Determinants of household poverty: empirical evidence from Pakistan.
Majeed, M. Tariq ; Malik, M. Nauman
This study examines household characteristics and personal
characteristics of the household head as the determinants of poverty in
Pakistan. The study decomposes education of the household into different
levels: primary, middle, matriculation, intermediate, bachelors and
higher studies and finds evidence that poverty is greatest among the
less literate households and declines as education level increases. The
probability of being poor reduces in urban area implying that incidence
and severity of poverty is more pronounced in rural areas. The role of
remittances appeared significant in reducing probability of being poor
and this is more striking in rural areas. The variables that are
negatively related with the probability of being poor are: experience,
age square and agriculture employment status. While the variables that
are positively related with the probability of being poor are: household
size, age of the household head, male-headed households and the
provincial residence.
JEL Classification: 13, J24
Keywords: Poverty, Education, Gender, Determinants, Household,
Pakistan
1. INTRODUCTION
The Millennium Development Goals (MDGs) aim at halving the
percentage of world population in 1990 with income less than US $ 1 a
day and halving the share of people who suffer from hunger by 2015.
Being a developing nation, poverty reduction should be our foremost
obligation. An appreciable decline has occurred recently, headcount
decreased from 34.46 percent in 2000-01 to 23.94 in 2004-05 [Pakistan
(2006-07)]. However, seeing only the statistics and the trends in
poverty we can just observe that what happened to poverty in different
periods and also the decomposition of poverty in different years gives
us a more appropriate picture of the incidence of poverty. This
knowledge is useful because it informs us whether poverty is increasing
or decreasing overtime. However, this information does not provide us
the details of the causes of poverty. For instance, is poverty high due
to low education attainment or large family size or due to any other
reason? Here is a need of research about the determinants of poverty
that are positively or negatively linked with the poverty status. This
is the area where research can be most useful because firstly we have to
understand the main determinants of poverty before designing the most
efficient policy to reduce poverty in the country.
A logistic regression technique has been used to evaluate the
determinants of poverty in Pakistan. An important determinant of
household poverty is education of the head of the household. In the
Millennium Development Goals (MDGs), originated from the United Nations
(UN) summit 1999, and the Poverty Reduction Strategy Paper (PRSP),
promoted by the World Bank and the IMF, education is considered as a
weapon against poverty. Therefore the idea that education is a
determinant of poverty has occupied much attention in the recent years.
Since 1960s when Shultz (1961) and Becker (1962) emphasised upon the
attainment of education and skills for human development,
education's role in the economic growth and development became
prominent and its importance in poverty reduction increased manifold. We
have to seek out such vital channels (both qualitatively and
quantitatively) between education and poverty reduction that will help
us in policy formulations for poverty reduction and educational
expansion.
Along with educational deprivation we will also analyse some other
personal and household characteristics. In this regard experience, age,
gender and employment status of the head of the household are important.
In most of developing countries gender discrimination is widely
prevalent. Females have less educational and earning opportunities as
compared to males. That's why female/male headed households can be
an important determinant of poverty. Region (urban/rural) as a
determinant of poverty is important in developing countries where
usually agriculture sector is dominant. Moreover, the rise of
industrialisation coupled with migration problems persuades us to
consider the region in poverty determinants. We will also extend the
location variable to cover the different provinces of a country.
Moreover, household size and remittance receiving status of household
will also be explored.
The main objective of this study is to determine the effect of
different educational levels upon the probability of being poor of
households (considering the expenditure side) in Pakistan. Similarly,
some other personal characteristics such as gender, age, experience and
employment status of the head of the household and some household
characteristics such as the household size, remittance receiving status,
regional and provincial location will also be analysed.
The study is structured as follows: Section 2 provides review of
the literature on determinants of poverty. Section 3 is related with the
data and methodology details while Section 4 includes the description of
regression technique and construction of variables. Section 5 provides
descriptive analysis of poverty assessment. Section 6 includes the
logistic estimations and interpretation of the results and finally,
Section 7 concludes with some recommendations.
2. REVIEW OF LITERATURE
After the contribution of Mincer (1958) in finding the role of
education in wage earnings, Schultz (1961) and Becker (1962) both viewed
investment in education attainment and in skill enhancement as the
necessary component of human capital accumulation. As human capital
formulation is necessary for poverty reduction that's why education
becomes the vital and prominent factor in reducing poverty both at
income level and also at other social and capability levels.
Coulombe and Mckay (1996) used multivariate analysis to analyse the
determinants of poverty in Mauritania based on household survey data for
1990. They estimated a multinomial logit model for the probability of
being poor depending on household specific economic and demographic
explanatory variables. The authors found that low education, living in a
rural area and a high burden of dependence significantly increase the
probability of being poor of the household.
Gundlach, et al. (2001) did a study on 102 countries using the
quality adjusted broader measure of human capital, which depends upon
the social returns of educational levels and an index of quality. The
findings show that the income of the poor (lowest quintile) increases
with the rising quality-adjusted human capital. They estimate that a 10
percent increase in the stock of quality adjusted human capital per
worker increases the average income of the poor by 3.2 percent.
Tilak (2002) has conducted a comprehensive analysis of the
approaches of development and well-being with respect to the
education's reflections upon poverty. He points out that the
inverse relationship between education and human poverty is well
recognised in many prominent approaches of development such as the human
capital approach, the basic needs approach, the human development
approach and the capability approach. The author argues that at micro
level incidence of poverty is greatest among the illiterate households
and tends to decline at higher levels of education in developing
countries. Moreover, (at macro level) the decline in poverty is possible
through higher level of education of the population. He also notes that
a mutually reinforcing relation persists between education poverty (lack
of education) and income poverty because income deprivation resists
persons from attaining education and absence of education causes
low-income level. Tilak vehemently mentions the direct linear
relationship between education and earnings. This relation is well
recognised universally i.e. with the rise of education earnings also
rise considerably.
Okojie (2002) further goes in to the details of educational levels
that affect the household's income poverty and the human poverty
using household data of 1980, 1985, 1992 and 1996 for Nigeria. In the
poverty model, the logistic regression was used and it was found that
all levels of education (primary, secondary and tertiary) are
significant in reducing the probability of being poor of the households.
The results show that maleheaded households are less likely to be poor
than female-headed households. In the welfare model, the mean per capita
expenditure was used qs dependent variable and educational variables
found to be significant in increasing the per capita expenditure of the
household.
Bundervoet (2006) conducts a study upon the household data of
1998-99 of Burundi. The results show that the incidence of poverty
(headcount measure), poverty gap and poverty severity are worse for the
female headed families as compared to male headed families, however, the
worse off element decreases as the head's educational achievement
increases. The binary logistic regression results show the poverty
status of household using explanatory variables of household and
community characteristics. At rural level higher educational level of
the head of the household significantly reduces the likelihood of being
poor. A literate mother in the household reduces the probability of
being poor. The probability of poor rises up to the age of 42 of the
head and then declines. The possible reason could be the accumulation of
assets for old age.
Zuluaga (2007) conducts a study on the 31,745 households of
Colombia to find the monetary and the non-monetary effects of education
upon income poverty and human poverty, respectively. The results show
that an additional year of schooling of the head of the household
increases total income of the household by the amount of 14.1 percent.
Female-headed household is more likely to have less income as compared
to male-headed but a rise in income quintile (towards non-poor)
diminishes such disadvantage. Residents of rural areas are significantly
poorer than those in urban areas. The interesting finding is this that
the effect of education is not same in affecting all income quintiles.
The return of education is bigger for the lowest quintile and decreases
as the quintile increases. This shows that people from the lowest
quintiles benefit more from the skills through formal education. In
other words poor persons benefit more from the education attainment. For
the non-monetary effects of education upon human poverty the author
considers housing and health. The results show that education improves
health through modifying the behaviour and decisions of persons with
respect to health. Housing conditions also improve with the increasing
educational level because education improves its decisions and behaviour
regarding housing and it can avail credit facility in a better way.
Abuka, et al. (2007) estimate the determinants of poverty in the
case of Uganda using logistic regression technique and the data from
Uganda National Household Survey (UNHS). The results showed that an
increase in the schooling of household heads not only has a positive
impact on the productivity and earnings but also enhances the
productivity of other members of the household. The household size and
being in rural areas significantly increase the likelihood of being
poor.
A further analysis of educational levels by Tilak (2007) has shown
noteworthy results. He argued that it would be wrong to say that for
growth, development and poverty reduction we should wait for the
universalising of primary education rather we should work on the
post-primary education because it has the same role as primary
education. Primary education is the threshold of human capital but
secondary and higher education, and investment in science and technology
gives rise to acceleration and sustenance in economic growth and
development. The coefficient of correlation in India suggests that
illiteracy, literacy and primary education are positively related with
the poverty ratios. While, on the other hand middle, secondary and
higher education levels are negatively related with poverty.
The above mentioned studies consider education as a vital weapon
against poverty but Dollar and Kraay (2002) argue that education
doesn't have any substantial or measurable effect on the income of
the poor except its effect upon the overall average growth. They conduct
a macro level study based upon the data of 137 countries for the years
1950-99. They reported that income of poor raises one for one with
average income (growth) but the primary education attainment has a very
limited impact upon the income of the poor. They conclude that economic
growth is a prominent factor in eliminating poverty and primary
education completion is not so much important.
The similar conclusion was proposed by Tilak (2007) in studying the
correlation coefficients between the poverty ratios of 1999-00 and
percentage of population having different educational levels in 1995-96
in India. The results show that illiteracy, literacy and primary
education are positively related with the poverty. Hence it casts doubt
upon the role of primary education in poverty reduction.
Majeed (2010) shows the poverty reducing effect of human capital in
the case of Pakistan using the data over the period 1970-2004. In a
recent study, Majeed (2012) finds mixed evidence for the relationship of
poverty and human capital using a sample of sixty five developing
countries over the period 1970-2008.
Through analysing different studies we can see that it is necessary
that we must know the determinants of poverty for an effective poverty
reduction strategy. Rather than focusing on macro level and cross
country analyses we have to go for the micro level research for the
proper evaluation of the poverty determinants. Dealing with micro level
data we are engaging in the ground realities and micro circumstances of
any particular country. Micro level data approach is very much relevant
for the poor developing countries whose main problems are widely
prevalent at grass root levels while macro data based studies do not
represent the effects of those problems in their data with aggregates or
averages.
3. METHODOLOGY
This study evaluates the personal and household characteristics as
determinants of poverty in Pakistan. We show that how the occurrence of
any particular event will affect the likelihood of the household being
poor. For instance, in what proportion the acquisition of primary
education will increase or decrease the likelihood of being poor with
respect to 'no education'.
Education is the most important factor regarding poverty reduction.
The attainment of education enhances the earning potential of
individuals and consequently, the increased earnings help reduce the
poverty. There are also non-pecuniary effects of education that are
effective in other dimensions of poverty such as deprivation in decision
making abilities, and awareness about the surrounding. Hence it is
expected that education is negatively linked with the poverty status and
higher levels of education are more effective in poverty reduction.
Experience can be taken as the improvement in expertise and skill
enhancement, which have positive implications for poverty elimination.
The 'feminisation of poverty' means women are much more
deprived and facing severe hardships in pulling themselves out of
poverty as compared to men therefore it is expected that being
female-headed household will increase the likelihood of the household
being poor. The age of the head of the household is going to be seen in
non-linear relation. Generally, in the working age of the head of the
household when one can accumulate human capital there are more chances
to be non-poor as compared to the old age. However, in the opposite case
it is said that until the old age (or after retirement) one can
accumulate enough resources or assets to be non-poor in old age as
compared to the working middle age.
A large portion of population in Pakistan is directly or indirectly
linked with our traditional agriculture sector hence its important to
find out that whether the agriculture employment status as compared to
non-agriculture employment status of the household head is effective in
reducing household poverty or not. Population is a resource but its huge
size and high growth rate in developing countries appeared to be a
problem due to low level of human capital. Hence usually family size is
positively related with the poverty status of the household.
Remittance, whether domestic or foreign, is a source of income for
the household and reduces household poverty. It is a widely prevalent
idea that in Pakistan the incidence and severity of poverty is high in
rural areas as compared to urban areas hence to verify such statement we
have to see whether the rural location of the household is associated
with being poor or not. In the same way we can analyse the provincial
location of the household as well.
Having been provision of theoretical consistent arguments, we have
developed this model with choice variables comprising of the personal
characteristics of the head of the household and household
characteristics. In this regard, education, experience, gender, age and
employment status of the household head are considered as personal
characteristics of the household while household size, provincial
location, regional location and remittance receiving status are
considered as the household characteristics.
POVERTY = f(EDU, GEN, AGE, EXP, EMP, HS, REM, REG, PRO)
Dichotomous dependent variable: Poor = 1, Non-poor = 0
Explanatory Variables: Personal Characteristics of the Household Head
* Education (EDU)
Primary Education Primary = 1, Otherwise = 0
Middle Education Secondary = 1, Otherwise = 0
Matric Education Matric = 1, Otherwise = 0
Intermediate Education Intermediate = 7, Otherwise = 0
Bachelors Education Bachelors = 1, Otherwise = 0
Professional Education Masters or above education=l, Otherwise = 0
No Education Reference Category
* Gender (GEN) Male = 1, Female = 0
* Employment Status (EMP) Agriculture = 7, Not agriculture = 0
* Experience (EXP) Age-School starting age-Years of schooling
* AGE Age, [Age.sup.2] (Square of Age)
Household Characteristics
* Region (REG) Urban = 1, Rural = 0
* Remittances (REM) Remittance = 7, Not = 0
* Household Size (HS) Number of individuals in family
* Province (PRO)
Punjab Punjab = 1, Otherwise = 0
Sindh Sindh = 7, Otherwise = 0
KPK KPK = 7, Otherwise = 0
Balochistan Reference Category
4. DATA, CONSTRUCTION OF VARIABLES AND ECONOMETRIC TECHNIQUE
The data for this study is taken from Household Income and
Expenditure Survey (HIES) 2001-02 which is conducted by the Federal
Bureau of Statistics (FBS) of Pakistan. It's the available gigantic
and meaningful source of information of its kind that has the household
level information in Pakistan. The selected data used for this study
covers the four provinces of Pakistan (Punjab, Sindh, Balochistan and
KPK).
The very first thing is to clarify the criteria through which we
classify the households into poor and non-poor. In other words we can
say that how we assign value of one (poor) or zero (non-poor) to the
dependent dichotomous variable. For this task, there are different
approaches such as the basic needs approach or the calorie-based
approach; but here we use the method of quartile. We make four quartile
of households depending upon the monthly per adult household
expenditure. The lowest quartile (25 percent) will have the households
with the lowest monthly per adult household expenditures. The households
in the lowest quartile are considered poor and consequently dependent
variable takes value one for them whereas each household in other three
quartiles take the value zero. The household expenditure variable is the
monthly per adult expenditure of the household considering all the food
and non-food items. To calculate the adult equivalents we make use of
the official calories chart (2003) with respect to age, provided by the
Government of Pakistan.
Considering the explanatory variables of our model the personal
characteristic variables will be used for the head of the household. The
educational variables are dummy variables and one of them will get the
value one in response to the household head's highest educational
attainment. It means the educational level of the household's head
will either fall in primary, secondary, matriculation, intermediate,
bachelors or professional (masters and above) category. Here 'no
education' is used as reference category. Other variables include
age, experience and employment status of the household's head. Here
the employment status is characterised into two broad categories whether
the status is related to the agriculture sector (owner cultivator, share
cropper, contract cultivator, livestock) or non-agriculture sector
(employer, self-employed, paid employee-reference category). The
experience variable is attained through subtracting the years of
schooling and school starting age from the age of a person. It is not
the actual but the potential experience. The personal characteristics
include male/female headed households where female headed will be the
reference category, household size*, whether the household is remittance
receiver or not where having no remittances is the reference category,
regional variable with rural as the reference category and provincial
location of the household with Balochistan as the reference category.
The dependency ratio also matters as a correlate of poverty, since
our main focus has been to investigate the role of human capital for
which we have introduced different categories. We followed a
parsimonious approach in selecting other control variables and have
chosen control variables which are closely related to poverty incidence.
This study takes the Logistic Regression Technique to identify some
determinants of poverty in Pakistan at household level. The model is
estimated using the information of the four provinces of Pakistan. The
binary logistic regression is used to identify the effect of explanatory
variables upon the probability of being poor of the household. The
dependent variable is dichotomous in which the value 1 for the poor
household and 0 for the non-poor household. The results will not be
interpreted through the coefficients but we will use the odd ratios in
logistic regression to see that the occurrence of any particular event
will increase or decrease the probability being poor of household and
with what proportion as compared to the reference category.
5. POVERTY ASSESSMENT: A DESCRIPTIVE ANALYSIS
This study examines the personal characteristics and household
characteristics as the determinants of poverty in Pakistan. Therefore it
would be convenient to understand the results if we see the graphical
representation of the poverty assessment in selected dimensions. The
descriptive analysis is based upon the demonstration of average number
of poor households in the particular dimensions. Hence the dimensions,
which are going to be demonstrated, are the education, gender and
regional location.
5.1. Poverty and Education
Investment in education is considered as the main source of human
capital accumulation, which is the least developed sector of many
developing countries including Pakistan. The acquisition of education
helps an individual to overcome the multidimensional poverty prevalence
and the education of the household head is also beneficial for other
family members. Through education availability we can break the mutually
reinforcing relationship between poverty and lack of education
(education poverty). In Figure 1, we can see that as the educational
level of the household head increases, on average the number of poor
households declines. There is a consistent reduction in poverty from no
education to the bachelor's level.
[FIGURE 1 OMITTED]
5.2. Poverty and Regional Location
Rural areas are much deprived as compared to the urban areas in
Pakistan. One of the important reasons is the low productivity and
consequently, the low incomes in the rural areas. Moreover, rural areas
are much more vulnerable to natural calamities especially the floods and
droughts. There is a huge gulf between the rural and urban areas in
terms of facilities and opportunities that shows the biased government
policies against rural areas. That's why we observe the regional
migration phenomenon especially for quality education and employment
opportunities.
Incidence, depth and severity of poverty are high in rural areas as
compared to urban areas in Pakistan [Jamal (2005)]. Our graphical
demonstration of the data in Figure 2 shows that on average poor
households are much more in rural areas as compared to urban areas.
[FIGURE 2 OMITTED]
5.3. Poverty and Gender
It is generally perceived that our society is characterised with
gender bias or gender discrimination. Women have unequal opportunities
in education attainment and earnings as compared to men. Generally,
female participation in society is low and it is observed that
female-headed households face difficult circumstances to escape the
poverty. The descriptive analysis in Figure 3 shows the surprising
result that on average the poor female-headed households are small as
compared to male-headed households. One reason behind this result could
be the under-representation of female-headed households because there
are cultural reasons to believe that many of the households that showed
themselves male-headed are actually the female-headed households.
[FIGURE 3 OMITTED]
The graphical activity demonstrates the results about poverty
incidence on average or aggregate basis. Cognizant of this descriptive
analysis, now we are able to relate this information with our regression
results to have a more vivid picture about the poverty determinants.
6. RESULTS
The logistic regression technique has been applied to evaluate the
personal characteristics of the household's head and household
characteristics as the determinants of household poverty in Pakistan.
The personal characteristics include education, gender, age, employment
status and experience. The household characteristics include regional
location, provincial location, household size and remittances.
Table 2 reports the result for four provinces. The separate
provincial and regional results are reported in the appendix. The Wald
test is used to test the significance of coefficients and interpret the
results using odd ratios. All the educational variables of the household
head are highly significant in reducing the probability being poor of
the household. The primary, middle, matriculation, intermediate,
bachelors and higher studies (professional category) education of the
household head reduce the likelihood of the household being poor by 22
percent, 54 percent, 64 percent, 87 percent, 90 percent and 89 percent
respectively as compared to the reference category of 'no
education'.
It is noteworthy that the chances of escaping poverty of the
household increase consistently as we increase the educational level of
the household head. However, little improvement is observed beyond the
attainment of intermediate education. All the educational variables in
the separate provincial regressions provided in the appendix show that
education is significantly and negatively related with the poverty
status of the household except the primary education in Sindh, KPK and
Balochistan. The same situation is with the middle and matric in
Balochistan. However, all coefficients have negative signs as expected.
Considering the separate regional regressions we observe that primary
education of the head of the household is significant in reducing
poverty in rural areas. In the rural areas primary education reduces the
chances of poverty by 29 percent in comparison to the base category of
no education. Moreover, all educational levels have shown that all
levels are reducing the chances of poverty in greater proportion in
urban areas as compared to the rural areas except the primary education.
If the employment status of the household head falls in the
category of agriculture status (owner cultivator, share cropper,
contract cultivator or livestock owner) then this reduces the
probability of household being poor by 27 percent as compared to the
base category of non-agriculture status. For age of the household head,
we observed the positive sign for the variable age and the negative
correlation is found between the poverty status of household and square
of age. The age variable shows that as the age of the head increase by
one year the chances of the household being poor will significantly
increase by 3.9 percent. However the age-square variable shown negative
sign which means in the older ages of the household head likelihood of
the household being poor declines by 0.1 percent. Although the
experience has a little effect but as the experience of the head of the
household increases by one year then it reduces the chances of the
household being poor by 0.9 percent.
The residence in urban areas was negatively associated with the
poverty status. If the household is situated in the urban region then
this reduces the likelihood of household being poor by 60 percent as
compared to the reference category of rural areas. This result is
significantly consistent in all separate provincial regressions given in
the appendix. The male-headed household significantly increases the
probability of that household to fall in poverty by 82 percent as
compared to the female-headed household. The overall result's
negative sign of the male-headed households also holds for the separate
provincial and regional regressions. A household is more likely to be
poor if it has a large number of members. If the family size increases
by one person then it increases the probability of the household being
poor by up to 22 percent. The same increase we observe for household
size in the separate provincial and regional results. If the household
is remittance receiver whether the remittances come from abroad or
within the country then it decreases the probability of the household
being poor by 43 percent than non-receiving households. In appendix the
remittance effect in the separate urban and rural regressions reduces
the chances of poor by 36 percent and 43 percent respectively.
Considering the provincial location variables, being in Punjab, Sindh
and KPK increases the chances of the household being poor by 94 percent,
31 percent and 91 percent respectively as compared with the base
category of Balochistan. In appendix one additional determinant is
evaluated considering the data of four provinces of Pakistan. The
additional variable named as earners, which counts the number of earners
per household that have any level of education. With the increase of one
educated earner significantly reduces the probability of household being
poor by 11 percent. However, almost all other results remain intact
except the experience variable, which becomes insignificant.
7. CONCLUSION
The purpose of this study was to estimate the determinants of
household poverty in Pakistan. The data used for this task is taken from
the Household Integrated Economic Survey (HIES 2001-02) conducted by the
Federal Bureau of Statistics. The determinants of poverty are explored
using the logistic regression technique.
The main findings of our analysis can be concluded as follows:
First, poverty is greatest among the less literate households and
declines as education level increases-primary, middle, matriculation,
intermediate, bachelors and higher studies. Therefore, educational
attainment is a critical determinant of the incidence of poverty and
should be considered closely in implementing poverty alleviation
programs. Second, the role of remittances appeared significant in
reducing probability of being poor and this is more striking in rural
areas. Third, the probability of being poor reduces in urban area
implying that incidence and severity of poverty is more pronounced in
rural areas.
Finally, the variables that are negatively related with the
probability of being poor are: experience, age square and agriculture
employment status. While the variables that are positively related with
the probability of being poor are: household size, age of the household
head, male-headed households and the provincial residence.
This analysis has certain limitations: First, it is a
cross-sectional analysis using household survey data and it does not
take into account time dynamics. Second, this study mainly focuses on
the different levels of education and some selected control variables.
Some control variables such as dependency ratio and training are
missing. Future research can make a comparison of poverty determinants
between different household surveys. In addition, research can be
extended to incorporate more control variables. Similarly, a time series
analysis can be conducted.
This analysis purposes following policy implications:
* There is a need to implement an appropriate policy measure in
order to achieve the negative impact of education on poverty through
increasing share of education expenditures at all levels.
* It is recommended that policy makers need to focus more on
facilitating the remittances flows in rural areas through increasing
financial access and reducing the costs associated with transfers of
money.
APPENDIX
Table 3
Separate Gender Level
Explanatory Variables
Gender Age [Age.sup.2] Primary Middle
Male
Coefficient .035 .000 -.299 -.727
P-values .000 .000 .087 .000
Odd Rations 1.036 1.000 .795 .484
Female
Coefficient .55 -.001 .146 -.133
P-values .155 .075 .809 .841
Odd Rations 1.057 .999 1.157 .875
Explanatory Variables
Gender Matric Inter BA Prof. Urban
Male
Coefficient -.980 -1.966 -2.387 -2.162 -.786
P-values .000 .000 .000 .000 .000
Odd Rations .375 .140 .092 .115 .456
Female
Coefficient -7.288 -7.203 -6.918 -7.194 -.850
P-values .553 .709 .725 .780 .000
Odd Rations .001 .001 .001 .001 .427
Explanatory Variables
Gender Exp. Mem. Rem. Constant
Male
Coefficient -.007 .181 -.329 -2.290
P-values .000 .000 .000 .000
Odd Rations .993 1.199 .719 .101
Female
Coefficient -.028 .199 -.604 -2.228
P-values .156 .000 .002 .013
Odd Rations .973 1.220 .546 .108
Table 4
Separate Provincial Level
Explanatory Variables
Province Age [Age.sup.2] Primary Middle Matric
Punjab
Coefficient .040 -.001 -.328 -1.092 -1.556
P-values .007 .000 .095 .000 .000
Odd Rations 1.041 .999 .720 .336 .211
Sindh
Coefficient .007 .000 -.302 -.627 -.822
P-values 692 .352 .225 .023 .001
Odd Rations 1.007 1.000 .739 .534 .439
KPK
Coefficient .73 -.001 -.246 -.432 -.725
P-values .002 .000 .478 .187 .016
Odd Rations 1.075 .999 .782 .649 .484
Balochistan
Coefficient .58 -.001 .166 .022 -.151
P-values .030 .010 .683 .957 .642
Odd Rations 1.060 .999 1.180 1.022 .859
Explanatory Variables
Province Inter BA Prof. Urban Male
Punjab
Coefficient -2.728 -2.526 -2.812 -.849 .865
P-values .000 .000 .000 .000 .000
Odd Rations .065 .080 .060 .428 2.376
Sindh
Coefficient -1.594 -2.208 -2.459 -1.40 .238
P-values 000 .000 .000 .000 .577
Odd Rations .203 .110 .085 .246 1.268
KPK
Coefficient -1.915 -1.963 -1.831 -.650 .350
P-values .000 .000 .001 .000 .053
Odd Rations .147 140 .160 .522 1.419
Balochistan
Coefficient -1.137 -1.870 -1.384 -.778 .658
P-values .025 .001 .015 .000 189
Odd Rations .321 .154 .251 .459 1.932
Explanatory Variables
Province Exp. Mem. Rem. Agri. Constant
Punjab
Coefficient -.005 .224 -.384 -.596 -2.931
P-values .361 .000 .000 .000 .000
Odd Rations 995 1.251 .681 .551 .053
Sindh
Coefficient -.007 .246 .150 -.051 -2.396
P-values .300 .000 .722 .610 .000
Odd Rations .993 1.279 1.161 .950 .091
KPK
Coefficient -.013 .131 -.681 -.199 -3.066
P-values .127 .000 .000 .095 .000
Odd Rations .987 1.140 .506 .820 .047
Balochistan
Coefficient -.029 .171 -.370 -.255 -3.623
P-values .010 .000 .286 .059 .000
Odd Rations .972 1.186 .690 .775 .027
Table 5
Separate Region Level
Explanatory Variables
Region Age [Age.sup.2] Primary Middle
Urban
Coefficient .047 -.001 -.220 -.812
P-values .007 .000 .310 .000
Odd Rations 1.048 .999 .802 .444
Rural
Coefficient .027 .000 -.340 -.647
P-values .019 .001 .037 .000
Odd Rations 1.027 1.000 .712 .523
Explanatory Variables
Region Matric Inter BA Prof. Urban
Urban
Coefficient -1.107 -2.261 -2.555 -2.739 .528
P-values .000 .000 .000 .000 .001
Odd Rations .331 .104 .078 .065 1.696
Rural
Coefficient -1.06 -2.211 -2.384 -1.763 .454
P-values .000 .000 .000 .000 .001
Odd Rations .345 110 .092 .172 1.575
Explanatory Variables
Region Exp. Mem. Rem. Agri. Constant
Urban
Coefficient -.009 .186 -.443 -2.679
P-values .113 .000 .000 .000
Odd Rations .991 1.205 .642 .069
Rural
Coefficient -.006 .186 -.552 -.288 -2.831
P-values .159 .000 .000 .000 .000
Odd Rations .994 1.204 .576 .750 .059
REFERENCES
Abuka, C. A., M. A. Ego, J. Opolot, and P. Okello (2007)
Determinants of Poverty Vulnerability in Uganda. Institute for
International Integration Studies. (Discussion Paper No. 203).
Becker, G. S. (1962) Investment in Human Capital: A Theoretical
Analysis. The Journal of Political Economy 9-49.
Bundervoet, T. (2006) Estimating Poverty in Burundi. Households in
Conflict Network (HiCN). (Working Paper No. 20).
Coulombe, H. and A. McKay (1996) Modeling Determinants of Poverty
in Mauritania. World Development 24:6, 1015-31.
Dollar, D. and A. Kraay (2002) Growth is Good for the Poor. Journal
of Economic Growth 7, 195-225.
Gundlach E, J. N. D. Pablo, and N. Weisert (2001) Education is Good
for the Poor. World Institute for Development Economics Research
(WIDER). (Discussion Paper No. 2001/137).
Haughton, J. H. and S. R. Khandker (2009) Handbook on Poverty and
Inequality. World Bank Publications.
Jamal, H. (2005) In Search of Poverty Predictors: The Case of Urban
and Rural Pakistan. The Pakistan Development Review 44:1, 37-55.
Majeed, M. Tariq (2010) Poverty and Employment: Empirical evidence
from Pakistan. The Forman Journal of Economic Studies 6:1.
Majeed, M. Tariq (2012) Poverty Consequences of Globalisation in
OIC Countries: A Comparative Analysis. The Pakistan Development Review
51:4, 479-492.
Mincer, J. (1958) Investment in Human Capital and Personal Income
Distribution. The Journal of Political Economy 281-302.
Okojie, C. E. (2002) Gender and Education as Determinants of
Household Poverty in Nigeria. World Institute for Development Economics
Research (WIDER). (Discussion Paper No. 2002/37).
Pakistan, Government of (2006-07) Pakistan Economic Survey.
Ministry of Finance Islamabad.
Schultz, T. W. (1961) Investment in Human Capital. The American
Economic Review 1-17.
Tilak, J. B. (2002) Education and Poverty. Journal of Human
Development 3;2, 191--207.
Tilak, J. B. (2007) Post Elementary Education, Poverty and
Development in India", International Journal of Educational
Development 27:4, 435-445.
Zuluaga, B. (2007) Different Channels of Impact of Education on
Poverty: An Analysis for Colombia. Discussion Paper, Available at SSRN
958684.
Comments
It is a nice effort to emphasise the importance of education in
poverty but has serious issues which are to be addressed by the author.
The issues are listed below:
(i) Surprisingly the study presented in December 2015 is based on
data of 200102. After 2001-02, five data sets of HIES are released by
FBS and a lot research is carried out with new available data sets, then
what is the significance of analysis based on 14 years old data.
(ii) Employment status is taken as Agriculture and non-agriculture.
This is industry rather than employment status. Moreover the study is
based on rural as well as urban areas of Pakistan. As per HIES used in
present study only 6.46 percent earners from urban areas are employed in
agriculture. So this classification makes no sense for employed persons
of urban areas.
(iii) One of the explanatory variables is Household size. It is
measured by household members. In poverty analysis absolute number is
not important; it is rather age and gender composition which is
important. Moreover in poverty analysis merely household size does not
matter, it is dependency ratio that matters.
(iv) The household belonging to lowest quintile are considered as
poor, (it 5th quintile (not 4th) means lowest 20 percent households). It
means around 20 percent households are considered as poor, which is
vague, as all studies reported that for 2001-02 around 35 percent
population was below poverty line.
(v) On page 13 "If employment status falls in agriculture then
it reduces the probability of being poor by 27 percent as compare to
base category of non-agriculture".
(vi) This result does not match with the published data of HIES for
2001-02. As per data average monthly income in Agriculture is 2062 and
for non-agriculture it is 3303. Moreover it is less than all other
industries. Almost same is true for all quintiles
(vii) As per data 62.25 percent earners in rural areas belongs to
agriculture that why poverty is more intensive in rural areas, On the
other hand only 6.46 percent earners from urban areas are engaged in
agriculture. What does these results indicates for urban people.
(viii) Taking the variables of age and experience of head at the
same time makes no sense, as experience is down scaling of age, i.e.,
age minus four or five. AND surprisingly both variables have different
signs.
(ix) Households are categorised in two groups as receivers of
remittances or non-receivers. This is too broader classification, as
volume of remittances do matter. Secondly the micro analysis of data
shows that overwhelming majority of household belonging to top quintile
are not receivers of remittances.
(x) The results on one side show that education of head has
negative impact on poverty, on the other side households with
agriculture industry are better off. These findings are again
contradicting as mostly educated people are engaged in non-agri
industry.
Muhammad Idrees
Quaid-i-Azam University, Islamabad.
M. Tariq Majeed <m.tariq.majeed@gmail.com> is Assistant
Professor, Quaid-i-Azam University, Islamabad. M. Nauman Malik is
Lecturer, University of Sargodha, Sargodha.
* Haughton and Khandker (2009) argue that household size is also an
important correlate of poverty.
Table 1
Construction of Explanatory Variables
Categories Variables Explanation
Education Primary 1 = primary, 0 = otherwise
(Dummy) Middle Similarly, we make other
Matric (Matriculation) education variables.
Inter (Intermediate)
Ba (Bachelors)
Prof (Professional)
No education (reference
category)
Age Age Age
[Age.sup.2] Square of age
Experience Exp EXP = Age--years of
schooling -school starting
age
Household Size Mem Number of family members
Employment Agri 1 = agriculture status,
Status Non-agri. Status 0 = otherwise.
(Dummy) (reference category)
Remittances Rem 1 remittance receiver, 0 =
(Dummy) Without remit. not receiver.
(reference category)
Gender Male lrnale, 0 = not male
(Dummy) Female (reference (female).
category)
Province Punjab l = Punjabi, 0 = not Punjabi
(Dummy) Sindh Similarly, we construct the
KPK Sindh and KPK variables.
Balochistan (reference
category)
Region Urban l = Urban,
(Dummy) Rural (reference 0 = otherwise (rural).
category)
Table 2
Logistic Estimates of Poverty Determinants of Pakistan
Variables Coefficients P-values Odd Ratios
Age .038 .000 * 1.039
[Age.sup.2] -.001 .000 * .999
Primary -.243 .063 *** .784
Middle -.769 .000 * .463
Matric -1.026 .000 * .358
Inter -2.020 .poo * .133
Ba -2.340 .000 * .096
Prof -2.227 .000 * .108
Urban -.906 .000 * .404
Male .597 .000 * 1.816
Punjab .661 .000 * 1.937
Sindh .267 .000 * 1.306
KPK .647 .000 * 1.909
Exp -.009 .008 *** .991
Mem .196 .000 * 1.216
Rem -.554 .000 * .575
Agri -.316 .000 * .729
Constant -3.269 .000 * .038
* denotes statistically significant at the 1 percent level.
** denotes statistically significant at the 5 percent level.
*** denotes statistically significant at the 10 percent level.
