Characteristics of
Households by Status of Assistance
Never
Characteristics Attempt Received Attempt
Household size (number) 7.5 8.0 7.8
Education of head (average years) 3.9 2.7 3.1
Heads employed (%) 79.0 76.5 82.1
HH facing shock in last 5 years (%) 86.4 80.9 86.6
Disabled person in home (%) 3.8 4.0 5.7
Under debt households (%) 23.5 34.0 38.1
Not owned house (%) 8.5 10.1 12.0
Katcha house (%) 61.3 75.5 70.9
Persons per room (number) 3.7 4.3 4.4
Large animal (number) 1.6 1.2 1.0
Small animal (number) 1.5 1.8 1.4
Land owned (acres) 3.5 2.1 2.0
Source: Authors' estimation from the micro-data of PPHS 2010.
Note: 1-Numbers represent average numbers, percentages and
the proportion of each characteristic in the three stated
categories, respectively.
Table 7
Rales of the Stains of Receiving DISP Assistance by Socio-economic
Characteristics of Households
Characteristics Never Attempt Received Attempt
Sex of the Head of the Household
Male 76.7 6.8 16.5
Female 71.7 13.8 14.5
Education of the Head of the Household
Illiterate 74.4 7.9 17.6
1-5 70.5 9.7 19.8
6-10 83.1 4.0 12.9
11+ 84.3 4.1 11.6
Dependency Ratio by Category
Low 78.7 6.9 14.4
Medium 76.7 7.0 16.3
High 72.6 7.4 20.0
Presence of Permanent Disabled Person in Home
No 76.9 7.2 15.9
Yes 70.6 6.9 22.5
Experienced Shock over Last 5 Years
No 74.6 9.8 15.6
Yes 76.9 6.7 16.4
Persons per Room
Up to 2 person in a room 84.7 4.6 10.7
>2 to 3 person in a room 79.9 5.7 14.4
>3 and above 70.7 8.8 20.5
Debt Status
No 80.4 6.2 13.4
Yes 68.4 8.8 22.7
Land Ownership by Category
No land 74.2 7.6 18.2
Up to 3 acres 78.9 6.8 14.3
3< to 10 acres 81.7 5.7 12.6
10< acres 84.8 4.9 10.4
Livestock (Large Animals Only)
No Animal 73.9 7.4 18.7
1/ 2 Animal 77.6 7.2 15.2
3/ 5 Animal 84.4 6.1 9.5
6 and above Animal 93.5 3.3 3.3
Farm Households (Rural Area Only)
Own land 81.2 6.0 12.8
Sharecropper 58.4 12.0 29.7
p-value
Characteristics Total (chi-square)
Sex of the Head of the Household
Male 100.0 0.005
Female 100.0
Education of the Head of the Household
Illiterate 100.0
1-5 100.0 0.000
6-10 100.0
11+ 100.0
Dependency Ratio by Category
Low 100.0
Medium 100.0 0.003
High 100.0
Presence of Permanent Disabled Person in Home
No 100.0 0.088
Yes 100.0
Experienced Shock over Last 5 Years
No 100.0 0.035
Yes 100.0
Persons per Room
Up to 2 person in a room 100.0 0.000
>2 to 3 person in a room 100.0
>3 and above 100.0
Debt Status
No 100.0 0.000
Yes 100.0
Land Ownership by Category
No land 100.0
Up to 3 acres 100.0 0.000
3< to 10 acres 100.0
10< acres 100.0
Livestock (Large Animals Only)
No Animal 100.0
1/ 2 Animal 100.0 0.000
3/ 5 Animal 100.0
6 and above Animal 100.0
Farm Households (Rural Area Only)
Own land 100.0 0.000
Sharecropper 100.0
Source: Authors' estimation from the micro-data of PPHS 2010.
Table 8
Average per Capita Monthly Expenditures and Expenditures
by Quintiles by Status of Received BISP Assistance
Never Attempt Received Attempt
Per Capita Monthly Expenditure on (in Rs)
Food 1752.3 1602.8 1534.1
Non-food 1312.2 991.8 931.4
Total 3105.2 2615.7 2478.5
Per Capita Monthly Expenditure by Quintiles (%)
First 69.3 7.9 22.8
Second 74.7 7.6 17.8
Third 76.3 6.3 17.5
Fourth 77.4 8.4 14.2
Fifth 84.1 5.2 10.7
p-value (chi-square)= 0.000
Poverty [Level.sup.1] (%)
18.2 25.2 27.2
Source: Authors' estimation from the micro-data of PPHS
2010.
Note: Measured through headcount method at Rs 1,671.89 per
adult per month.
Table 9
Status of Current Received Cash Assistance and Poverty Dynamics:
2001, 2004 and 2010 (Rural Punjab and Sindh only) (1)
Never
Attempt Received Attempt Total
Poor in Three Periods 66.7 6.7 26.7 100
Moving Out 67.4 11.5 21.2 100
Falling In 69.3 10.6 20.1 100
Moving Out and Falling In 72.3 8.7 19.0 100
Non Poor in Three Periods 83.0 7.6 9.5 100
Source: Authors' estimation from the micro-data of PRHS
2001, PRHS 2004 and PPHS 2010.
Note: 1- Only rural Punjab and Sindh are included in this
part of the analysis as they are the only regions where all
three rounds of the panel survey have been conducted.
Table 10
Determinants of the BISP Cash Transfer: Logistic Regression
Covariates Coefficients Standard Error
Education of head (years) -0.045 * 0.017
Female to male ratio 0.213 * 0.072
Household size (in numbers) 0.044 ** 0.020
Unexpected shock in last five years
(yes=1) 0.513 * 0.181
Presence of disabled person (yes=1) -0.038 0.331
Number of room per person -0.046 0.036
Land ownership (in acres) -0.036 ** 0.015
Total large animals -0.034 0.034
Total small animals 0.018 0.019
Region (urban=1) -0.004 0.179
Sindh/Punjab 1.421 * 0.182
KP/Punjab 0.535 ** 0.260
Balochistan/Punjab 0.940 * 0.257
Constant -3.160 * 0.286
LR chi2 120.75 (14)
Log likelihood -792.70636
Prob > chi2 0.0000
Pseudo [R.sup.2] 0.0708
N 3,379
Source: Authors' estimation from the micro-data of
PRHS 2001, PRHS 2004 and PPHS 2010.
Table 11
Average Treatment Effects of BISP Under Various Measures
of PSM and Socio-economic Indicators of Household
Food Expenditure Health Expenditure
Poverty per Capita per Capita
Method (Yes=l) (Monthly) (Monthly)
Nearest Neighbour Method
ATT -0.015 48.36 88.16
N. Treated 235 235 235
N. Control 236 236 236
St. Error 0.042 24.25 41.11
Bootstrap
t-stat -0.359 1.99 2.14
Kernel Method
ATT 0.014 20.57 55.70
N. Treated 235 235 235
N. Control 2992 2992 2992
St. Error 0.028 12.38 20.54
Bootstrap
t-stat 0.505 1.66 2.71
Radius Method
ATT 0.014 29.11 19.31
N. Treated 191 191 191
N. Control 730 730 730
St. Error 0.046 14.98 12.16
Bootstrap
t-stat 0.296 1.94 1.59
Stratification Method
ATT -0.012 22.92 62.36
N. Treated 235 235 235
N. Control 2992 2992 2992
St. Error 0.028 10.87 32.79
Bootstrap
t-stat -0.422 2.11 1.90
School
Enrolment
of Children Employment Status of
of Age 5-14 Women of Age 15-64
Method (Yes=1) (Yes=1)
Nearest Neighbour Method
ATT 0.03 0.013
N. Treated 517 568
N. Control 417 489
St. Error 0.05 0.038
Bootstrap
t-stat 0.52 0.34
Kernel Method
ATT 0.006 0.075
N. Treated 517 568
N. Control 6430 6339
St. Error 0.019 0.080
Bootstrap
t-stat 0.32 0.94
Radius Method
ATT 0.05 0.03
N. Treated 273 387
N. Control 684 753
St. Error 0.07 0.05
Bootstrap
t-stat 0.714 0.60
Stratification Method
ATT 0.048 0.075
N. Treated 517 568
N. Control 6324 6376
St. Error 0.033 0.19
Bootstrap
t-stat 1.45 0.39
Source: Authors' estimation from the micro-data of PPHS 2010.
Fig. 1. Number of Households by Source of
Assistance Received and Region
National Urban Rural
264 77 187
67 16 51
29 10 19
27 12 15
Source: Authors' estimation from the micro-data of PPHS 2010.
Note: Table made from bar graph.
