Impact evaluation of remittances for Pakistan: propensity score matching approach.
Siddiqui, Rizwana
This study attempts to uncover the biases in the impact evaluation
of remittances when the problems relating to selection bias and counter
factual are not taken into account. Taking migration as an intervention
and foreign remittances as an input, the study measures the
socioeconomic impact using an approach which yields more accurate
non-experimental estimates in self-select cases through multiple output
and outcome indicators such as income, expenditure, saving, and capital
accumulation which, directly and indirectly, affect households'
welfare, poverty incidence and growth prospects of a country. Using PIHS
data, the study first calculates the difference in socioeconomic
characteristics of treated or remittances beneficiary households (RBH)
and control or remittances non-beneficiary households (NRBH) ignoring
endogeneity and observable differences. Second, it calculates the
propensity score and evaluates the impact using data from common support
area for both RBH and NRBH households. Third, it evaluates the impact
using the propensity score matching approach which replicates the
experimental benchmark. The difference in the first and the third
estimates reveals the bias originating from the issues of selection and
difference in observable characteristics. The results show that after
controlling for observable characteristics of households, regional
difference, networking and applying the selection correction technique,
the average impact of remittances is significantly reduced. A
disaggregated analysis shows that the socioeconomic impact of
remittances differs by the level of skills. The impact is significant
for relatively low skilled poor households but for high skilled
households it remains significant only in case of bank deposits. The
paper concludes that estimates are biased upward if the selectivity
issue and endogeniety problems are ignored which may lead to wrong
policy implications.
JEL Classification: F24, 015, P36
Keywords: Propensity Score Matching, Remittances, Poverty, and
Capital Accumulation
1. INTRODUCTION
The extent to which foreign remittances affect welfare, poverty and
growth has been a matter of considerable debate. (1) Pakistan is among
the top five countries whose foreign exchange earnings comprise a
significant amount of foreign remittances. Growing by about 20 percent
annually the foreign remittances now form 5 percent of Pakistan's
GDP in 2010-11. (2) Their importance can be viewed from the fact that
remittances do not have to be paid back like other foreign exchange
receipts such as official development assistance. Therefore, its
integration into overall development planning is essential to maximise
its benefits. A comprehensive analysis using the most appropriate
technique is needed to form appropriate policies [White (2005)].
Foreign remittances play an important role at the macro as well as
micro levels. They are a major source of income of the recipient
households in Pakistan and help mitigate the financial hardships of the
households. The recipient households put them to various uses that have
welfare, poverty, and growth implications. The existing literature (3)
measures the impact of the remittances using methodologies that vary
from the most complicated ones such as the economy wide computable
general equilibrium (CGE) model [Siddiqui and Kemal (2006) and Siddiqui
(2009)] to the simplest as descriptive statistics. (4) The CGE method is
most demanding not only in overcoming the problem of data scarcity and
capturing multi-round effects (5) but also in finding appropriate
elasticities and skills for programming [Knerr (1992)]. Some studies
explore just one dimension or the other in the partial equilibrium
framework. (6) However, the majority of these studies do not account for
selection to migration (7) and ignore the counterfactual or differences
in the observable characteristics, hence they tend to overstate the
impact. Therefore, it is obligatory to take into account the selectivity
issue and the difference in observable characteristics that measure the
actual impact of treatment. Any ambiguity in the impact raises need for
empirical research. To correctly measure the socio economic impact of
remittances, one must compare the socio-economic indicators such as
income, expenditure, saving and capital accumulation (human, financial
and physical) of the migrant-households (8) to what they have if they
are not migrated. The latter has not been observed. Recognising this
difficulty, Rosenbaum and Rubin (1983) were the first to propose the
propensity score matching (PSM) approach for more accurate
nonexperimental estimates in self-select cases. In the following years,
the method was also recommended by Heckman, et al. (1997); Dehejia and
Wehba (2002); White (2006); and McKenzie and Gibson (2006); Deininger
and Liu (2008) for this type of analysis.
Considering migration as an intervention--a case of non-random
selection of remittance beneficiary households [individual self-select
to migrate]--this author adopted the PSM approach to evaluate the impact
of remittances on the socio economic condition of households which,
directly and indirectly, affect welfare, poverty, and growth prospects
of the country. For this purpose data from the Pakistan Integrated
Households Survey (PIHS) [Pakistan (2002)] on income, consumption,
saving, asset holdings, indebtedness, capital accumulation--human,
physical and financial, and domestic economic activity for both groups
i.e. [RBH and NRBH] was used employing the same methodology. This study
assumes that households which receive foreign remittance are treated or
remittance beneficiary households (RBH) and the control group which does
not receive remittance income are called non-treated or remittance
non-beneficiary households (NRBH).
Here three approaches are used to calculate bias in attribution of
remittances. First, the naive approach to calculate the mean difference
in socio-economic indicators using full sample of all RBH and NRBH
ignoring selection bias and counter factual. Second, the difference in
the indicators is calculated using data from common support area after
allowing for the propensity score. Third, after pairing observation from
RBH and NRBH groups based on PSM to balance treatment and control group
on observable characteristics, the difference in the mean value of
socio-economic indicators is calculated. The difference in the three
estimates reveals the bias that originates due to selection bias and the
difference in observable characteristics.
The rest of the paper has been organised as follows. The next
section presents impact evaluation methodology, selection variables and
multiple socio-economic indicators. Data used for the analysis are
discussed in Section 3. Section 4 discusses distribution of beneficiary
and control group. The results are discussed in Section 5. Sections 6
and 7, respectively, discuss heterogeneity in the impact by skill level
and compare the results of this study with earlier ones. Section 8
concludes the paper.
2. METHODOLOGY
In impact evaluation studies, bias originates from three sources;
(i) selection bias, (ii) self-selection, and (iii) difference in
observable characteristics.
First, the naive approach is used to measure the difference in
socio-economic impact of remittances. In this approach, the impact is
measured using all households-- RBH and NRBH ignoring selection bias and
counter factual. (9)
Second, the conceptual framework from Rosenbaum and Rubin (1983)
and Heckman, et al. (1997, 1998); which has been widely used in this
type of analysis [Dehejia and Wehba (2002); McKenzie and Gibson (2006);
Deininger and Liu (2008) etc.] is used to reveal the bias (if any) in
the estimates. The framework consists of PSM and difference methods. The
PSM approach has many advantages over the other methods:
(1) It overcomes the problem of multi dimensionalities and develops
an index of propensity score P(X) for the treated (RBH) and control
(NRBH) groups to match. In the presence of a large number of explanatory
variables, matching all variables becomes difficult. The PSM method
renders the multidimensional matching problem to one-dimensional i.e.
instead of matching on a vector X of variables.
(2) It gives more accurate non-experimental estimates, where
households self-select into the programme [Dehejia and Wehba (2002);
McKenzie and Gibson (2006); Deininger and Yanyan (2008); White (2006)].
(3) It replicates the experimental benchmark if the outcome from
the treatment and control groups is (i) compared over a common support
area (the distribution of households likely to receive the treatment is
similar in both groups), (ii) Data is collected from both groups in a
similar fashion [Dehejia and Wehba (2002)].
(4) The method does not require a parametric model and allows the
estimation of mean impacts without arbitrary assumptions about
functional forms and error distribution [Jalan and Ravallion (2001)].
In this study the remittance-response function or selection
equation is estimated first. The major concern in the PSM approach
concerns which explanatory variables should be included in remittance
response function to estimate the probability of a household receiving
remittances or not. The probability depends on households and community
based characteristics of RBH and NRBH. The dependent variable represents
the status of households receiving remittance income (decision to
migrate) or not i.e., a dichotomous variable taking the value
'1' when household receive remittances and '0' when
it does not.
[D.sub.REM] = [b.sub.i] [x.sub.i] + [g.sub.j] [z.sub.j] ... ... ...
... ... ... (1)
[D.sub.REM] is a dichotomous variable where [D.sub.REM] = 1 if a
household receive remittances, otherwise 0.
[x.sub.i] is a vector of individual or household level
characteristics
[Z.sub.j] is a vector of community characteristics
In the absence of information about migrated labour, it is
worthwhile to examine family characteristics that motivate the migrated
worker's decision to remit income. These variables are chosen in
such a way that they affect remittance income (migration decision) but
not the outcome variables.
The most important variable that determines remittance from
migrated labour is their education [Nishat and Bilgrami (1993); Adams
(2008). (10) This information is not available from the existing data.
However, the correlation between the education of the head of the
households and average education of the earners is 0.75. Therefore, here
the education of the head of household has been used as a determinant of
the remittance income. Five categories of education [(1-5), (6-9),
(10-13), (14-15), and 16 and above including all professional
categories] are defined with base category of education of less than one
year.
The principal migration motivation comes from household size, which
determines the need for migration. If a household has a large family
size, labour is expected to migrate to earn more owing to the fact that
labour receives higher wages abroad. The age of the head of the
household is included in the equation as an explanatory variable.
People living in the same community are more likely to have many
characteristics ([Z.sub.j]) in common including community norms,
infrastructure, leadership, physical environment, social structure,
household strategies. Therefore, they behave in similar fashion. The
existence of migratory network affects migration from that community. In
this study community remittance income per household has been used to
indicate the existence of migratory network. (11) It indicates that the
larger the value of remittances per household, the stronger is the
migratory network and more people are expected to migrate abroad from
that community. (12)
Region also reflects a certain skill level. For instance, labour
from rural area belongs to lower education level and more likely to send
a higher proportion of low skill (low educated or unskilled) labour
compared to urban labour. Language is also an important factor in
determining the type of labour migrating to different parts of the
world. In this case, workers from more developed provinces with high
literacy rate are more likely to send skilled labour. (13) In this study
one dummy variable has been employed for region -[D.sub.Region] with
rural as base category and three dummy variables (D,) for three
provinces, Punjab, Sindh, Khyber Pakhtunkhwa (KP) with rest of Pakistan
(ROP) (14) as the base category to control for regional differences,
assuming that characteristics mentioned above are region specific and
vary across the regions [Nishat and Bilgrami (1993)].
Rizwana Siddiqui <rizwana@pide.org.pk> is Senior Research
Economist at the Pakistan Institute of Development Economics, Islamabad.
Author's Note: I am very thankful to Dr Howard White,
Executive Director of International Initiative for Impact Evaluation
(3ie), for his continuous help and invaluable comments. I am also
thankful to anonymous referees for their detailed comments.
The likelihood of being a recipient family is presented by reduced
form equation which includes above mentioned households level and
community level characteristics.
The model is defined as follows:
[D.sub.REM] = [alpha] + [[beta].sub.2] [Y.sup.com.sub.REMH] +
[[beta].sub.3]Hsize + [5.summation over EDU=1] [[beta].sub.EDU] +
[summation over i] [[gamma].sub.i] [D.sub.i] + [phi][D.sub.region] ...
(2)
[D.sub.i] 1 for ith province and 0 otherwise,
where
i = P(Punjab), Sindh (S), Khyber Pakhtunkhwa (KP)
= 0 otherwise
[D.sub.EDU] = 1 for kth education level of head of the household
and 0 otherwise,
where
EDU = primary (1-4), Middle (5-9), FA(10-13), BA (14-15), 16 and
above with base category of less than one year of education.
[D.sub.region] = 1 for urban and 0 otherwise, base category rural
Hsize = Household size--total members present in a household
[Y.sup.com.sub.REMH] = Community remittance income per household
[Age.sub.HH] = Age of the head of household
In this study the SPSS programme has been used to estimate the
logistic function defined in Equation 2.
The second concern in this approach is to choose treatment (RBH)
and comparison or control group (NRBH). The SPSS-PSM-macros developed by
Levesque to match PSM of the treated (RBH) with control group (NRBH) are
employed and the common support area (S) is defined selecting the
observation following Heckman, et al. (1998).
S = Supp(X|[D.sub.rem] = 1) [intersection] Supp(X|[D.sub.rem] = 0)
... (3)
It defines the area with the common range dropping all observations
from RBH and NRBH whose P values are beyond the range defined in
Equation 3.
Third, the exact matching approach in which each RBH is paired with
NRBH has been used which minimises the difference of their PSM within
the common support area and drops the rest of the households.
The next goal is to calculate the attribution of remittances to
socio-economic outcome. Classic evaluations focus on two parameters:
average impact on the units that are given the opportunity to take it up
(non-participant-NRBH) and the average impact on those who receive it
(participants-RBH) [Ravallion (2009)].
Let T be the vector of socio-economic variables that are defined as
output and outcome variables. The outcomes corresponding to [D.sub.REM]
= I and [D.sub.REM] 0 are denoted by (YI, Y0), respectively, and X is
the vector of variables that are time invariant characteristics of the
treated unit RBH. The assumption underlying the matching estimator is
that all relevant differences between the two groups are captured by
their observables X. The treatment assignment DRHM (household receiving
remittance income) is independent of Y(Y0 and Y1) given X (observable
characteristics). It can be written as
([Y.sub.0] [Y.sub.1]) [??][D.sub.REM] | X ... (4)
This implies that
([Y.sub.0]) [??][D.sub.REM] | P(X) ... (5)
Where P(X) is propensity score, and defined as P(X) = Pr
([D.sub.REM] = | X) which by definition lies between 0 and 1. Another
implicit assumption required by the matching estimator is the stable
unit treatment value assumption (SUTVA), which states that the outcome
of ith unit given treatment is independent of the outcome of unit jth
unit given treatment. To satisfy this assumption we have to ignore the
general equilibrium effects [Ham, et al. (2005)]. In the absence of
baseline data, the remittance impact (REMP) is measured as follows:
REMI = E(Y1|[D.sub.REM] = 1) - E(Yo|[D.sub.REM] = 0)
This expression measures mean difference in the impact of
remittance income on RBH over the control group NRBH.
The effects of remittances vary with the education of head of the
households. (15) This study tests the hypotheses: Does the effect of the
treatment vary by education level? Let Edu denote schooling and s denote
the different levels of schooling. The effect of remittances income on
different educational groups is estimated for each education level in
the following way:
[[DELTA].sub.s] = E(Y1 - Y0) [paragraph] = 1, Edu=s) = E(Y1
[paragraph] [D.sub.REAM] = 1, Edu=s)-E(Yo [paragraph] [D.sub.REAM]=0,
Edu = s) (6)
We define s = 0, 1,2, 3,4, 5
S=0 if education is less than one year
S=1 if education is below primary, (1-4) year
S=2 if education is between (5-9) year
S=3 if education is between (10-13) year
S=4 if education is between (14-15) year
S=5 if education is 16 years and above including professionals such
as doctors, engineers etc.
This study measures the attribution of remittances to socioeconomic
aspects of households such as income, expenditure, saving, investment,
welfare, and poverty. These indicators are discussed in detail in the
next section.
Third, the bias in the impact of remittances is calculated ignoring
the differences in observable characteristics. It is calculated as
difference in difference.
Let D1, D2, and D3 be the differences measured using full sample,
data from common support area, and using PS matching of RBH and NRBH,
respectively. The difference between Dl and D3 reveals the bias in the
estimates if one ignores the issues of endogeniety and differences in
the observable characteristics.
Bias = D1 - D3
2.1. Socio-economic Indicators
In this study multiple socio-economic indicators (including basic
need indicators (BNIs such as calorie intake, housing, safe drinking
water, sanitation facilities, education) have been used to measure
attribution of remittances. Satisfaction of basic needs determines a
country's capability development [Siddiqui (2006)] and poverty
reduction.
(a) Income Effects
Migrants are expected to receive higher income as workers leave
their home country to take the advantage of higher wages [Farchy (2009)]
and remit a significant amount of their earnings; about 78 percent of
their total earnings [Siddiqui and Retrial (2006)]. Remittances are not
exogenous transfers but a substitute for the domestic earnings that
migrants had earned if they had not migrated. Income per adult
equivalent has been used here to measure the income effect of migration.
(16)
The RBH group has three choices to use these receipts: consume,
save or invest, which directly and indirectly affect poverty and growth
prospects of a country.
(b) Consumption
Earlier literature on socio-economic impact of remittances
[Gillani, et al. (1981) and Amjad (1988)] show that remittances (57 to
62 percent) are generally, used for consumption purposes. (17) The
expenditure pattern of households is central to any meaningful
discussion on welfare and poverty. If households increase the demand for
food and non-food items, remittances are more likely to improve the
welfare of households and reduce poverty. Here food and non-food
expenditure in rupees per adult equivalent term and calorie intake (BN)
per adult equivalent have been used which directly determine the welfare
and poverty effects and indirectly determine the growth effects as
increase in expenditures boosts the economy through multiplier effects.
Similarly, higher expenditure on consumer durables [households'
equipment] such as washing machine, TV, oven, refrigerator, automobiles
also indicates higher standard of living. Ownership of households'
equipment is measured in rupee value at the household level.
(c) Investment
If remittances ease working capital constraint, it is expected to
improve capability and growth prospects of a country by increasing human
and physical capital.
Investment in Human Capital: Remittances are expected to improve
the capability of a household if migrant households spend more on
children's education to improve the quantity and quality of their
education. It compensates for loss in human capital due to migration of
labour in the long run and improves literacy rate (as indicator used to
measure capability of a country). In this study 'average class of
school-going children in a household' and expenditure on education
per class have been used to measure quantity and quality of human
capital formation, respectively i.e., the key outcome from the
perspective of economic growth in the long run.
Investment in Physical Capital: Empirical studies show that migrant
households largely invest in housing. Housing is one of the basic needs.
This effect has been captured in terms of adults/room. In addition,
existence of facilities like availability of clean water (BN),
sanitation, electricity, gas, and telephone indicates higher standard of
living. These facilities are partially dependent on infrastructure
development by the government. (18)
Investment in productive capital is captured through agriculture
farming/land holding, livestock holdings, and entrepreneurial activity
etc. If remittance income increases accumulation of productive capital,
it is expected to have a growth promoting impact.
(d) Saving
Households save by buying jewellery, keep cash at home or save in
bank schemes, (i) Jewellery is one form of investment in unproductive
capital though it indicates leakage from the economy but can be used in
growth enhancing activities. For instance, It may be used for investment
purposes on the return of migrant labour. However, for the year under
analysis, this is idle money and indicates households' financial
condition. This indicator is measured at the household level in terms of
rupees, (ii) Households cash holdings at home are measured in rupees,
(iii) Households' bank profit receipts measuring the size of the
bank deposits (19) are used as outcome indicators of financial saving
that determine financial development--financial resources available for
credit distribution. Remittances via financial development can also
positively affect poverty and growth [Aggarwal, et al. (2006)]. If the
deposit level is higher for RBH, it may also have growth-enhancing
effect through the banks' intermediation process--credit expansion.
It can be indirectly inferred that higher bank deposits have a growth
promoting impact.
(e) Poverty
Poverty is measured by head count ratio i.e. the percentage of
population below the poverty line which is officially prescribed poverty
line for rural and urban areas.
(f) Growth
The growth impact is deduced indirectly front growth oriented
activities such as increased demand for goods and services,
entrepreneurial activity, livestock activities, land farming etc.
Entrepreneurial activities alone are considered as a driver of growth.
If these activities increase, one may expect to have growth promoting
effects of remittances. These channels of remittances' impact on
outcome indicators are comprehensively presented in log Frame in
Appendix 1 Table 1.
3. DATA
This study relies on data from Pakistan Integrated Household Survey
(P1HS) for the year 2001-02 conducted by Federal Bureau of Statistics
(FBS) [Pakistan (2002)]. The data provides detailed information on
household size, income, consumption (food, non-food, and durable
commodities), asset endowment (land, buildings, livestock), loans,
education status and expenditure on education, work status by gender,
and small scale entrepreneurial activities. The sample consists of 16182
randomly selected households. The sample is restricted to households
whose income is greater than Rs 1000 per month. Households whose
consumption data is missing have been dropped. Out of this sample of
15924, 802 households (5 percent of the total) are remittance
beneficiary households (RBH) and 15122 are nonremittance beneficiary
households (NRBH). Table 2 in Appendix I presents the set of variables
along with their definition that have been included in the analysis.
The major characteristics of households have been presented in
Tables 3-5 in Appendix I. The geographic distribution of the RBH show
that majority of migrated households are located in two provinces of
Pakistan, Punjab and Khyber Pakhtunkhwa, 33.2 and 30.7 percent,
respectively (Table 3 in Appendix 1). However, RBH are largely from
rural areas -56.1 percent of the total (Table 4). This implies that
migrated labour can largely be categorised as unskilled or low skilled
labour. The majority of migrants consist of unskilled and semiskilled
workers, i.e., 52.24 percent, while highly qualified migrants are only
2.52 percent in 2007 [Siddiqui (2011)].
The average size of the households is 7.2 individuals with average
age of head of the household being 45.7 years having education of 4.2
years (Table 5 in Appendix I). The income per adult equivalent per year
is Rs 28063.7. Food expenditure is high relative to non-food expenditure
consuming 3732 calories per day per adult. Households own household
equipment worth Rs 19851.5. The human capital accumulation indicated by
the education level of currently school going children is 7.2 years with
very low average for the household education level of 2.7 years. Average
expenditure on education of children currently going to school is Rs
3807.9 per year.
The living condition is not good -2.5 adults / room. On average,
76.8 percent of households have tap water and 35.6 percent have access
to sanitation facilities, 69.9 percent have electricity. Average gas and
telephone facilities are very low as a whole-21.3 and 12.1
percent--respectively. Households, on average, own assets including
residential and commercial buildings, and land worth Rs 0.35 million.
They own 1.4 acres of land per household. Household save on jewellery
purchases and cash worth Rs 16619.6 and Rs 10355.5 respectively and they
owe money amounting to Rs 29814.9 and receive profit on bank deposits of
Rs 332. Entrepreneurial activity is low as households, on average, hold
0.2 enterprises. Two employed persons per household indicate a
dependency ratio of 3.6 per earner. With poverty line for rural and
urban areas at Rs 705 and Rs 850 expenditure per adult per month,
respectively, the poor households are 39.1 percent and 29.8 percent of
the total in rural and urban areas in 2001-02.
4. DISTRIBUTION OF THE TREATMENT AND COMPARISON SAMPLES
First, a binary logistic function [Equation 2] is tested to
calculate the probability that a household receives remittances. The
results are reported in Table 1. The results show that a majority of
variables are significant at the 5 percent level.
Second, the paired t-test is employed to examine whether the mean
of each element of X vector for the treatment is equal to that of the
matched sample. The results show that prior to matching, the difference
between the mean values of explanatory variables of the two groups was
very significant, but the difference becomes insignificant for all
variables after PSM (see Table 2). This indicates that the distribution
of the covariates is approximately the same across the RBH and NRBH.
Prior to matching, the comparison of the mean values of the
indicators across the treated (RBH) and control group (NRBH) reveal a
crude difference (that includes difference due to observed
characteristics) in living standards. The results show that household
size is larger for RBH i.e., 7.7 compared to 7.2 of the NRBH, the larger
family size indicates the need for migration (Table 5 in Appendix I). On
average, the head of the households is older with higher education level
in the RBH. Treated units receiving remittance income have higher
expenditure per adult per year compared to NRBH. Their expenditure on
food is lower than expenditure on non-food item in contrast to the
expenditure pattern of NRBH. The human capital indicators support the
positive relationship of remittance income and human capital formation
[see calorie intake, average class of school going children at present,
and expenditure on education per year]. On average, RBH households own
equipment that 2.7 times higher in worth than NRBH. RBH have 3.4 rooms
per household compared to 2.4 rooms for NRBH. RBH own houses with more
facilities such as electricity, safe drinking water, and sanitation. The
higher percentage of RBH also has gas and telephone facilities. All
these indicators show higher standard of living of treated units
compared to non-treated ones. The RBH hold larger assets which include
residential buildings, non-residential buildings and livestock, have
more cash and jewellery and are less indebted. The profit on bank
deposits of RBH is about three times higher than that of NRBH. On
average, they hold fewer acres of land holdings. The results support the
view that remittances have positive impact on housing and consumer
durables and non-land assets [Quisumbing and McNiven (2007)]. Low
entrepreneurial activities among RBH do not support the growth impact of
remittances. It may affect growth through credit expansion. However, the
poverty impact of remittances is very strong with only 5 percent of RBH
being below the poverty line compared to 23.9 percent of NRBH. This is a
naive valuation approach that overstates the remittances impact as the
difference in mean value which includes the impact of the difference in
observables characteristics.
5. RESULTS
Table 3 reports the difference in the mean value of socio-economic
indicators of the treated and control groups of households under three
definitions. First, the differences in the mean values of socio-economic
indicators of the treated and control groups are calculated using all
observations. Second, these differences are calculated based on a set of
observations from common support area. Third, households that minimise
the difference between PS of the two groups--treated (RBH) and control
(NRBH)--are matched. The differences are tested statistically using
t-ratios. These results are compared by taking the difference in
difference of mean of first and third exercise to find the bias in the
estimated values if the endogeneity problem and difference in observable
characteristics are ignored.
In Table 3, Column 1 and 2, respectively, the mean differences in
socio-economic indicators are reported which are based on the whole
sample of the treated or remittances receiving households (802) and the
control group consists of all households who do not receive remittances
(15924) and their t-values. Column 3 and 4 present the results for the
trimmed sample (common support area) with the sample of 802 and 15122,
respectively, for beneficiary and non-beneficiary households. Finally,
the 5th and 6th columns present the average treatment (remittances)
effect on the treated (RBH) after exact matching of propensity score
with control group (NRBH) which minimises the difference between treated
and control groups of households after controlling for observables'
characteristics along with their t-ratio to measure significance.
The results of this impact evaluation reveal that RBHs are in
better position than NRBH. The results show that difference in the mean
income per adult equivalent is negative but not significant in the first
two exercises. After exact matching, this difference in mean values
becomes positive and significant (Col. 5 and 6). This proves the theory
that workers migrate to take the advantage of higher wages. The results
show that households with same qualification and social background earn
higher income abroad than in the domestic country. In all the three
exercises, the RBH have higher expenditure per adult equivalent but the
difference is minimum when the PS, i.e. the exact matched samples have
been used. This result also holds for food expenditure, calorie intake
and non-food expenditure. After exact matching of RBH and NRBH, the
difference in expenditure per adult equivalent reduces to Rs 8619.3-24
percent.
With reference to human capital indicators, RBH appear to be better
educated. The difference in education level of children currently going
to school reduces from 4.4 classes to 2.4 classes. This is also
reflected in average expenditure per class. Like previous studies, the
results support the hypotheses that remittances have positive impact on
human capital accumulation. However, results also show that if
differences in the observable characteristics are not controlled, the
impact would be 43 percent and 45.9 per cent larger over the actual
impact on human capital accumulation. The results may misguide policy
makers if issues of endogeneity and counterfactual are ignored.
Other differences are associated with ownership of durable goods
and other amenities of life. On average, RBH households own more
equipment than the NRBH. A higher proportion of the RBH has access to
electricity, telephone facilities and room per adult equivalent than the
NRBH. However, gas and tap water facilities are not significantly
different in both groups may be because of lack of public
infrastructure.
In case of different types of physical capital accumulation, the
results suggest that remittances do not have a statistically significant
impact on the accumulation of livestock, land holdings, jewellery, and
loans (Table 3) but have higher assets of residential buildings, cash
holdings (significant at 5 percent) and profit receipts from banks
(significant at 10 percent level). (20) These results make the role of
remittances in generating economic growth doubtful. Some of these
results confirm the earlier findings of Amjad (1988); Gilani (1981) and
Arif (1999) that migrant households invest in housing but reject that
they are used for land, jewellery, and repayment of loans. However, the
results are not comparable as earlier studies did not take into account
counterfactual.
The level of female and male economic activity in RBH is
significantly lower than in the NRBH. This suggests that both men and
women in the households are less likely to work if they receive
remittances. This may also imply that the control group of households
are relatively poor and women are forced to work to meet their basic
needs. Men's lower economic activity in RBH is self-evident since
it is they who are working abroad. The lower participation of both men
and women also indicates the loss of production due to migration.
Non-agriculture establishments (enterprises) are largely owned by
non-migrant households or NRBH. The difference between the two groups is
significant. This indicates that remittances are not invested in
productive enterprises and the hypotheses that remittances influence
growth is not correct. The results of earlier studies by Gilani, et al.
(1981), Tinsabad (1988) for Thailand, and Rodrigo and Jayatissa (1988)
for Sri Lanka show that remittances are used for non-agriculture
investment by 8.2 percent, 29.5 percent, and 3.6 percent, respectively.
Therefore, government should promote local businesses so that households
predominantly engaged in consumption or unproductive investment have the
option to engage in productive activities.
These results show that the living standard of remittance receiving
households is higher than that of the non-treated group. But, if we
ignore the difference in observable characteristics, we overstate the
impact extensively. This leads us to conclude that remittances raise the
standard of living. The results associated with basic needs such as
calorie intake, housing, sanitation facilities etc. also have a role in
poverty reduction. Poverty, measured by head count ratio, shows that
among the RBH would be, on average, 0.1 points lower than among the NRBH
i.e., a difference of 0.1 point (p.d) between the two groups. This
finding is supports the earlier finding by Siddiqui and Kemal (2006),
which shows that remittances reduce poverty by 0.1 percent over the base
year with one percent increase in remittances. The difference in poverty
is of 0.2 points when we compare poverty of two groups--all treated and
all non-treated. This method overstates the impact of remittance on
poverty. PSM overcomes the bias problem cutting the impact down to 0.1
p.d--a reduction of 33 percent. The bias in other estimates can be
observed from the last column of Table 3 which shows that the estimates
are biased upward. If one ignores the issues of selection and
differences in observable characteristics, remittances would look like
having a greater than actual impact (see Table 3) which is likely to
result in wrong policies. The last column of Table 3 shows that the
existing literature measuring the impact of remittances belongs to the
first group. This study for Pakistan is the first which evaluates the
impact of remittances overcoming the problem of endogeniety and
counterfactuals and provides an experimental benchmark. Therefore, the
results of earlier studies need careful consideration if used for policy
formulation.
6. HETEROGENEITY IN IMPACT OF REMITTANCES BY EDUCATION LEVEL
It is important to examine heterogeneity in treatment effect on
socioeconomic aspects of households grouped by education level using a
methodology that renders an experimental benchmark. Here households are
defined in two strata on the basis of education of the head of the
household--low skill (less than 10 years) and high skill (10 years and
above). In each group, households are further classified into three sub
groups. In the lower strata of education (below matriculation) three
skill levels are classified as: L-LS (less than one year of education),
L-MS (1-4 years of education) and L-HS (5-9 years of education). In the
upper strata of education [matriculation and above] three groups are
classified as: H-LS (10-13), H-MS (14-15), H-HS (16 years and above).
(21)
The overall results show that the relatively poorer group of
households (first four groups) register larger gain from foreign
remittances in terms of income and expenditure per adult equivalent
which increases with the education level except for L-MS. In this group
(L-MS) the difference in the income is significant at 10 percent level.
Poverty reduces the most among household groups in lower strata where
education of the head of the household is below matriculation. In the
upper strata, income expenditure and poverty impact are observed in
households classified as low skill (10-13 years of education). The other
two groups do not register any significant impact of remittances. These
groups belong to the richest group of households.
The poverty effect of remittance is estimated to be 0.1-point
difference (p.d) for the whole group in the aggregate analysis. The
poverty reduction effect is estimated to be -0.2 p.d for L-LS and L-HS,
and larger than average effect for the whole group (-0.1 p.d.). The
poverty impact decreases (in absolute term) from -0.2 p.d. to -0.1 p.d.
for below matriculation to above matriculation group. This finding is
consistent with the findings of Siddiqui and Kemal (2006), which show
that poverty impact is larger among relatively poor households and has a
smaller impact on relatively rich households. There is no poverty impact
for the richest group of households. This does not imply that migration
is an irrational decision for these groups of households. Some earlier
studies show that the positive effect of migration is not realised until
five or six years after the original migration Ham, et al. (2005). The
initial returns are not significant. Siddiqui (2011) shows that
migration of skilled labour has increased in recent years. So the
benefits have not been significantly realised yet, or the sample of
these households is very small.
The detailed results for these households are presented in Table 6
in Appendix I. The results show that the impact of remittances is still
positive in terms of income, expenditure and all types of capital
accumulation for the households with less than one year of education. In
the upper strata, households with education of matriculation and above,
i.e., with 10-13 years of education benefit. The other two groups show
significant positive impact only on [profit from bank at 10 percent
level of significance] and [expenditure on education, room occupancy],
respectively [see Table 6 in Appendix I], From this it can be concluded
that two households in the upper strata belong to richer group of
households and do not register the impact in the basic needs'
variable. But the impact is significant in bank accounts. However,
insufficient data for these groups may be the major reason for the
insignificant results. The overall results show that aggregate analysis
hides the variation in impact by education level.
7. COMPARISON WITH EARLIER STUDIES
The main difference between the results of this study and the
earlier studies is that the change in outcome indicators in this study
is unlikely to be correlated with the migration decision, while in the
earlier studies it is correlated. The characteristics that influence the
migration decision are likely to influence the decision of other
households. The majority of earlier works do not take into account the
issues of selection and differences in observable characteristics.
Therefore they are likely to overestimate the impact.
Empirical estimates from earlier studies are compiled in Table 7 in
Appendix 1. The table reports major results along with data and
methodology used in the analysis. It shows that disparities in
estimation techniques and data affect the conclusion. It also shows that
more than 90 percent labour migrated from Pakistan, India, Sri Lanka,
and Bangladesh to Middle East in the 1980s.
Income and Consumption: Many studies conducted in the 1980s,
especially in South Asian countries, have focused on the use of
remittances based on existing migration data collected at the household
level. Amjad (1988), Kazi (1988) and Gilani, et al. (1981), for Pakistan
have found that migrant households allocate about 63 percent of
resources to total consumption expenditure and 56.8 percent when
remittances through unofficial channels are also included. While the
results of the present study show that RBH spend 44 percent on food
measured in per adult equivalent term, which is lower than the food
expenditure of NRBH at 53 percent. Similarly, Gilani, et al. (1981) show
that RBH allocate 62 percent of their remittances to recurrent
consumption (57 percent), durable goods (2.8 percent) and other
expenditure (2.3). The results of these studies do not compare the
expenditure pattern with counterfactual or control group expenditure.
Hence their findings cannot be used to conclude that remittances
contribute to higher consumption or lower consumption. Malik and Sarwar
(1993) overcome this problem and estimate demand functions for three
types of consumption expenditure--total consumption expenditure,
recurrent consumption and expenditure on durable goods for RBH and NRBH
for various regions of Pakistan. The study concludes that the
expenditure pattern is different for migrant and non-migrant households.
But the study ignores the differences in observable characteristics and
estimate the function by using the whole sample. The result of the
present study shows that estimates are biased if the difference is
measured using all migrant and non-migrant households. The difference in
consumption of RBH and NRBH decreases by 25 percent in total consumption
and in expenditure on durables, while the expenditure on food decreases
by 24 percent. Therefore it is necessary to use a methodology which at
least minimises if not eliminates the bias.
Empirical evidence shows that more than 50 percent of the migrated
labour to Middle East were unskilled labour. Mahmood (1988), Hyun
(1988), and Tan and Canlas (1988) show that migrated labour from
Bangladesh, Korea and Philippines are earning three to six times higher
than wages in their country of origin. If we control for the selection
bias and the observable characteristics, the difference in income (22)
becomes significant and positive (see Table 3).
Welfare, Poverty and Inequality. Higher income and consumption are
expected to reduce poverty and inequality. Siddiqui and Kemal (2006)
show that remittances reduce poverty and improve welfare by 0.01 percent
and 0.06 percent, respectively. Rodrigo and Jayatissa (1988) show that
inequality increases with remittance inflow. The study by Jongwanich
(2007) using cross country data shows that remittances through direct
and indirect channels reduce poverty by 0.03 percent. The results of the
present study--reduction in poverty by 0.01 percent--match with the
results of Siddiqui and Kemal (2006) who show that the poverty impact
varies by type of household, rich and poor. These results are confirmed
by the results of this study as the impact varies by type of households
i.e., the poor register larger impact. However, the results are not
exactly comparable as the base year is different. This study shows that
bias reduces poverty estimates by 33 percent, when we correct for the
selection bias and observable characteristics.
Investment: The studies show that investment in real estate, land,
and housing are higher for RBH and ranges between 20.7 and 35.4 percent
for Pakistan. Another study for Thailand shows that 33 percent migrant
households own houses compared to 20 percent non migrant households.
Overall, 75 percent migrant households own assets compared to 39 percent
non migrant households. The studies also measure the impact of
remittances or their allocation to different types of assets, physical,
financial, and human (see Table 7) but the fact remains that they
over-estimate the impact due to uncontrolled difference in observable
characteristics or counterfactuals. This type of analysis does not
measure the sole benefits of migration or remittances, but also include
the effects of uncontrolled differences in socio economic
characteristics of households.
Growth: Burney (1987) using demand composition and Iqbal and Sattar
(2005) using the Chami, Fullenkamp, Jahjah model show positive
relationship between growth and remittances. Burney (1987) shows that
the contribution of official remittances from Middle East to GNP growth
was 13.6 percent during 1973-4 to 1976-7. This contribution increased to
24 percent when remittances through unofficial channels were also taken
into account. Iqbal and Sattar (2005) show that increase in remittances
by one percentage point increase growth by 0.44 percentage point.
Jongwanich (2007) estimates the neoclassical model of Barro using cross
country data. The study could not find any direct and significant impact
on growth, but indirect effect of remittances on growth works through
human capital investment (0.02 percent) and physical capital (0.01
percent). The growth impact remains inconclusive in the present study.
The results show that remittances affect human capital accumulation
positively, which have a growth promoting impact [Jongwanich (2007)]. In
addition higher bank deposits also point to growth promoting effects
through the banks' intermediary role i.e. credit expansion. But low
entrepreneurial activity and less land holding shows the opposite.
This writer constructed a table of expected outcome indicator of
remittances impact giving value of '1' if a study includes the
indicator and zero if it ignores, measuring data quality with
l=secondary, 2=primary, 3=data on both treated and control groups.
Similarly the methodology is ranked as 1 if data uses only descriptive
statistics, and 2 if both descriptive and statistical estimation
analysis are used, while 3 means rigorous. An index based on the
information has been developed. An ideal situation (hypothetical) is
when comprehensive data with treated and control groups is used, for
rigorous impact evaluation to measure the impact on all expected outcome
indicators. A comparison of the ideal study with the previous empirical
studies shows deviation from the optimal analysis (see Figure 1). The
figure shows that majority of studies divert from optimum evaluation
level. There is a need to motivate researchers to conduct impact
evaluation using method which reduces biases if not eliminate them in
the impact and renders an experimental benchmark.
[FIGURE 1 OMITTED]
8. CONCLUSION
Given the multi dimension impact of remittances, its integration
into overall development planning is essential. For that purpose, it is
required to conduct a comprehensive analysis using the most appropriate
techniques to draw lessons for suitable policies. Rosenbaum and Rubin
(1983) indicate that robust estimates can be obtained by overcoming the
problem of selection bias and difference in observable characteristics
using PSM and difference method which replicates experimental bench mark
in self-select cases.
This paper contributes to the literature of remittance in Pakistan
by analysing the impact of remittances using the propensity score
matching and difference method. The major finding of the study is that
robust estimates that take into account both selection and endogeniety
problems in estimating the average impact of remittances are
substantially different from the estimates which disregard these issues
and so overstate the actual impact. A comparison of impact corrected for
selection with those where it is not shows a very large and significant
bias. In policy-making it is the unbiased results that are needed.
The paper quantifies the benefits of migration (remittance), in
terms of income, expenditure, savings, human capital and physical
capital accumulation, poverty, and growth. After balancing for the
differences in observable characteristics, migration is found to be
beneficial. The number of migrant households with income levels below
poverty line reduces by 0.1 points over non-migrants in the aggregate.
Their higher human and physical capital ownership, savings in cash and
profit from bank point to the growth promoting impact of remittances,
whereas lower entrepreneurial activity and low men and women
participation in economic activities in the RBH group illustrate the
opposite. The growth impact of remittances therefore remains
inconclusive.
Aggregate analysis hides heterogeneity in impact by education level
and underestimates/over-estimate the effect for poor/rich households.
The results show remittances have significant impact on poor households
(with less than one year of education) but have no impact on highly
educated households. Therefore, matching is a useful way to control for
observable heterogeneity too.
The pattern of use of remittances determines the impact on poverty
and growth. Therefore, if the objective is to achieve higher growth, the
remittances can be redirected from current consumption towards
productive investment by offering higher interest rate on deposits or
subsidies for productive investment. However, further analysis requires
more demographic and economic information on migrants and return
migrants, their stay abroad, how they send money back home, over what
period of time and from where. That analysis would be helpful to devise
migration policies for poverty reduction and growth enhancing
strategies.
APPENDIX I
Table 1
Impact in the Log Frame: Remittances Inflow from Abroad
Level Indicators
Activities Migration
Input Remittances
Outputs 1. Accumulation of Capital
Stock: Human, Physical and Financial.
Intermediate 1. Better nutrition
Outcomes 2. Higher enrolment
3. Higher physical capital stock
4. Higher Bank Deposits
Final Outcomes Improved Social and Economic Indicators:
literacy rate and health status and growth
Short Run Impact 1. Reduce poverty
2. Improve welfare of households
Long Run Impact Higher Productivity and Earnings
Table 2
Detail of Variables Used in the Analysis
Variable Name Definition
1. Remittances Households' remittance income from
abroad in (Rs).
2. Income per Adult Household income from all
sources-domestic and foreign,
divided by number of adult
equivalent (Rs).
3. Total Expenditure per Adult Total households expenditure
divided by number of adult
equivalent (Rs).
4. Expenditure on Food per Adult Food expenditure per adult
equivalent in Rs.
5. Calorie Intake Calculated by multiplying quantity
of good consumed with calorie per
unit.
6. Expenditure on Non-food Items Non Food Expenditure in Rs per
per Adult year per adult equivalent
7. Expenditure on Durables such Expenditure on durables per adult
as Clothing and Footwear equivalent per year
8. Expenditure of Education per Households Expenditure on
Class Education divided by level
(classes) of school going
individuals
9. Average Class of School Going Total number years of schooling of
Children currently going to school children
divided by number of school going
children
10. Household Size Number of households members
11. Females Economic Activity Female Employment
12. Education of the Head of Highest level of Education of head
the Household of the household
13. Capital Stock Accumulation
13a. Iluman Capital Measured by education of currently
going to school (years of
schooling), average level of
education of households and
expenditure on Education per
class.
13b. Physical Capital Asset: Buildings (completed or
under construction),- land,
residential buildings, commercial
buildings
13c. Equipment Durable goods: Tangible asset
accumulation such as refrigerator.
TV, automobile and other durables.
13d. Financial Capital Profit on Bank Deposits measure
size of deposits
13e. Savings Jewellery and Cash
14. Poverty Plead Count Ratio, Percentage of
population below poverty line
14.a Poverty Line Rural and urban poverty line are
calculated based on the assumption
that the gap between rural and
urban poverty line is same as in
1990 Official national poverty
line is used to calculate poverty
line for rural and urban areas.
Poverty lines are Rs 748. Rs 850
and Rs 705 for Pakistan. Urban and
Rural areas, respectively
15. Household Condition
(measured by amenities)
15a. Electricity Electricity direct connection
15b. Gas Gas direct connection
15c. Tap Water Piped, Fland Pump. Tube well
direct
15d. Sanitation Facilities Flush connected to public
sewerage. Flush connected to pit
15e. Telephone Telephone direct connection
15f. Occupancy Room per adult
Table 3
Geographic Distribution (%)
Control(NRBH) Treated(RBH) Total
Punjab 39.8 33.2 39.4
Sindh 24.2 4.0 23.2
KP 15.9 30.7 16.6
ROP 20.1 32.2 20.8
Total 100 100 100
Source: Author's Calculations.
Table 4
Distribution of RBH and NRBH by Region
Urban Control Treated Total
Punjab 39.5 57.9 40.3
Sindh 40.8 41.0 41.3
KP 31.3 32.9 31.4
ROP 28.0 32.9 28.4
Total Urban 36.2 43.9 36.6
Rural
Punjab 60.5 42.1 59.7
Sindh 59.2 0.0 58.7
KPK 68.7 67.1 68.6
ROP 72.0 67.1 71.6
Total Rural 63.8 56.1 63.4
Pakistan 100.0 100.0 100.0
Source: Author's Calculations.
Table 5
Mean Values of Output and Outcome Variables
Full
Variables Sample Treated Control
No. of Observation 15924.0 802.0 15122.0
H-size 7.2 7.7 7.2
Age 45.7 48.6 45.5
Education of Head of the Household 4.2 4.5 4.2
Remittances per adult per year 778.1 15450.0 0.0
Households in a district 155.4 175.0 154.4
Income per Adult 28063.7 27714.3 28082.2
Expenditure per Adult 16053.0 26958.1 15474.6
Food intake per adult per year 8865.5 12344.9 8681.0
Non-food consumption (durables) 7187.4 14613.2 6793.6
per adult per year
Calorie intake per adult per day 3732.2 4741.7 3678.7
Average class of currently going 7.2 11.4 7.0
to School
Expenditure on education per year 3807.9 9271.8 3518.1
Average class of households 2.7 3.9 2.7
Household Equipment 19851.5 49862.1 18259.9
Room per households 2.4 3.4 2.4
Electricity 69.9 88.7 68.9
Gas 21.3 22.0 21.3
Telephone 12:1 38.5 10.7
Tap water 76.8 72.3 77.1
Toilet 35.6 58.9 34.4
Asset 351314.6 882440.2 323146.2
Livestock 0.2 0.2 0.2
Land Ownership 1.4 0.8 1.5
Jewelry 16619.6 149963.3 9547.6
Cash 10355.5 28763.3 9379.3
Loan 29814.9 28774.0 29870.2
Bank Deposit profit 332.0 1039.8 294.5
Employment 1.9 0.9 1.9
Enterprises 0.2 0.2 0.2
Poverty based on expenditure per 23.0 4.99 23.9
adult equivalent
Source: Author's Calculations.
Table 6
Impact Evaluation of Remittances by Education Level
Less than one Year
Education level Outcome/ Difference T-
Output Indicators in mean Statistics
1. Income per adult equivalent 6374.36 7.99
2. Expenditure per adult
equivalent 5396.05 7.94
3. Expenditure on Food per
Adult per Year 1685.03 5.64
4. Calorie intake per adult per
year 770.47 1.96
5. Non food consumption
(durables)per adult per year 3711.02 7.54
6. Expenditure on education 3311.67 6.09
7. Average class of Households 1.02 4.55
8. Average class of currently
going to School children 4.34 6.10
9. Household Equipment 13742.28 5.97
10. Room Occupy 0.86 7.61
11. Electricity 0.16 4.94
12. Gas -0.01 -0.58
13. Telephone 0.21 8.61
14. Safe Drinking Water 0.05 1.39
15. Sanitation facilities 0.18 5.43
16. Asset 246351.34 5.79
17. Livestock 0.21 1.50
18. Land holdings (acres) 0.00 0.03
19. Jewelry (RS) 251106.20 1.06
20. Saving in Cash (Rupees) 13687.21 3.84
21. Loan 23780.64 1.77
22. Profit on bank deposits -76.19 -0.79
23. Men employed -0.51 -6.06
24. Women employed -0.14 -2.96
25. Employed total -0.65 6.21
26. Enterprises -0.08 -2.37
27. Poverty (Head Count Ratio) -0.15 -5.84
1-4 year
Education level Outcome/ Difference T-
Output Indicators in mean Statistics
1. Income per adult equivalent 4725.54 1.73
2. Expenditure per adult
equivalent 637.83 0.23
3. Expenditure on Food per
Adult per Year 664.59 0.60
4. Calorie intake per adult per
year -140.26 -0.08
5. Non food consumption
(durables)per adult per year -26.77 -0.01
6. Expenditure on education 3333.83 1.98
7. Average class of Households 0.37 0.65
8. Average class of currently
going to School children 1.86 0.83
9. Household Equipment 7194.39 1.25
10. Room Occupy 0.52 1.54
11. Electricity -0.12 1.68
12. Gas 0.02 0.20
13. Telephone 0.09 0.95
14. Safe Drinking Water 0.00 0.00
15. Sanitation facilities -0.02 -0.20
16. Asset 237325.76 0.97
17. Livestock -0.36 -0.80
18. Land holdings (acres) -4.85 -0.80
19. Jewelry (RS) 5470.45 0.46
20. Saving in Cash (Rupees) 37196.97 1.30
21. Loan 43602.41 1.54
22. Profit on bank deposits 795.45 1.16
23. Men employed -0.28 -0.88
24. Women employed 0.00 0.00
25. Employed total -0.28 -0.68
26. Enterprises -0.20 -1.75
27. Poverty (Head Count Ratio) -0.09 -1.12
5-9 year
Education level Outcome/ Difference T-
Output Indicators in mean Statistics
1. Income per adult equivalent 8011.45 248
2. Expenditure per adult
equivalent 9212.72 4.95
3. Expenditure on Food per
Adult per Year 2476.38 3 52
4. Calorie intake per adult per
year 87 90 0.10
5. Non food consumption
(durables)per adult per year 6736.34 4.85
6. Expenditure on education 5507.90 2.89
7. Average class of Households 0.59 1.44
8. Average class of currently
going to School children 2.92 1.75
9. Household Equipment 51042.25 2.11
10. Room Occupy 0.73 3.68
11. Electricity 0.08 2.55
12. Gas 0.04 0.73
13. Telephone 0.35 7.16
14. Safe Drinking Water 0.10 1.98
15. Sanitation facilities 0.15 2.68
16. Asset 646653.19 1.45
17. Livestock 043 1.97
18. Land holdings (acres) 0.02 0.06
19. Jewelry (RS) 21636.55 4.39
20. Saving in Cash (Rupees) 24624.74 1.65
21. Loan -14039.49 -0.76
22. Profit on bank deposits 530.70 1.32
23. Men employed -0.43 -3.20
24. Women employed -0.10 -1.77
25. Employed total -0.53 -3.49
26. Enterprises -0.10 -1.59
27. Poverty (Head Count Ratio) -0.16 -4.39
10-13
Education level Outcome/ Difference T-
Output Indicators in mean Statistics
1. Income per adult equivalent 21040.96 3.51
2. Expenditure per adult
equivalent 18456.22 3.57
3. Expenditure on Food per
Adult per Year 5719.69 4.77
4. Calorie intake per adult per
year 948.61 0.96
5. Non food consumption
(durables)per adult per year 12736.53 3.00
6. Expenditure on education 2179.34 1.49
7. Average class of Households 0.25 0.45
8. Average class of currently
going to School children -1.42 -0.84
9. Household Equipment 28240.48 2.01
10. Room Occupy 0.11 0.44
11. Electricity 0.01 0.40
12. Gas 0.02 0.36
13. Telephone 0.25 4.41
14. Safe Drinking Water -0.08 -1.78
15. Sanitation facilities 0.08 1.62
16. Asset 268842.66 0.84
17. Livestock 0.01 0.05
18. Land holdings (acres) -0.36 -0.77
19. Jewelry (RS) 26957.60 1.92
20. Saving in Cash (Rupees) 33712.75 2.55
21. Loan -8412.13 -1.26
22. Profit on bank deposits 1822.15 1.42
23. Men employed -0.65 -5.06
24. Women employed -0.15 -2.00
25. Employed total -0.80 -4.90
26. Enterprises -0.19 3 12
27. Poverty (Head Count Ratio) -0.09 -2.93
14-15
Education level Outcome/ Difference T-
Output Indicators in mean Statistics
1. Income per adult equivalent 13569.31 1.01
2. Expenditure per adult
equivalent 11935.50 0.77
3. Expenditure on Food per
Adult per Year 5642.87 1.51
4. Calorie intake per adult per
year 665.87 0.23
5. Non food consumption
(durables)per adult per year 6292.63 0.51
6. Expenditure on education 12313.57 0.72
7. Average class of Households 0.54 0.42
8. Average class of currently
going to School children -1.49 -0.32
9. Household Equipment 54.49 0.00
10. Room Occupy 0.34 0.75
11. Electricity -0.04 -1.00
12. Gas 0.00 0.00
13. Telephone 0.02 0.17
14. Safe Drinking Water 0.05 0.50
15. Sanitation facilities 0.01 0.08
16. Asset 979637.04 1.11
17. Livestock 0.02 0.08
18. Land holdings (acres) -0.68 -0.89
19. Jewelry (RS) 7958.52 0.83
20. Saving in Cash (Rupees) -118451.85 -0.97
21. Loan -16068.15 -1.38
22. Profit on bank deposits 1814.81 1.68
23. Men employed -0.30 -1.29
24. Women employed -0.17 -1.64
25. Employed total -0.47 1.87
26. Enterprises -0.01 -0.08
27. Poverty (Head Count Ratio) 0.00 0.05
16 and above
Education level Outcome/ Difference T-
Output Indicators in mean Statistics
1. Income per adult equivalent 22259.93 0.98
2. Expenditure per adult
equivalent 13012.64 0.56
3. Expenditure on Food per
Adult per Year 6204.74 1 63
4. Calorie intake per adult per
year 2930.88 0.83
5. Non food consumption
(durables)per adult per year 6807.90 0.34
6. Expenditure on education 13585.78 1.98
7. Average class of Households -0.29 -0.24
8. Average class of currently
going to School children -1.62 -0.37
9. Household Equipment -54404.74 -0.77
10. Room Occupy 0.70 1.69
11. Electricity 0.05 1.00
12. Gas -0.06 -0.46
13. Telephone 0.14 i.n
14. Safe Drinking Water -0.04 -0.40
15. Sanitation facilities 0.15 1.60
16. Asset 11444.81 0.02
17. Livestock -0.57 -2.12
18. Land holdings (acres) -0.36 -0.47
19. Jewelry (RS) 27948.05 1.45
20. Saving in Cash (Rupees) 6919.16 0.15
21. Loan -15250.00 -0.82
22. Profit on bank deposits 4175.32 0.68
23. Men employed -0.69 -2.58
24. Women employed 0.10 0.79
25. Employed total -0.58 -1.95
26. Enterprises -0.04 -0.36
27. Poverty (Head Count Ratio) 0.00 0.00
Source: Author's Calculations.
Table 7
Empirical Estimates from Existing Literature Focus on Data Results
Growth (5,6,21) TS official 0.44 to one
(unofficial) = percentage
1.49 (2.89) to point of
6.59 (11.01) in remittances
1970 and 1980s
Earning 469 HH survey 223.6
Estimate of
non Migrant
Per Capita
Remittances (1)
Earnings ARTEP 4908Rs/Month
(remittances)
r(8)
Remittances ARTEP/ILO 2589 Rs/m 27083
(2,4) (20416)
Wage ratio 2.65, Bangladesh
after migration = 5.77
/Domestic
(13,15,19)
Variation in 469 HU survey 1-13.8%
Remittances
income = ratio
of poorest/
richest 20%,
(1)
Consumption Total 63.3% -56.8% 62.19
out of remit/ recurrent 53.50% 57.00
Share of marriages 9.80% 2.35
consumption Consumer included 2.84
(2,7,9,8, Durables in recurrent
18,20,99)
Real Estate Total 35.40% 21.68
Construction/ 12.14
Purchase of
Residential
House
Improvement in 2.27
House &
Construction
Commercial 5.72
Real Estate
Agriculture Land 1.55
Investment/ Total 24.2% saving 12.95
Saving, after/
Before (2)
Agricultural 3.3
Investment
Industrial/ 8.21
commercial
Investment
Financial
Investment/
Saving = foreign
currency account
Residual 8.5 318
Human Capital 0
Povertv(12, 22) 0 01,0.03 2275752.63
Welfare(12) -0.06 0
average cost of 1534 in $ 1983 38979 to
migrant(15,20) 43518 baht
Growth (5,6,21) TS Human
capital
= 0.02
Physical
Investment
= 0.01
Earning 469 HH survey
Estimate of
non Migrant
Per Capita
Remittances (1)
Earnings ARTEP
(remittances)
r(8)
Remittances ARTEP/ILO 5909.00
(2,4)
Wage ratio Philippines
after migration = 6.35
/Domestic
(13,15,19)
Variation in 469 HU survey
Remittances
income = ratio
of poorest/
richest 20%,
(1)
Consumption Total 0.57m 52.1
out of remit/ recurrent 0.52 nm
Share of marriages 0.53m 2.9
consumption Consumer 0.026m 5.9
(2,7,9,8, Durables 0.03 nm
18,20,99)
Real Estate Total
Construction/
Purchase of
Residential
House
Improvement in 14.2
House &
Construction
Commercial
Real Estate
Agriculture Land 15.6
Investment/ Total 35 1
Saving, after/
Before (2)
Agricultural
Investment
Industrial/ 29.5
commercial
Investment
Financial
Investment/ 14.2
Saving = foreign
currency account
Residual 0.003 loan = 4 3%,
Jewellery =
Human Capital 5.1
Povertv(12, 22) Education =
Welfare(12) 2.4,
average cost of Health = 5.9
migrant(15,20)
Growth (5,6,21) TS
Earning 469 HH survey
Estimate of
non Migrant
Per Capita
Remittances (1)
Earnings ARTEP
(remittances)
r(8)
Remittances ARTEP/ILO
(2,4)
Wage ratio
after migration
/Domestic
(13,15,19)
Variation in 469 HU survey
Remittances
income = ratio
of poorest/
richest 20%,
(1)
Consumption Total 6.99/ 57%
out of remit/ recurrent (0.55) (99)
Share of marriages
consumption Consumer 11.04
(2,7,9,8, Durables
18,20,99)
Real Estate Total
Construction/
Purchase of
Residential
House
Improvement in 33.13/20.73
House &
Construction
Commercial
Real Estate
Agriculture Land 6.12/225.24
Investment/ Total M/NM = 75/
Saving, after/ 39 Asset
Before (2) Ownership
Agricultural
Investment
Industrial/ Year = 1981
commercial 5.41 (6 65)
Investment = transport
Financial equipment,
total
invest
3.61
Investment/ 13.42/2.20
Saving = foreign
currency account
Residual loan = Loan =
23.54. 2.6%(99).
Human Capital Jewellery Jewellery =
Povertv(12, 22) = 0.26 26.8(99)
Welfare(12) Other
average cost of saving =
migrant(15,20) 5.1%
Sources: 1. Adams (1998), 2. Amjad (1986), 3 Amjad (1988), 4 Arif
(1999), 5. Burney (1988) 6. Iqbal and Sattar (2005), 7. Gilani,
et al. (1981), 8 Kazi (1988), 9. Malik and Sarwar (1993), 10.
Maqsood and Sirajeldin (1994), 11. Nishat and Bilgrami (1993), 12
Siddiqui and Kemal (2006), 13. Hyun (1988), 14 Jongwanich (2007),
15. Mahmud (1988), 16. Nayar (1988) 17. Quisumbing and McNiven
(2007), 18 Rodrigo and Jayatissa (1988), 19. Tan and Canlas
(1988), 20. Tingsabad (1988), 21. Aggarwal, et al. (2006),
Jongwanich (2007).
Note: Number in parentheses in the first and second column
indicates reference study described below
APPENDIX II
Histograms Before and After Propensity Score Matching
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
[FIGURE 4 OMITTED]
REFERENCES
Adams, J. R. H. (1998) Remittances, Investment, and Rural Asset
Accumulation in Pakistan. Economic Development and Cultural Change 47,
155-173.
Adams, J. R. H. (2008) The Demographic, Economic and Financial
Determinants of International Remittances in Developing Countries.
Development Economics Department, Developing Prospects Group. The World
Bank, Washington, D. C. (Policy Research Working Paper 4583).
Aggarwal, R., A. D. Kunt, and M. S. M. Peria (2006) Do Worker
Remittances Promote Financial Development? World Bank, Washington, DC.
(World Bank Policy Research Paper 3957).
Amjad, R. (1986) Impact of Workers' Remittances from the
Middle East on Pakistan's Economy: Some Selected Issues. The
Pakistan Development Review 25:4, 757-782.
Amjad, R. (1988) To the Gulf and Back: Studies on the Economic
Impact of Asian Labour Migration (eds.). New Delhi: 1LO/ARTEP.
Arif, G. M. (1999) Remittances and Investments at the Household
Level in Pakistan. Pakistan Institute of Development Economics,
Islamabad. (Research Report No. 166).
Burney, N. A. (1987) Workers' Remittances from the Middle East
and their Effect on Pakistan's Economy. The Pakistan Development
Review 26:4, 745-763.
Burney, N. A. (1988) A Macro-Economic Analysis of the Impact of
Workers' Remittances from the Middle East on Pakistan's
Economy. In R. Amjad (ed.) To the Gulf and Back: Studies on the Economic
Impact of Asian Labour Migration. New Delhi: 1L0/ARTEP. 197-222.
Dehejia, R. and S. Wahba (2002) Propensity Score Matching Methods
for Non-Experimental Causal Studies. Review of Economics and Statistics
84:1, 151-161.
Deininger, K. and L. Yanyan (2008) Economic and Social Impacts of
Self-Help Group in India.Paper presented at the American Agricultural
Economic Association Annual Meeting, Orlando, FL, July 2007-29.
Fajnzylber, P. and J. H. Lopez (2007) Close to Home, The
Development Impact of Remittances in Latin America, Conference Edition,
The International Bank for Reconstruction and Development / The World
Bank, Washington, D.C.
Farchy, E. (2009) The Impact of EU Accession on Human Capital
Formation: Can Migration Fuel a Brain Gain? The World Bank, Development
Economics, Policy Review Unit. Washington, D.C. (Policy Research Working
Paper No. 4845).
Gilani, I., M. K. Faheem, and M. Iqbal (1981) Labour Migration from
Pakistan to the Middle East andIts Impact on the Domestic Economy.
Pakistan Institute of Development Economics, Islamabad. (Research Report
Series No. 126).
Ham, C., X. L. John, and B. R. Patricia (2005) Propensity Score
Matching, a Distance--Based Measure of Migration and the Wage Growth of
Young Men. Federal Reserve Bank of New York. (Staff Report No. 212).
Heckman, J. J., I. Hidehiko, and T. Petra (1997) Matching as an
Econometric Evaluation Estimator: Evidence from Evaluating a Job
Training Programme. Review of Economic Studies 64, 605-654.
Heckman, J., H. Ichimura, and P. Todd (1998) Matching as an
Econometric Evaluation Estimator. Review of Economic Studies 65,
261-294.
Hyun, O. S. (1988) The Impact of Overseas Migration on National
Development: The Case of the Republic of Korea. In R. Amjad (ed.) To the
Gulf and Back: Studies on the Economic Impact of Asian Labour Migration.
New Delhi: ILO/ARTEP. 143-166.
Iqbal, Z. and A. Sattar (2005) The Contribution of Workers'
Remittances to Economic Growth in Pakistan. Pakistan Institute of
Development Economics, Islamabad. (Research Report No. 187).
Jalan, J. and M. Ravallion (2001) Does Piped Water Reduce Diarrhea
for Children in Rural India? Development Research Group, Poverty. The
World Bank, Washington, D. C. (Policy Research Working Paper 2664).
Jongwanich, J. (2007) Worker's Remittances, Economic Growth
and Poverty in Developing Asia and the Pacific Countries, United Nations
Economic and Social Commission for Asia and the Pacific (UNESCAP)
(Working Paper, WP/07/01).
Kazi, S. (1988) Domestic Impact of Overseas Migration: Pakistan. In
R. Amjad (ed.) To the Gulf and Back: Studies on the Economic Impact of
Asian Labour Migration. New Delhi: ILO/ARTEP.
Knerr, B. (1992) Methods for Assessing the Impact of Temporary
Labour Emigration. The Pakistan Development Review 31:4, 1207-1239.
Mahmud, W. (1988) The Impact of Overseas Labour Migration on the
Bangladesh Economy: A Macro-Economic Perspective. In R. Amjad (ed.) To
the Gulf and Back: Studies on the Economic Impact of Asian Labour
Migration. New Delhi: ILO/ARTEP.
Malik, S. J. and N. Sarwar (1993) Some Tests for Difference in
Consumption Pattern: The Impact of Remittances using Household Income
and Expenditure Survey data of Pakistan 1987-88. The Pakistan
Development Review 32:4, 699-711.
Maqsood, A. and 1. Sirajeldin (1994) International Migration,
Earnings, and the Importance of Self-selection. The Pakistan Development
Review 33:3, 211-227.
Mansuri, G. (2007) Does Work Migration Spur Investment in Origin
Communities? In CaglarOzden and Maurice Schiff (eds.) Entrepreneurship,
Schooling, and Child Health in Rural Pakistan in International
Migration, Economic Development and Policy'. Washington, DC.: World
Bank and Palgrave Macmillan.
McKenzie, D., J. Gibson, and S. Stillman (2006) How Important is
Selection? Experimental vs. Non-experimental Measures of Income Gains
from Migration. Department of Economics. (Working Paper in Economics
3/06).
Nishat, M. and N. Bilgrami (1993) The Determinants of Workers
Remittances in Pakistan. The Pakistan Development Review 32:4,
1235-1245.
Nayyar, D. (1988) International Migration from India: Arnacro
Economic Analysis. In R. Amjad (ed.) To the Gulf and Back: Studies on
the Economic Impact of Asian Labour Migration.' New Delhi:
ILO/ARTEP.
Pakistan, Government of (2002) Pakistan Integrated Household
Survey. Islamabad: Federal Bureau of Statistics.
Pakistan, Government of (2008-09) Pakistan Economic Survey.
Islamabad: Planning Commission, Finance Division.
Quisumbing, R. A. and S. McNiven (2007) Moving Forward, Looking
Back: The Impact of Migration and Remittances on Assets, Consumption and
Credit Constraints in the Rural Philippines. Agriculture Development
Economics Division, The Food and Agriculture Organisation of the United
States. (ESA Working Paper No. 07-05).
Rodrigo, C. and R. A. Jayatissa (1988) Maximising Benefits from
Labour Migration: Sri Lanka. In R. Amjad (ed.) To the Gulf and Back:
Studies on the Economic Impact of Asian Labour Migration. New Delhi:
ILO/ARTEP.
Ravallion, M. (Forthcoming) Evaluating Three Stylised
Interventions, Forthcoming. Journal of Development Effectiveness.
Rosenbaum, P. R. and D. B. Rubin (1983) The Central Role of the
Propensity Score in Observational Studies for Causal Effects. Biometrika
70:1, 41-55.
Siddiqui, R. (2011) Pakistan: Migration, Remittances, and
Development. Chapter 3. In Saman Kelegama (ed.) Migration, Remittances,
and Development in South Asia. India: SAGE.
Siddiqui, R. and A. R. Kemal (2006) Remittances, Trade
Liberalisation and Poverty in Pakistan, The Role of Excluded Variable.
The Pakistan Development Review 45:3, 383-415.
Tan, E. A. and D. B. Canlas (1988) Migrant's Saving
Remittances and Labour Supply Behaviour: The Philippines Case. In R.
Amjad (ed.) To the Gulf and Back: Studies on the Economic Impact of
Asian Labour Migration. New Delhi: ILO/ARTEP. 223-254.
Tingsabadh, C. (1988) Maximising Development Benefits from Labour
Migration: Thailand. In R. Amjad (ed.) To the Gulf and Back: Studies on
the Economic Impact of Asian Labour Migration. New Delhi: ILO/ARTEP.
304-343.
White, H. (2005) Challenges in Evaluating Development
Effectiveness. Institute of Development Studies, Brighton. (IDS Working
Paper 242).
White, H. (2006) Impact Evaluation: The Experience of the
Independent Evaluation Group of The World Bank. Washington, DC.: World
Bank.
(1) Remittances currently represent about one-third of total
financial flows to developing countries, which are larger than official
development assistance flows. In many countries, they are also larger
than foreign direct investment. Therefore, the interest in the impact of
remittances is growing to better understand how remittances resulting
from migration contribute to poverty reduction [Fajnzylber and Lopez
(2007)].
(2) In absolute term, remittances have increased from $1087 million
to $13186.58 over 2001-12 [Pakistan (2008-09, 2012-13)].
(3) Adams (1998), Aggarwai, et al., (2006); Amjad (1986); Anijad
(1988); Arif (1999); Burney (1988); Gilani, et al., (1981); Hyun (1988);
Iqbal and Sattar (2005); Jongwanich (2007); Kazi (1988); Mahmud (1988);
Malik and Sarwar (1993); Maqsood and Sirajeldin (1994); Nayar (1988);
Quisumbing and McNiven (2007); Rodrigo and Jayatissa (1988); Siddiqui
and Kemal (2006); Tan and Canlas (1988); Tingsabad (1988).
(4) Amjad (1986, 1988); Burney (1987, 1988), Gilani, et al.,
(1981); Kazi (1988) for Pakistan, Oh-Seok (1988) for Korea, Mahmud
(1988) for Bangladesh, Nayar (1988) for India, Rodrigo and Jayatissa
(1988) for Sri Lanka, Tan and Canlas (1988) for Philippines, Tingsabad
(1988) for Thailand.
(5) An inflow of remittances increases household income and
expenditure, which may, in turn, generate new income and employment
opportunities--multiplier effect [Adams (1998)].
(6) For instance, Iqbal and Sattar (2005) estimate the relationship
between growth and remittances. Arif (1999) investigates investment
behaviour of remittances beneficiary households (RBH) and Malik and
Sarwar (1993) compare the consumption pattern of RBH and NRBI-1
[non-remittance beneficiary households], Adam (1998) has conducted a
Tobit analysis to explore remittances impact on rural asset
accumulation-land, livestock and non-farm assets. All these studies
ignore the problem of selection to migration. Though Maqsood and
Sirajeldin (1994) account for selection correction terms and focus on
one aspect, wage earnings and used explanatory variables which are
correlated with migration such as wealth.
(7) Gilani, el at. (1981); Amjad (1986); Irfan (1986); Various
studies in Amjad (1988); Burney (1987); Malik and Sarwar (1993); Arif
(1999); Iqbal and Sattar (2005); Siddiqui and Kemal (2006); Jongwanich
(2007); Some of them have analysed the impact of remittances on macro
and micro aggregates quantitatively using regression analysis. For
instance Maqsood and Sirajeldin (1994) consider migration as
endogenously determined, therefore made corrections in their earnings
function. However, all these studies overstate the impact because they
ignore the differences in observable characteristics i.e., measure the
impact of remittances on consumption without taking into account the
impact of income what they have earned in the domestic economy before
migration.
(8) Migrant households are those who receive remittance income from
abroad and non-migrants are those who do not receive income from abroad.
(9) Malik and Sarwar (1993) have compared consumption of RBH and
NRBH using this method. The results show that total consumption and
recurrent consumption of RBH are higher by 0.05 points, whereas
expenditure on durable goods is higher for NRBH.
(10) Education may also be an important variable to determine,
whether migrant send money through formal or informal channel. The
highly educated are expected to send remittances through formal
channel-susing financial institution. Whereas illiterate or low educated
labour send remittance through informal channels such as
'hundi'. Education and occupation are highly correlated.
(11) This indicates migration prevalence rate and is used as an
instrument for the opportunity to migrate [Mansuri (2007)]. Migratory
network increase migration opportunities by providing information to
potential migrants and existing migrant worker relax financial
constraints [Mansuri (2007)].
(12) Remittance income per households along with education level
will also determine how these remittances are sent and from where.
However, all these are assumptions, for real analysis there is a need to
collect data on these issues.
(13) Education may also be important determinant of labour migrated
to specific region. For instance, labour with high education level may
migrate to English speaking countries, whereas labour with lower
education level may migrate to Middle East countries.
(14) ROP includes Balochistan, Federal Administered Tribal Areas
and Azad Kashmir.
(15) Quisumbing and McNiven (2007) show that countries exporting
unskilled labour receive more remittances per capita than the
remittances per capita received by the countries exporting skilled
labour.
(16) According to the theory of migration, migration itself is
nothing but investment in human capital, which contributes to growth on
their return. But that analysis is beyond the scope of this paper.
(17) 57 percent of total remittances (through official and
unofficial channels) are allocated to recurrent consumption and 62
percent of remittances through official channels only.
(18) Multiplier effects of remittances also generate
growth-enhancing impact. Through back ward and forward
linkages--investment of one household could generate an increase in
income of the other, for example, investment in housing generates
employment for construction workers and income. Existing literature show
that this sector boost at the macro level.
(19) Aggarwal, et al. (2006) use level of deposits to measure
financial development that affect poverty and growth via credit
expansion.
(20) The reason can be that majority of RBH belong to rural area,
and they may be receiving remittances through informal channels. Even if
the migrant send through formal channels (Banks), household may not
report.
(21) Where in the lower strata, L-LS =Low-Low skill,, L-MS=Low
medium skill, L-HS= Low-high skill, In the upper strata H-LS=High-
low-skill, H-MS=High-medium-skill, and H-HS=High-high-skill.
(22) This difference is in total earned income.
Table 1
Results from Estimated Logistic Function
Coefficients T-Statistics
Community Characteristics
D_P 0.32 3.1
D_S 0.49 2.3
D_N 0.15 1.5
D_UR 0.10 1.2
[LY.sup.com.sub.REMH *] 0.89 21.8
Households Characteristics
D_EDU1 0.16 1.0
D_EDU2 0.18 1.7
D_EDU3 0.03 0.3
D_EDU4 0.40 1.8
D_EDU5 0.39 1.8
LHSIZ * 0.18 2.3
LAGE * 0.01 0.1
Constant -12.35 -17.0
*-Variables are in log form.
Table 2
Mean of the Covariates of Remittances Income
T-Test for Equality of Means
Before PSM After PSM
Mean Mean
Covariates (X-Vector) Difference T Difference T
Age 3.1 5.2 1.0 1.3
Education of Head of the 0.3 1.6 -0.1 -0.4
Household
Province 0.7 10.6 0.0 0.5
Region -0.1 -4.3 0.0 -0.6
Household Size 0.5 3.2 0.2 0.8
Remittance per Household 8439.8 28.0 785.0 1.8
by District
*** The range of estimated probability that a household receives
remittance income is between 0.0002 -0.35476. The distribution of
propensity scores (PS0 for the treated and control groups before and
after PSM are presented in Figures 1 to 4 in Appendix I. The common
support area is defined by dropping observation from the RBH group
whose P-values are larger than that of NRBH and the non treated
observation of which P-values are smaller than that of treated i.e.;
unmatched PS. in other words we select a common field for both
players, which is with PS in the range of 0.00035-0.35187. I drop the
cases that have probability less than 0.00035 and larger than 0.35187
from both groups. Prior to matching, the mean of estimated PS for
migrant and non-migrant households were, respectively, 0.14867 and
0.045161. In the trimmed sample the mean of PS for control is 0.06342,
the gap between the two reduces. But after the matching there is
negligible difference in the mean values of propensity scores of the
two groups-0.14867 for the control and 0.13853 for the RBH.
Table 3
Comparison of Differences in Means for Households--Treated (TRH) Vs
Control (NRBH)
Full Samples
Outcome and Output Indicators Mean t-statistics
Difference
(1) Income per adult equivalent -368.0 -0.1
(2) Expenditure per adult 11483.5 9.1
equivalent
(3) Expenditure on food per 3663.9 11.4
adult per year
(4) Non-food consumption 7819.6 7.8
(durables) per adult per year
(5) Calorie intake per adult per 1063.1 4.0
day
(6) Expenditure on education 5753.7 7.7
(7) Average class 1.2 8.6
(8) Average class of currently 4.4 9.4
going to school children
(9) Household equipment 31602.3 5.8
(10) Room occupy 1.1 17.0
(11) Electricity 0.2 16.7
(12) Gas 0.0 0.5
(13) Telephone 0.3 16.0
(14) Safe drinking water 0.0 -2.9
(15) Sanitation facilities 0.2 13.7
(16) Asset 559293.9 5.7
(17) Livestock 0.1 0.8
(18) Land holdings (acres) -0.7 -4.8
(19) Jewelry (Rs) 140415.7 1.2
(20) Saving in cash (Rupees) 19384.1 4.2
(21) Loan -1096.2 -0.1
(22) Profit on bank deposits 745.3 2.5
(23) Men employed -0.8 -19.7
(24) Women employed -0.3 -14.8
(25) Employed total -1.1 -22.9
(26) Enterprises -0.1 -4.3
(27) Poverty (Head Count Ratio) -0.2 -22.4
Number of Observation 802 vs 15122
Trimmed Sample-Common
Spoil Area
Outcome and Output Indicators Mean t-statistics
Difference
(1) Income per adult equivalent -6889.4 -0.8
(2) Expenditure per adult 10433.9 8.2
equivalent
(3) Expenditure on food per 3280.2 10.1
adult per year
(4) Non-food consumption 7153.7 7.1
(durables) per adult per year
(5) Calorie intake per adult per 868.4 3.2
day
(6) Expenditure on education 5152.9 6.8
(7) Average class 1.0 6.9
(8) Average class of currently 3.7 7.8
going to school children
(9) Household equipment 28515.4 5.2
(10) Room occupy 1.0 15.5
(11) Electricity 0.1 11.3
(12) Gas 0.0 -1.3
(13) Telephone 0.3 14.5
(14) Safe drinking water 0.0 -3.0
(15) Sanitation facilities 0.2 10.3
(16) Asset 519566.9 5.3
(17) Livestock -0.1 -1.0
(18) Land holdings (acres) -0.4 -2.5
(19) Jewelry (Rs) 139428.0 1.2
(20) Saving in cash (Rupees) 17449.0 3.7
(21) Loan 11115.0 1.6
(22) Profit on bank deposits 651.1 2.2
(23) Men employed -0.7 -16.6
(24) Women employed -0.2 -10.4
(25) Employed total -0.9 -18.4
(26) Enterprises -0.1 -5.9
(27) Poverty (Head Count Ratio) -0.2 -19.9
Number of Observation 802 vs 10756
After Propensity Score
Matching
Outcome and Output Indicators Mean t-statistics
Difference
(1) Income per adult equivalent 9948.1 5.9
(2) Expenditure per adult 8619.3 5.6
equivalent
(3) Expenditure on food per 2788.3 7.4
adult per year
(4) Non-food consumption 5831.0 4.7
(durables) per adult per year
(5) Calorie intake per adult per 695.5 1.9
day
(6) Expenditure on education 4094.9 4.7
(7) Average class 0.7 3.5
(8) Average class of currently 2.4 3.7
going to school children
(9) Household equipment 20046.5 3.1
(10) Room occupy 0.7 7.3
(11) Electricity 0.1 4.6
(12) Gas 0.0 0.0
(13) Telephone 0.2 10.1
(14) Safe drinking water 0,0 1.1
(15) Sanitation facilities 0.1 5.2
(16) Asset 336624.2 2.9
(17) Livestock 0.1 1.5
(18) Land holdings (acres) -0.3 -1.0
(19) Jewelry (Rs) 137027.2 1.2
(20) Saving in cash (Rupees) 15671.4 2.4
(21) Loan 9212.7 1.2
(22) Profit on bank deposits 645.1 1.9
(23) Men employed -0.5 -8.7
(24) Women employed -0 1 -3.8
(25) Employed total -0.6 -8.7
(26) Enterprises -0.1 -4.3
(27) Poverty (Head Count Ratio) -0.1 -7.7
Number of Observation 802 vs 685
Outcome and Output Indicators Bias
(1) Income per adult equivalent -2803.5
(2) Expenditure per adult -24.9
equivalent
(3) Expenditure on food per -23.9
adult per year
(4) Non-food consumption -25.4
(durables) per adult per year
(5) Calorie intake per adult per -34.6
day
(6) Expenditure on education -28.8
(7) Average class -13.4
(8) Average class of currently -45.9
going to school children
(9) Household equipment -36.6
(10) Room occupy -39.3
(11) Electricity -55.5
(12) Gas -114.1
(13) Telephone -19.2
(14) Safe drinking water -156.8
(15) Sanitation facilities -45.0
(16) Asset -39.8
(17) Livestock 172.2
(18) Land holdings (acres) -53.1
(19) Jewelry (Rs) -2.4
(20) Saving in cash (Rupees) -19.2
(21) Loan -940.4
(22) Profit on bank deposits -13.4
(23) Men employed -34.2
(24) Women employed -58.6
(25) Employed total -40.8
(26) Enterprises 60.4
(27) Poverty (Head Count Ratio) -33.1
Number of Observation
Table 4
Impact Evaluation of Remittances by Education Level
Education less Education below
than 1 year Primary (1-4 years)
Difference T- Difference T-
Outcome Indicators in Means Statistics in Means Statistics
Number of
Observation 403 44.0
Income per Adult
Equivalent 6374.36 7.99 4725.54 1.73
Expenditure per
Adult Equivalent 5396.05 7.94 637.83 0.23
Poverty
(Head Count Ratio) -0.2 -5.84 -0 1 -1.12
Education Primary Education Matric to
but below Matric below BA (10-13 years
(5-9 years)
Difference T- Difference T-
Outcome Indicators in Means Statistics in Means Statistics
Number of
Observation 156.0 146.1
Income per Adult
Equivalent 8011.45 2.48 21040.96 3.51
Expenditure per
Adult Equivalent 9212.72 4.95 18456.22 3.57
Poverty
(Head Count Ratio) -0.2 -4.39 -0.1 -2.93
Education BA to below Education MA and above
MA (14 to 15 years) including Professionals
(16 years and above)
Difference T- Difference T-
Outcome Indicators in Means Statistics in Means Statistics
Number of
Observation 27.0 28.0
Income per Adult
Equivalent 13569.31 1.01 22259.93 0.98
Expenditure per
Adult Equivalent 11935.50 0.77 13012.64 0.56
Poverty
(Head Count Ratio) 0.00 -0.05 0.00 0.00
