Beyond the Poverty Line: A Multidimensional Analysis of Poverty in Pakistan
January 10, 2014
1.
Introduction
The issue of
poverty is both simple and complex: at one level, the implicit or explicit
objective of most economic policymakers in developing countries is to reduce
the levels of poverty in a country; at another level, the tools that should be
used are constantly open to debate. The debate in almost all developing
countries concerns the correct policy mix between those that target economic
growth, which have the potential to reduce poverty as overall income levels
rise, and those policies that target poverty directly, such as social safety
nets or income transfers.
There is
little question that long-term growth reduces poverty, but in a country such as
Pakistan, where growth is sporadic at best, the question that arises is what
can be done to reduce poverty for those who will not benefit from growth for
years or even decades. If we add to this the fact that development has been
devolved to the provinces in Pakistan after the 18th Constitutional Amendment,
the future of poverty alleviation initiatives is quite simple: Either we
explicitly acknowledge that the state is only concerned with economic growth
and wait for growth to reduce poverty—while pursuing intermittent and
idiosyncratic poverty interventions that assist the poor but do little to move
them out of poverty—or we clearly focus the limited resources of the state on
explicitly targeted poverty interventions.
The purpose of
this chapter is not to evaluate poverty reduction initiatives in Pakistan or to
predict what will happen to poverty in the country over the next decade.
Rather, we want to see what has been happening to poverty over the last decade
by looking at poverty in unique ways: First, we look at what has been happening
to poverty if we expand the definition of poverty to include not just income
but also other basic necessities such as health and education. After this, we
decompose poverty into two distinct categories in Pakistan: poverty as a result
of the circumstances into which one is born (such as where one was born or the
education of one’s parents) as opposed to poverty as a result of one’s own
efforts (such as the level of education one obtains). The principle objective
of our analysis is move the debate in Pakistan beyond the “poverty line” as
defined in income terms and toward a more comprehensive discussion of what
really makes people “poor” and the best ways of targeting poverty in Pakistan.
We believe that, in an environment characterized by constrained resources and
intermittent poverty reduction interventions, the only way one can make an
active difference to the level of poverty is to clearly identify which areas to
focus on.
The chapter is
structured as follows: Section 2 discusses some of the literature on poverty in
Pakistan as well as the more recent literature on poverty measurement. In
Section 3, we look at a wider view of poverty, which includes health and
education, and then at poverty trends from this wider perspective. In Section
4, we decompose poverty in that which can be explained by circumstances and
that which can be explained by effort. We then see what happens to poverty if
we were to equalize the circumstances in which people were born. Finally, we
present some conclusions from our analysis.
2.
Measuring Poverty
and the Literature on Poverty in Pakistan
2.1.
Trends in Poverty
and Inequality in Pakistan
On average,
poverty has decreased in Pakistan over the last decade or so, after having
risen in the 1990s. The proportion of people living below the international
poverty line, i.e., on less than USD 1.25 a day, went from 58.5 percent in 1990
to 36 percent in 2001/02, finally falling to 22.6 percent in 2004/05 (World
Bank, 2008, 2009). According to the national poverty line, poverty rose
steadily from 17.32 percent in 1987/88 to 25 percent in 1993/94, dipping to
21.6 percent in 1996/97 before rising sharply to 30.6 percent in 1998/99, and
finally peaking at 34.5 percent in 2000/01. After that, significant declines
occurred, culminating in a poverty headcount of 22.3 percent in 2005/06—almost
identical to the poverty rate seen in 1990/91. Households are vulnerable to
shocks, such as the food price crisis in the latter part of the last decade,
economic slowdowns, political turmoil, and other disruptions that can reverse
gains in poverty alleviation, moving vulnerable households back into poverty.
Khan (2011) and Anwar (2009) reach similar conclusions about the overall
fall in poverty when looking at other dimensions of wellbeing besides income
per capita. Khan (2011) finds that multidimensional poverty fell by five
percentage points between 1998 and 2008 to 38 percent. However, it was not a
steady decline as steep increases in poverty occurred both in 2001/02 and
2005/06. At the end of the period studied, poverty continued to be higher in
rural than urban areas, but the gap had narrowed. Naveed and Islam (2011) also
consider multidimensional poverty in Pakistan, noting that the most common
deprivations faced by households (in order of importance) are due to child
mortality, lack of land ownership, and children not enrolled in school.
Anwar (2009)
uses household data over the 2000–05 period to look at trends in consumption
and other measures of wellbeing across income deciles, finding that inequality
falls in terms of “opportunities” as measured by literacy, school enrollment
rates (at all levels), child immunization, pre- and postnatal care utilization,
access to electricity, and access to modern water and sanitation. However, he
notes that income inequality likely increased over the period since the richest
two deciles increased their share of consumption.
There are also
seasonal aspects to the relationship between income and household wellbeing,
particularly in rural areas. The vulnerability of low-wealth households can be
more acute during certain times of the year, particularly in rural areas.
Behrman, Foster, and Rosenzweig (1997) point out, in their analysis on rural
households’ joint or linked production-consumption decisions in Pakistan, that
food consumption is sensitive to wage income for low-wealth families (<1.5 acres) in the planting (lean) season, mainly due to
the high prices of food and credit. However, in a plentiful harvest season,
households’ consumption decisions are mostly invariant to income shocks. The
authors measure a small positive productivity effect (in terms of additional
output at harvest) of additional calorie consumption at the planting stage.
Shocks such as those to food prices can lead to large increases in
poverty, though these are usually transient. Ivanic, Martin, and Zaman (2012)
find that the spike in food prices in late 2010 that caused wheat prices to
double and led to 65–75 percent increases in sugar, maize, and soybean and palm
oils, led to a net increase in poverty of 44 million people around the world.
Further, they estimate that the poverty headcount in Pakistan increased by two
full percentage points (the second largest increase in the sample) due
primarily to the fact that the steep rise in global wheat prices was largely
passed on in the form of higher local prices.
The World Bank
(2011) also studies the effect of increasing food prices (particularly wheat)
between 2006 and 2010, finding that households would have needed 27 percent
more income between 2006 and June 2008 to maintain household utility levels in
the presence of the price increases (increasing further over the course of the
year), and that caloric availability fell by around 8 percent at the height of
the food price crisis in 2008. The study finds that households sold assets to
absorb the shock of higher food prices, and that those that owned land were
somewhat protected.
Chaudhry and
Chaudhry (2008) calculate the elasticity of the poverty gap (depth of poverty)
with respect to food prices as 2.1 percent and 0.44 percent with respect to
energy. Kurosaki (2006) also focuses on the effects of risk and shocks on
poverty when decomposing poverty into chronic and transient components. He
finds in Khyber Pakhtunkhwa (KP) that during 1996–99, more than half the
population was always poor, 13 percent were usually poor, and 16 percent were
occasionally poor. Earlier research by Alderman (1996) on the IFPRI panel from
the late 1980s of rural households in Pakistan demonstrates that, even though
households engage in precautionary saving, they experience a reduction in per
capita consumption and sell assets when faced with multiple shocks.
2.2.
Measurement of
Poverty
The simplest and
historically most commonly used indices to measure poverty include the poverty
headcount, the poverty gap, and the squared poverty gap. These indices are
known as the Foster-Greer-Thorbecke (FGT) measures of poverty. However, in
recent years, newer measures of poverty have been developed to capture
dimensions of poverty that are not contained in the FGT indices, such as
poverty across time or different dimensions of poverty. These include measures
that capture the detrimental impact of income fluctuations (see Kurosaki, 2006)
or measures that capture multiple dimensions of wellbeing (such as Alkire &
Foster, 2011) in the spirit of Sen’s capabilities approach. Alkire and Foster
(2011) first calculate deprivation on individual dimensions, such as
consumption, health, education, or empowerment, after which the number of
dimensions for which a person is deprived can be summed.
As we have seen, there is increasing recognition among development
economists that poverty is more than simply the shortfall of income or
consumption below a certain threshold, but rather the deprivation of households
along multiple dimensions. In fact, the relationship between consumption levels
and deprivation can be weak, as noted by Naveed and Islam (2010) in their
analysis of selected districts in Punjab and KP.[1]
According to their analysis, the official poverty line has missed a large share
of those who would be considered among the multidimensional poor. Measuring
poverty by the official poverty line, only around 18 percent would be
considered poor in these two provinces, but the share of households deprived in
either five or six dimensions is 36 or 25 percent, respectively.
Similarly,
McLeod (2006) finds for a larger group of countries including Pakistan that
calculations of poverty through household expenditure surveys do not correlate
highly with capabilities; rather, national accounts estimates of consumption
growth are better predictors of improvements in wellbeing.
Kurosaki (2006) notes that one of the unattractive features of the FGT
squared poverty gap measure is that it does not capture that income
fluctuations have more serious welfare implications the greater the depth of
poverty. Further, he finds that decompositions of poverty into chronic and
transient components using the FGT squared poverty gap is not robust to changes
in the poverty line, while the Clark-Watts measure which incorporates constant
relative risk aversion, performs significantly better.
3.
Incorporating
Education and Health into Poverty Measures
In this section,
we analyze unidimensional and multidimensional measures of poverty using
household-level data on income, health, and education in Pakistan. The reason
for this is that poverty is a multidimensional concept in that certain segments
of the population may be simultaneously deprived in numerous dimensions but
their poverty can be underestimated if one looks only at their income.
Similarly, certain segments of the population may be better off when assessed
across various dimensions of deprivation as compared to a simple analysis of
their income.
For the
purposes of this analysis, we focus on three dimensions of poverty: income,
education, and health. As is the standard practice (see Labar & Bresson,
2011), when looking at educational deprivation, we have restricted our sample
to individuals above the age of 20. In order to measure each type of
deprivation, we use the following data for 2004, 2008, and 2010 from the
Pakistan Social and Living Standards Measurement Survey: for income deprivation
we use household per capita income, obtained by dividing the total household
income by na, where n is the number of household members and
a is an equivalence factor. For
education, we use the number of years of education obtained by the household
member. For health, we use a combination of indicators, including household
sources of water and household sanitary conditions.
The income-based poverty line we use is USD 1.08 per day in 2004 and USD
1.25 per day in 2008 and 2010. For education-based poverty, we take a person to
be below the poverty line if they have not completed primary education; for
health-based poverty, we take a person to be below the poverty line if they do
not have a clean source of drinking water.
First, we look
at unidimensional comparisons of poverty over time. Here, we look at the
poverty trend using each of these indicators across gender, provinces, and
rural/urban divides. We then apply unidimensional stochastic dominance tests to
gauge what has been happening to poverty, using these indicators separately.
Following
this, we look at multidimensional poverty using all the indicators. Following
Alkire and Foster, we use both the intersection and union approaches to poverty
identification. The intersection approach implies that a person is considered
poor if they are deprived in all dimensions of poverty, namely income, health,
and education. In the union approach, a person is considered poor if they are
deprived in at least one dimension of poverty. As for the unidimensional
analysis, we initially look at the trend in multidimensional poverty across
gender, provinces, and between rural and urban areas.
3.1.
Analysis of Unidimensional
Poverty in Pakistan over Time
The
unidimensional poverty analysis checks to see if poverty has fallen when one
looks at the three indicators of deprivation, income, health, and education,
separately over time. We start with a look at the statistics on poverty levels,
using each of these three indicators over time, and then carry out a stochastic
dominance analysis to see if poverty has fallen significantly over the last
decade in Pakistan.
Looking at
poverty from the perspective of income, Table 17.1 shows the overall percentage
of people below the poverty line for the years 2004, 2008, and 2010. What the
numbers show is that that there is a significant decline in the number of
people below the poverty line between 2004 and 2008 from approximately 20 to 16
percent. From 2008 to 2010, the percentage of people below the poverty line
rises to approximately 16.5 percent but this number is not statistically
different from the corresponding number from the 2008 data, which implies that
the percentage is approximately the same between 2008 and 2010.
Table 17.1: Percentage of population below the income-based
poverty line for the entire population
2004
|
2008
|
2010
|
|
Below poverty line
|
20.8
|
16.2
|
17.5
|
Above poverty line
|
79.2
|
83.8
|
82.5
|
Total
|
100.0
|
100.0
|
100.0
|
Source: Authors’ calculations.
Table 17.2
shows what has happened in the rural and urban areas in terms of
income-measured poverty: Between 2004 and 2008, there is a significant decrease
in the percentage of the urban population living below the poverty line, from
approximately 11.5 to 9.2 percent. After this fall, the level of urban poverty
stays approximately the same from 2008 to 2010. A greater fall occurs in rural
poverty from 2004 to 2008, with the percentage of the rural population living
below the poverty line falling from approximately 25 to 19 percent. The level
of rural poverty rises between 2008 to 2010 from approximately 19 to 20
percent, but this increase is again not statistically significant.
Table
17.2: Percentage of population below the income-based
poverty line for the entire population: Rural vs. urban breakdown
2004
|
2008
|
2010
|
||||
Urban
|
Rural
|
Urban
|
Rural
|
Urban
|
Rural
|
|
Below
poverty line
|
11.84
|
25.8
|
9.48
|
19.82
|
10.09
|
21.44
|
Above
poverty line
|
88.16
|
74.12
|
90.52
|
80.18
|
89.91
|
78.56
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
Finally Table
17.3 shows the breakdown of income-based poverty across the provinces. Again,
there is a significant decrease in the percentage of people living below the
poverty line in all four provinces between 2004 and 2008, with the greatest
decrease occurring in KP where the percentage of people living below the
poverty line falls from 34 to 22 percent. In Punjab, Sindh, and Balochistan,
this trend of falling poverty continues between 2008 and 2010, but what is
striking is that the percentage of people living below the poverty line rises
significantly in KP between 2008 and 2010 from 22 to 26 percent.
Table
17.3: Percentage of population below the income-based
poverty line for the entire population: Provincial breakdown
2004
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Below poverty line
|
20.00
|
16.02
|
34.66
|
12.22
|
Above poverty line
|
79.99
|
83.97
|
65.33
|
79.19
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
2008
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Below poverty line
|
16.46
|
14.27
|
22.42
|
11.79
|
Above poverty line
|
83.53
|
85.72
|
77.57
|
88.20
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
2010
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Below poverty line
|
18.8
|
12.9
|
28.7
|
14.1
|
Above poverty line
|
83.2
|
87.1
|
71.2
|
85.9
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
Moving on to
the second indicator of poverty, education, we focus on the percentage of
people without primary education. Because individuals below a certain age may
either still be in the process of obtaining education or still have the
opportunity to complete their primary education, we have concentrated on
individuals above the age of 20. The overall breakdown of people above the age
of 20 who have not completed their primary education is shown in Table 17.4. As
the numbers illustrate, there was a significant decrease in the number of
people above the age of 20 without primary education between 2004 and 2008,
from approximately 59 to 55.5 percent, but there is no statistically
significant decrease in this number between 2008 and 2010.
Table
17.4: Breakdown of population above the age of 20 who have
not completed their primary education (%)
Only for ages 20 and above
|
2004
|
2008
|
2010
|
Education below primary
|
59.1
|
55.4
|
54.9
|
Education above primary
|
40.9
|
44.6
|
45.1
|
Total
|
100.0
|
100.0
|
100.0
|
Source: Authors’ calculations.
Similarly,
Table 17.5 shows that there is a significant fall in the number of men and
women above the age of 20 without primary education from 2008 to 2010, with the
percentage of men without primary education falling by almost four percentage
points from 44.6 to 40.6 percent and the percentage of women without primary
education falling from 73.9 to 70.5 percent. Again, there are no significant
changes in these numbers between 2008 and 2010.
Table
17.5: Breakdown by gender of population above the age of 20
who have not completed their primary education
2004
|
2008
|
2010
|
||||
Only for ages 20 and above
|
Male
|
Female
|
Male
|
Female
|
Male
|
Female
|
Education below
primary
|
44.67
|
73.88
|
40.62
|
70.52
|
40.45
|
69.28
|
Education above
primary
|
55.33
|
26.12
|
59.38
|
29.48
|
59.55
|
30.72
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
Tables 17.6
and 17.7 show the breakdown of people above the age of 20 without primary
education in rural and urban areas and across provinces between the years 2004
and 2010. The three important points to note are: First, the significant
decreases in the percentage of people without primary education between 2004
and 2008 (with a fall from 40.4 to 37.3 percent in the percentage of adults
without primary education in urban areas and a corresponding fall from 70.4 to
66 percent in the percentage of adults without primary education in rural
areas); second, the disparity in the percentage of adults without education
when comparing rural and urban areas and the provinces of KP and Balochistan
versus the provinces of Punjab and Sindh (with the former having a significantly
higher percentage of adults without primary education than the latter); third,
the lack of any significant change in all these percentages between 2008 and
2010.
Table
17.6: Rural/urban breakdown of population above the age of
20 who have not completed their primary education
Only for ages 20 and above
|
2004
|
2008
|
2010
|
||||||
Urban
|
Rural
|
Total
|
Urban
|
Rural
|
Total
|
Urban
|
Rural
|
Total
|
|
Education below
primary
|
40.44
|
70.42
|
59.10
|
37.35
|
65.98
|
55.37
|
36.60
|
65.79
|
54.88
|
Education above primary
|
59.56
|
29.58
|
40.90
|
62.65
|
34.02
|
44.63
|
63.40
|
34.21
|
45.12
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
Table 17.7: Province-wise breakdown of
population above the age of 20 who have not completed their primary education
2004
|
||||
Only for ages 20 and above
|
Punjab
|
Sindh
|
KP
|
Balochistan
|
Education below primary
|
53.39
|
56.40
|
66.77
|
73.40
|
Education above primary
|
46.61
|
43.60
|
33.23
|
26.60
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
2008
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Education below primary
|
48.70
|
52.32
|
62.53
|
70.59
|
Education above primary
|
51.30
|
47.68
|
37.47
|
29.41
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
2010
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Education below primary
|
47.95
|
52.26
|
61.61
|
72.66
|
Education above primary
|
52.05
|
47.74
|
38.39
|
27.34
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
The final dimension of poverty that we focus on is health, as measured
by the source of drinking water available to individuals. As discussed above,
we have characterized the sources of drinking water into categories above and
below the poverty line. The overall picture for the years 2004 to 2010 is given
in Table 17.8. Here we see an interesting reversal in the trend observed for
the other indicators of poverty: From 2004 to 2008, the percentage of people
below the poverty line as measured by their source of drinking water increased
from approximately 11 to 15 percent, and this higher level was also observed in
2010.
Table 17.8: Breakdown of population by
access to drinking water (%)
2004
|
2008
|
2010
|
|
Source of drinking water below poverty cut-off
|
10.9
|
15.1
|
14.6
|
Source of drinking water above poverty cut-off
|
89.1
|
84.9
|
85.4
|
Total
|
100.0
|
100.0
|
100.0
|
Source: Authors’ calculations.
Tables 17.9 and 17.10 show where this increase in poverty has occurred:
Table 17.9 shows the breakdown between rural and urban poverty in terms of
sources of drinking water and the numbers show a significant increase in the
poverty rate (or, in other words, a significant deterioration in the sources of
drinking water) in rural areas with an increase from 16 to 22 percent of the
rural population falling below the poverty line in terms of sources of drinking
water, while the percentage of people with poor sources of drinking water in
the urban areas is significantly lower at around 1.5 percent.
Table 17.9: Rural-urban breakdown of
population by access to drinking water
2004
|
2008
|
2010
|
|||||||
Urban
|
Rural
|
Total
|
Urban
|
Rural
|
Total
|
Urban
|
Rural
|
Total
|
|
Source of drinking
water below poverty cut-off
|
1.52
|
16.23
|
10.91
|
1.48
|
22.30
|
15.07
|
1.27
|
21.85
|
14.63
|
Source of drinking
water above poverty cut-off
|
98.48
|
83.77
|
89.09
|
98.52
|
77.70
|
84.93
|
98.73
|
78.15
|
85.37
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
Table 17.10 shows the provincial breakdown of people below the poverty
line in terms of sources of drinking water; the numbers are striking because of
the significantly lower levels of poverty in Punjab and Sindh compared to KP
and Balochistan. Table 17.10 also shows where the greatest increase in poverty,
as measured by the sources of drinking water, has occurred in Balochistan, with
the number increasing significantly from 31 to 49 percent of the population
being classified as poor in terms of their sources of drinking water. Finally,
Tables 17.9 and 17.10 show that, like income-based and health-based poverty,
there has been no significant change in poverty as measured by sources of
drinking water between the years 2008 and 2010.
Table 17.10: Province-wise breakdown of
population by access to drinking water
2004
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Source of drinking
water below poverty cut-off
|
1.14
|
6.41
|
24.06
|
31.89
|
Source of drinking
water above poverty cut-off
|
98.86
|
93.59
|
75.94
|
68.11
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
2008
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Source of drinking
water below poverty cut-off
|
1.41
|
7.95
|
24.07
|
49.21
|
Source of drinking
water above poverty cut-off
|
98.59
|
92.05
|
75.93
|
50.79
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
2010
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Source of drinking
water below poverty cut-off
|
1.83
|
7.83
|
21.77
|
49.92
|
Source of drinking
water above poverty cut-off
|
98.17
|
92.17
|
78.23
|
50.08
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
Thus, the numbers show that income-based poverty and education-based
poverty declined between 2008 and 2010 while poverty measured by sources of
drinking water increased between these years. Additionally, the numbers for all
three poverty indicators show no significant changes between 2008 and 2010.
The obvious question that arises is whether these
changes over time are significant for poverty when measured in terms of income
as well as when measured in terms of education and health. Table A17.1 gives
the results of unidimensional stochastic dominance tests for each of the
poverty measures. The results are interesting in that they show that
income-based and education-based poverty fell significantly between 2004 and
2008 while health-based poverty did not change significantly (even though the
results imply that it actually rose, these results were not significant).
Moreover, even though the results seem to be implying an increase in
all three poverty measures between 2008 and 2010, none of these are statistically
significant, implying that there was no significant change in poverty between
2008 and 2010. A comparison of these different types of poverty measures
between 2004 and 2010 yields the same results as those between 2004 and 2008:
income- and education-based poverty decreased but the health-based poverty
measure did not show a significant change over this period.
The rural/urban breakdowns of these unidimensional poverty measures
given in Tables A17.2 and A17.3 are informative: Between 2004 and 2008, all
three measures of poverty indicate that urban poverty fell significantly,
whereas only income- and education-based poverty fell during in rural areas.
What is interesting is that there was no significant change in income-based
poverty in either rural or urban areas between 2008 and 2010, but there was a
significant decline in education- and health-based poverty in urban areas
between 2008 and 2010, which was not matched by any significant changes in
education- and health-based rural poverty. Over the longer period 2004–08,
urban poverty declined using all three measures of poverty while only income-
and education-based poverty declined in the rural areas during this period.
The province-wise breakdowns of unidimensional poverty measures are
given in Tables A17.4–A17.7. It is worth noting, first, that while income-based
poverty fell in all the provinces between 2004 and 2008, education-based
poverty fell in three of the four provinces (the outlier being Balochistan) and
health-based poverty only fell in Punjab and KP. The second interesting
observation is that, while there was no significant change in income-based
poverty for all three provinces between 2008 and 2010, the only province that
experienced a decline in education- and health-based poverty was Sindh.
Finally, the statistics show that only KP did not experience a net decline in
income-based poverty over the longer period 2004–10, while Balochistan was the
only province that did not experience a net fall in education-based poverty
during this period. Over the longer period, it seems that only Punjab
experienced a significant decline in health-based poverty levels.
3.2.
Analysis of Multidimensional Poverty
This section looks at multidimensional poverty, using a combination of
income, health, and education indicators, and then performing tridimensional
dominance tests (using these indicators) to see if multidimensional poverty has
fallen over time in Pakistan.
Our approach uses the “intersection” of the various poverty measures to
measure the poverty rate. By this we mean that we consider an individual to be
above the poverty line if they fall below the poverty cutoff in at least one of
the indicators while they are considered to be above the poverty line if all
three of their indicators lie above the poverty cutoff. So an individual is
considered to be above the poverty line if they are considered not poor in
terms of income, health, and education; if any of these indicators lie below
the poverty cutoff, they are considered poor.
Though we acknowledge that this is a far more stringent criterion for
being above the poverty line than most commonly used measures, our rationale is
that it allows us to build a more comprehensive picture of the state of poverty
in Pakistan. So if 70 percent of the population lies below the poverty line
according to this multidimensional measure, it does not imply that the figures
we obtained for poverty above are incorrect; rather, it implies that, from the
point of view of development, 70 percent of the population is still deprived in
at least one dimension of poverty.[2]
Table 17.11 shows the percentage of the overall population that is poor
in terms of at least one criterion. Here, we see that, according to our
multidimensional view of poverty, almost 73 percent of the overall population
is considered to be deprived in at least one of our key criteria (income,
health, and education). This number falls to approximately 70 percent in 2008
after which is remains constant in 2010.
Table 17.11: Percentage breakdown of
overall population below the multidimensional poverty line (income, education,
and health)
2004
|
2008
|
2010
|
|
Above cut-off in all three
indicators
|
27.4
|
30.2
|
30.5
|
Below cut-off (poor) in at
least one indicator
|
72.6
|
69.8
|
69.5
|
Total
|
100.0
|
100.0
|
100.0
|
Source: Authors’ calculations.
Table 17.12 shows the rural/urban breakdown of multidimensional
poverty, indicating the stark difference between multidimensional poverty in
the urban and rural areas: Urban multidimensional poverty is almost 25 percent
lower than rural multidimensional poverty (56 percent as opposed to 82
percent), and though there is a slight fall in both poverty levels between 2004
and 2008 (after which it stays relatively constant), the wide gap between urban
and rural multidimensional poverty persists.
Table 17.12: Rural/urban breakdown of
overall population below the multidimensional poverty line (income, education,
and health)
2004
|
2008
|
2010
|
|||||||
Urban
|
Rural
|
Total
|
Urban
|
Rural
|
Total
|
Urban
|
Rural
|
Total
|
|
Above cut-off in all
three indicators
|
43.99
|
18.01
|
27.39
|
47.48
|
20.99
|
30.18
|
48.48
|
20.78
|
30.49
|
Below cut-off (poor)
in at least one indicator
|
56.01
|
81.99
|
72.61
|
52.52
|
79.01
|
69.82
|
51.52
|
79.22
|
69.51
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
Table 17.13 shows the provincial breakdown of multidimensional poverty.
What is striking is the large gap between the level of multidimensional poverty
in Punjab and Sindh versus that in KP and Balochistan. So while
multidimensional poverty falls in Punjab, Sindh, and KP between 2004 and 2010
(with the most significant falls taking place in Punjab and Sindh), there is no
significant decrease in multidimensional poverty in Balochistan over this
period.
Table 17.13: Province-wise breakdown of
overall population below the multidimensional poverty line (income, education,
and health)
2004
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Above cut-off in all three
indicators
|
33.38
|
29.77
|
18.33
|
16.76
|
Below cut-off (poor) in at
least one indicator
|
66.62
|
70.23
|
81.67
|
83.24
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
2008
|
||||
Punjab
|
Sindh
|
KP
|
Balochistan
|
|
Above cut-off in all three
indicators
|
37.78
|
33.15
|
22.91
|
15.22
|
Below cut-off (poor) in at
least one indicator
|
62.22
|
66.85
|
77.09
|
84.78
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
2010
|
||||
Punjab
|
Sindh
|
KPK
|
Balochistan
|
|
Above cut-off in all three
indicators
|
46.46
|
42.31
|
28.30
|
17.60
|
Below cut-off (poor) in at
least one indicator
|
53.54
|
57.69
|
71.70
|
82.40
|
Total
|
100.00
|
100.00
|
100.00
|
100.00
|
Source: Authors’ calculations.
4.
Separating Circumstances from Effort in Income Determination
A critical but often ignored question in the discussion on poverty is
the impact of opportunities available to individuals on their poverty status.
Obviously, there is a significant difference between the economic opportunities
available at birth to individuals in Pakistan and the same amount of effort may
be enough to take one individual out of poverty but insufficient in another’s
case. Roemer (1998), who formalized this idea, suggested decomposing economic
outcomes such as income into circumstances and effort. By circumstances, we
mean those exogenous factors that a person has no control over, such as gender,
family background, or place of birth; by effort, we mean those factors that are
affected by the choices made by individuals (or families), such as education or
occupation.
In this way, economic outcomes such as income, and therefore poverty,
are jointly determined by factors that are beyond the control of individuals
and factors that a person can control through his/her choices. This is a
powerful idea because not only can one say that a person’s income lies below
the poverty line, but one can also associate this income with observed
circumstances beyond the control of individuals. In other words, we divide the
determinants of income in Pakistan into factors determined by the circumstances
into which an individual was born, and factors that are determined in his/her
own life.
The small budgets and limited capacity that developing countries such
as Pakistan have to work with limit their ability to equalize economic outcomes
for individuals. For these reasons, in addition to notions of redistributive
justice, a policy that aims to equalize opportunities is an attractive approach
because it combines the idea of personal responsibility (that outcomes should
vary with effort) with the recognition that many individuals are born into
disadvantaged circumstances through no fault of their own.
Drawing on the concept of effort versus circumstances delineated by
Roemer (1998), Bourguignon, Ferreira, and Menendez (2007) use data from Brazil
on men’s hourly earnings and determine that between 10 and 37 percent of the
earning inequality for men aged 26 to 60 could be eliminated if they faced
equal circumstances, which included race, place of birth, parents’ education,
and father’s occupation. In their exercise, income is determined by three major
factors: circumstances, effort, and unobserved individual characteristics (such
as intrinsic ability and motivation). Circumstances are fully exogenous to
income. However, they allow individual effort, including education, occupation,
and migration, to be determined in part by circumstances. The motivation for
this, in the case of education for instance, is that the value that parents
place on education may depend in part on the family’s socioeconomic status,
thus influencing the amount of education attained by their offspring,
regardless of ability or intrinsic motivation.
This analysis is intuitively appealing, but it requires very specific
data on circumstances such as place of birth and parents’ education levels, as
well as data on effort, such as education level and occupation. Though much of
the latter information is available in household-level datasets in Pakistan,
the former is not available in virtually any dataset. Thus, we take a unique
approach to this problem: Using the PSLM data for 2010/11, we restrict our
sample to working men aged 20 to 30 who are living with their fathers. This
helps minimize the sample selection issue that would arise if men living with
their fathers differed fundamentally from those who had left, in ways that were
not exogenous to income.
For example, sample selection bias could occur if men who are more
intrinsically motivated are more likely to move away from their parents and
earn higher incomes. Since the majority of men in the 20–30-years-old age
bracket will not yet have moved from their natal household, we feel that taking
this subsample will minimize the sample selection bias. Given the different
circumstance and effort variables that we consider, we lose additional
observations due to missing data. We use the characteristics of the fathers in
the households and separate the impact on income of the circumstances into
which these men were born from that of their efforts.
4.1. Persistence
of Income and Educational Attainment over Generations
However, before studying the joint effects of circumstance and effort
variables, it is useful to take a step back and look at the persistence of key
economic characteristics across generations. These are referred to as Galtonian
regressions and take the form
(1)
where refers to the
attainment of the current generation, in this case income or education, and refers to the
attainment of the previous generation. The coefficients in each case
indicate the persistence across generations and therefore allows us to
measure intergenerational mobility. The greater the elasticity , the greater is the persistence of inequality across
generations; an elasticity of 1 would indicate that incomes are perfectly
correlated between generations of the same family.
Table 17.14 reports the results of these Galtonian regressions of
intergenerational persistence of income and educational attainment. For each of
the regressions, we run the Galtonian regression on two samples; in the first,
we use the largest sample available for men aged 20 to 30, and in the second we
use a smaller sample that is comparable with that used in the upcoming
regressions to compute the effects of unequal opportunity on income.
Table 17.14: Galtonian regressions of
intergenerational persistence: Income and education
(1a)
|
(1b)
|
(2a)
|
(2b)
|
(3a)
|
(3b)
|
|
Ln(yearly
total income)
|
Ln(yearly
total income)
|
Ln(1+education)
(in years)
|
Ln(1+education)
(in years)
|
Ln(1+education)
(in years)
|
Ln(1+education)
(in years)
|
|
Ln (father’s annual
total income)
|
0.40
(23.5)
|
0.36
(15.1)
|
||||
Ln (1 + father’s
education)
(in years)
|
0.09
(10.4)
|
0.1
(3.6)
|
||||
Ln (1 + mother’s
education)
(in years)
|
0.31
(34.2)
|
0.5
(16.7)
|
||||
Sample
|
All men aged 20-30 living with their fathers
|
Same sample as in full regressions (below)
|
All men aged 20-30 living with their fathers
|
Comparable to sample used in full regressions (below)
|
All men aged 20-30 living with their fathers
|
Comparable to sample used in full regressions (below)
|
Number of
observations
|
2844
|
1493
|
14741
|
1204
|
13753
|
1137
|
Source:
Authors’ calculations.
The results of Table 17.14 tell us that the income of men aged 20 to 30
has an elasticity of approximately 0.4 with respect to their fathers’ income.
That is, on average, a young man’s income tends to be 4 percent higher if his
father’s income is 10 percent higher. When it comes to education, the
attainment of men aged 20 to 30 is much more highly correlated with that of
their mothers (elasticity 0.3–0.5) compared to their fathers (elasticity of
0.1). A young man’s education tends to be more significantly affected by the
level of education of his mother compared to that of his father.
On the whole, the results imply that parents’ income and education are
significant in determining if young men lie above or below the poverty line.
4.2.
The Effect of Effort and Circumstances on Income
In this analysis, following Bourguignon et al. (2007), we analyze
separately the factors that affect a person’s income level into those factors
that a person faces at birth and has no control over, or “circumstances”, and
those that are determined during the life of a person, or “effort”. We then
analyze how the equalization of circumstances can reduce the inequality of
outcomes. In determining these, two critical points emerge: First, we can
assess which factors should be targeted in the effort to reduce poverty in
Pakistan. Second, we can see that, in an environment of limited resources, what
would be the impact on inequality of trying to simply reduce the unequal
circumstances into which people are born instead of generally targeting the
poor equally throughout Pakistan.
In our estimations, the circumstances that we consider are the average
educational attainment of the man’s parents (AEP), the father’s total annual income from all sources (FY), occupation dummies for the father (FOD), the household’s wealth index (W), and regional dummies for place of
residence (RD). The regression
equation for the total effect of circumstances on income is:
(2)
where w is the yearly total
income from all sources for men aged 20 to 30.
These results show that average parental educational attainment and
household wealth has no real impact on earnings. However, the father’s income
and occupational status have impacts that are both large and statistically
significant. More specifically, the children of sharecroppers have the lowest
incomes while the children of farmers who own their land have slightly higher
incomes. Interestingly, people with parents who are wage-paid employees tend to
have higher incomes than the children of farmers while the highest incomes are
those of children with parents who are self-employed or are employers. So, if
one wanted to target poverty reduction funds, one should aim to focus on areas
with significant populations of landless agricultural workers.
The results also show that the region of residence plays an important
role. Men living in the urban areas of northern Punjab have the highest
incomes, followed by those in Karachi. With the exception of KP, men in urban
areas generally have higher incomes than their rural counterparts. Thus, in
terms of targeting, central Punjab, southern Punjab, and northern Punjab are
areas in which income levels tend to be the lowest after controlling for all
other factors.
If, instead, we want to compute the
direct effect of circumstances, in other words controlling for individual
efforts, we start with the following expression:
(3)
The circumstance variables, C, are as above. The effort variables, E, chosen by the individual male aged 20
to 30 are education (ED) and the
decision to work in the agricultural sector (AG). The regression specification we use to estimate the direct
effect of circumstances C,
controlling for efforts is:
(4)
The estimation results for equation
4 can be found in Table A17.9. The results show that the individual’s own
education has the expected positive and nonlinear relationship with earnings;
an additional year of education increases annual total income by about 3.9
percent, emphasizing the important role of education in reducing poverty.
The father’s income and occupational
status have important effects on earnings. The elasticity of own income with
respect to the father’s income is 31 percent. Compared to men whose fathers are
sharecroppers, those with fathers with higher occupational status also earn
more, with the increase in wage rising with the occupational status. Men with
fathers who are owner-cultivators, wage earners, or have their own business
earn 20, 38, and 46 percent more than those whose fathers are sharecroppers.
The region of residence is also important, with the highest incomes earned by
those in northern urban Punjab and Karachi. Thus, the circumstances into which
people in Pakistan are born have a significant impact on their future income
levels.
Following Bourguignon et al (2007),
we compare the actual earnings distribution with the theoretical distribution
generated by equalizing circumstances and controlling for individual efforts
using an inequality measure, I. In
other words, we measure income inequality by equalizing the circumstances into
which every person in the sample was born. Indices of inequality for the
distribution of actual earnings, as well as the hypothetical earnings
distributions computed with equalized circumstances and the residual inequality
are summarized in Table 17.15.
In our results, equalizing all
circumstances reduces the Theil index from 0.5 to 0.46—a fall of about 8.2
percent—when using the regressed sample of 1,493 observations. If we expand the
sample to include observations for which effort variables are missing, the
reduction in inequality is 7 percent as measured by the Theil index. Thus,
differing circumstances at birth lead to significant differences in income, and
equating circumstances across Pakistan could have a significant impact on
poverty reduction.
Table 17.15: Decomposition of inequality
measures due to unequal opportunities
Gini
|
Theil
|
Residual
inequality after equalizing circumstances (share) GINI
|
Residual
inequality after equalizing circumstances (share) THEIL
|
Number of
observations
|
|
Regressed sample, 1,493 obs.
|
|||||
Actual income
distribution
|
0.479
|
0.501
|
Na
|
Na
|
1,493
|
Equalizing all
circumstances
|
0.454
|
0.460
|
0.052
|
0.082
|
1,493
|
Equalizing family
circumstances (only)
|
0.460
|
0.464
|
0.040
|
0.074
|
1,493
|
Equalizing all
circumstances, conditional on effort
|
0.454
|
0.460
|
0.052
|
0.082
|
1,493
|
Equalizing family
circumstances (only), conditional on effort
|
0.459
|
0.464
|
0.042
|
0.074
|
1,493
|
Expanded sample, 2,148 obs.
|
|||||
Actual income
distribution
|
0.469
|
0.471
|
Na
|
Na
|
2,148
|
Equalizing all
circumstances
|
0.449
|
0.438
|
0.043
|
0.070
|
2,148
|
Equalizing family
circumstances (only)
|
0.450
|
0.436
|
0.041
|
0.074
|
2,148
|
Source: Authors’ calculations.
5.
Conclusions
This chapter’s
analysis is significantly different from typical analyses of poverty. We have
looked at poverty as a multidimensional phenomenon and found that there are
more significant differences between the rural and urban areas and between
provinces if one expands the definition of poverty to include income levels,
health indicators, and educational attainment.
Urban multidimensional poverty is
almost 25 percent lower than rural multidimensional poverty and though there is
a slight fall in both poverty levels between 2004 and 2008, the gap between
urban and rural multidimensional poverty persists. Additionally, the provincial
breakdown of multidimensional poverty shows a large gap between the level of
multidimensional poverty in Punjab and Sindh versus that in KP and Balochistan.
While multidimensional poverty falls in Punjab, Sindh, and KP between 2004 and
2010, there is no significant decrease in multidimensional poverty in
Balochistan over this period.
We also adopt a new approach to
looking at the factors that affect income levels in Pakistan. By dividing the
factors that affect income into the circumstances into which people are born
(and which they have no control over) and those that people can influence
through their own efforts, we are able to gauge if targeting poverty reduction
initiatives can have a significant impact on income levels.
We find that factors such as
education have a significant impact on determining a person’s income. Thus, an
obvious way to reduce poverty would be to promote education in the least
developed areas. A more interesting result is that the circumstances into which
people are born have a significant impact on whether they will have low incomes
or not; put another way, controlling for all the usual factors such as education,
just being the child of a less educated father or sharecropper significantly
reduces the levels of income a person will have and increases the chances that
he/she will fall below the poverty line. We also find a significant decrease in
income inequality if we equalize the circumstances under which people are born.
This final result is critical
because normal policy reduction initiatives have a household-level focus: they
are less concerned about targeting regions or communities and more concerned about
targeting the lowest-income households. Our results imply that, in a situation
where resources are constrained, there can be significant reductions in income
inequality if one focuses on equalizing the circumstances under which people
are born. This may mean that one should not have to target households with the
lowest incomes, but rather target areas that have the most significant
persistence of low incomes.
Appendix
Table A17.1: Unidimensional stochastic
dominance tests for entire population
Comparison years
|
Income
|
Education
|
Health
|
2004-2008
|
≻
(0)***
|
≻
(0)***
|
≺
(1)
|
2008-2010
|
≺
(1)
|
≺
(0.6)
|
≺
(1)
|
2004-2010
|
≻
(0)***
|
≻
(0)***
|
≺
(1)
|
Note: ≻ (≺) indicates that the final distribution dominates (is
dominated by) the initial distribution at the corresponding order of dominance.
P-values are reported in parentheses and refer to the hypothesis of
nondominance of the initial distribution with respect to the final one when ≺ is reported and to the hypothesis of nondominance of
the final distribution with respect to the initial one in the other cases. For
simplicity, values less than 10-3 or greater than 1–10-3
are replaced by 0 and 1, respectively. *** indicates that the result is
significant at the 1 percent level, ** indicates that the result is significant
at the 5 percent level.
Table A17.2: Unidimensional stochastic
dominance tests for urban population
Comparison years
|
Income
|
Education
|
Health
|
2004-2008
|
≻
(0)***
|
≻
(0)***
|
≻
(0)***
|
2008-2010
|
≺
(1)
|
≻
(0.02)***
|
≻
(0.01)***
|
2004-2010
|
≻
(0)***
|
≻
(0)***
|
≻
(0)***
|
Note: ≻ (≺) indicates that the final
distribution dominates (is dominated by) the initial distribution at the
corresponding order of dominance. P-values are reported in parentheses and
refer to the hypothesis of nondominance of the initial distribution with respect
to the final one when ≺ is reported and to the
hypothesis of nondominance of the final distribution with respect to the
initial one in the other cases. For simplicity, values less than 10-3
or greater than 1–10-3 are replaced by 0 and 1, respectively. *** indicates
that the result is significant at the 1 percent level, ** indicates that the
result is significant at the 5 percent level.
Table A17.3: Unidimensional stochastic
dominance tests for rural population
Comparison years
|
Income
|
Education
|
Health
|
2004-2008
|
≻
(0)***
|
≻
(0)***
|
≺
(0.99)
|
2008-2010
|
≺
(1)
|
≻
(0.99)
|
≻
(0.16)
|
2004-2010
|
≻
(0)***
|
≻
(0)***
|
≺
(0.97)
|
Note: ≻ (≺) indicates that the final
distribution dominates (is dominated by) the initial distribution at the
corresponding order of dominance. P-values are reported in parentheses and
refer to the hypothesis of nondominance of the initial distribution with respect
to the final one when ≺ is reported and to the
hypothesis of nondominance of the final distribution with respect to the
initial one in the other cases. For simplicity, values less than 10-3
or greater than 1–10-3 are replaced by 0 and 1, respectively. *** indicates
that the result is significant at the 1 percent level, ** indicates that the
result is significant at the 5 percent level.
Table A17.4: Unidimensional stochastic
dominance tests for Punjab
Comparison years
|
Income
|
Education
|
Health
|
2004-2008
|
≻
(0)***
|
≻
(0)***
|
≻
(0)***
|
2008-2010
|
≺
(0.99)
|
≺
(0.76)
|
≻
(0.17)
|
2004-2010
|
≻
(0)***
|
≻
(0)***
|
≻
(0)***
|
Note: ≻ (≺) indicates that the final
distribution dominates (is dominated by) the initial distribution at the
corresponding order of dominance. P-values are reported in parentheses and
refer to the hypothesis of nondominance of the initial distribution with respect
to the final one when ≺ is reported and to the
hypothesis of nondominance of the final distribution with respect to the
initial one in the other cases. For simplicity, values less than 10-3
or greater than 1–10-3 are replaced by 0 and 1, respectively. ***
indicates that the result is significant at the 1 percent level, ** indicates
that the result is significant at the 5 percent level.
Table A17.5: Unidimensional stochastic
dominance tests for Sindh
Comparison years
|
Income
|
Education
|
Health
|
2004-2008
|
≻
(0)***
|
≻
(0.01)***
|
≺
(1)
|
2008-2010
|
≺
(1)
|
≻
(0)***
|
≻
(0)***
|
2004-2010
|
≻
(0)***
|
≻
(0)***
|
≺
(0.99)
|
Note: ≻ (≺) indicates that the final
distribution dominates (is dominated by) the initial distribution at the
corresponding order of dominance. P-values are reported in parentheses and
refer to the hypothesis of nondominance of the initial distribution with respect
to the final one when ≺ is reported and to the
hypothesis of nondominance of the final distribution with respect to the
initial one in the other cases. For simplicity, values less than 10-3
or greater than 1–10-3 are replaced by 0 and 1, respectively. *** indicates
that the result is significant at the 1 percent level, ** indicates that the
result is significant at the 5 percent level.
Table A17.6: Unidimensional stochastic
dominance tests for KP
Comparison years
|
Income
|
Education
|
Health
|
2004-2008
|
≻
(0)***
|
≻
(0)***
|
≻
(0.03)**
|
2008-2010
|
≺
(1)
|
≻
(0.84)
|
≺
(0.80)
|
2004-2010
|
≻
(0.11)
|
≻
(0)***
|
≺
(0.14)
|
Note: ≻ (≺) indicates that the final
distribution dominates (is dominated by) the initial distribution at the
corresponding order of dominance. P-values are reported in parentheses and
refer to the hypothesis of nondominance of the initial distribution with respect
to the final one when ≺ is reported and to the
hypothesis of nondominance of the final distribution with respect to the
initial one in the other cases. For simplicity, values less than 10-3
or greater than 1–10-3 are replaced by 0 and 1, respectively. *** indicates
that the result is significant at the 1 percent level, ** indicates that the
result is significant at the 5 percent level.
Table A17.7: Unidimensional stochastic
dominance tests for Balochistan
Comparison years
|
Income
|
Education
|
Health
|
2004-2008
|
≻
(0)***
|
≺
(1)
|
≺
(1)
|
2008-2010
|
≺
(1)
|
≺
(1)
|
≺
(1)
|
2004-2010
|
≻
(0)***
|
≺
(0.98)
|
≺
(1)
|
Note: ≻ (≺) indicates that the final
distribution dominates (is dominated by) the initial distribution at the
corresponding order of dominance. P-values are reported in parentheses and
refer to the hypothesis of nondominance of the initial distribution with respect
to the final one when ≺ is reported and to the
hypothesis of nondominance of the final distribution with respect to the
initial one in the other cases. For simplicity, values less than 10-3
or greater than 1–10-3 are replaced by 0 and 1, respectively. *** indicates
that the result is significant at the 1 percent level, ** indicates that the
result is significant at the 5 percent level.
Table A17.8: Regression of log total
yearly income on only circumstance variables for men aged 20–30
Variable
|
Min
|
OLS
|
Max
|
t-stat
|
Average education of
parents
|
-0.014
|
-0.014
|
-0.014
|
-2.63
|
Ln (father’s total
annual income)
|
0.307
|
0.308
|
0.308
|
11.93
|
Father is employer or
self-employed (dummy)
|
0.425
|
0.434
|
0.443
|
3.87
|
Father is wage-paid
employee (dummy)
|
0.346
|
0.355
|
0.364
|
3.40
|
Father is
owner-cultivator of land (dummy)
|
0.161
|
0.170
|
0.179
|
1.42
|
Wealth score of
household
|
0.012
|
0.013
|
0.015
|
0.36
|
Punjab
|
0.049
|
0.069
|
0.091
|
0.41
|
Sindh
|
-0.274
|
-0.268
|
-0.261
|
-2.56
|
Northern Punjab
|
-0.300
|
-0.281
|
-0.261
|
-1.58
|
Central Punjab
|
-0.400
|
-0.380
|
-0.360
|
-2.55
|
Southern Punjab
|
-0.311
|
-0.292
|
-0.274
|
-1.79
|
Western Punjab
|
-0.091
|
-0.069
|
-0.048
|
-0.41
|
Karachi
|
0.162
|
0.165
|
0.169
|
1.58
|
Urban northern Punjab
|
0.463
|
0.470
|
0.476
|
2.65
|
Urban central Punjab
|
0.271
|
0.274
|
0.276
|
4.02
|
Urban western Punjab
|
-0.061
|
-0.055
|
-0.049
|
-0.35
|
Urban southern Punjab
|
-0.246
|
-0.240
|
-0.236
|
-1.92
|
Urban Sindh
|
0.198
|
0.202
|
0.206
|
2.26
|
Urban KP
|
-0.252
|
-0.245
|
-0.238
|
-1.54
|
Constant
|
7.803
|
7.817
|
7.828
|
24.13
|
Note: Number of obs. = 14,94,
prob. ≻ F = 0.0000. KP is the excluded province category and
“sharecropper“ is the excluded father’s occupation
Table A17.9: Regression of log total
yearly income on effort and circumstance variables for men aged 20–30
Variable
|
Min
|
OLS
|
Max
|
t-stat
|
Education (years)
|
0.037
|
0.039
|
0.041
|
2.55
|
Education squared
|
-0.003
|
-0.003
|
-0.003
|
-2.47
|
Average education of
parents
|
-0.014
|
-0.014
|
-0.014
|
-2.53
|
Ln (father’s total
annual income)
|
0.307
|
0.308
|
0.308
|
11.95
|
Father is employer or
self-employed (dummy)
|
0.444
|
0.455
|
0.465
|
4.05
|
Father is wage-paid
employee (dummy)
|
0.370
|
0.379
|
0.389
|
3.62
|
Father is
owner-cultivator of land (dummy)
|
0.194
|
0.204
|
0.214
|
1.7
|
Wealth score of
household
|
0.013
|
0.015
|
0.016
|
0.4
|
Man aged 20–30 works
in agriculture
|
0.127
|
0.129
|
0.131
|
2.2
|
Punjab
|
0.124
|
0.145
|
0.167
|
0.85
|
Sindh
|
-0.266
|
-0.258
|
-0.248
|
-2.45
|
Northern Punjab
|
-0.368
|
-0.351
|
-0.330
|
-1.95
|
Central Punjab
|
-0.460
|
-0.444
|
-0.424
|
-2.95
|
Southern Punjab
|
-0.382
|
-0.365
|
-0.345
|
-2.22
|
Western Punjab
|
-0.153
|
-0.135
|
-0.115
|
-0.79
|
Karachi
|
0.153
|
0.157
|
0.160
|
1.5
|
Urban northern Punjab
|
0.490
|
0.491
|
0.495
|
2.77
|
Urban central Punjab
|
0.273
|
0.274
|
0.276
|
4.03
|
Urban western Punjab
|
-0.059
|
-0.053
|
-0.047
|
-0.34
|
Urban southern Punjab
|
-0.211
|
-0.205
|
-0.201
|
-1.64
|
Urban Sindh
|
0.220
|
0.224
|
0.228
|
2.49
|
Urban KP
|
-0.259
|
-0.251
|
-0.242
|
-1.58
|
Constant
|
7.654
|
7.666
|
7.680
|
23.5
|
Note: Number of obs. = 1,494,
F(22, 1471) = 14.63, prob. ≻ F = 0.0000. KP is the
excluded province, and “sharecropper” is the excluded father’s occupation.
* The author is an associate professor and dean of
economics at the Lahore School of Economics.
** The author is an associate professor of economics at
the Lahore School of Economics.
*** The author is a research associate at the Centre for
Economics and Business, Lahore School of Economics.
**** The author is an assistant professor of economics at
the Lahore School of Economics.
[1] The dimensions of deprivation
(and cutoff) considered were education (less than primary completed, child not
enrolled), health and nutrition (an underweight woman in household, one death
under age 5), housing (mud house), electrification, safe drinking water, (no
covered sources), sanitation (lack of proper toilet), assets (none), livelihood
(household head unemployed or in elementary occupation), child status (not
enrolled in school), cooking fuel (dirty fuel used: wood, dung, or coal), land
ownership (< 2 acres agricultural or some nonagricultural land), and
consumption (less than the official poverty line).
[2] We could also use the “union” approach in which an
individual is considered poor if they lie below the poverty cutoff in all three
dimensions of poverty. However, there are two problems with this approach:
First, it might underestimate poverty; and second, the number of observations
in our sample that meet this criterion is severely limited.
Labels: Pakistan, Pakistan Economy, Pakistan: Moving the Economy Forward, Publications
posted by S A J Shirazi @ 1/10/2014 12:00:00 AM,
City Campus
104 - C, Gulberg III,
Lahore, Pakistan.
Phones: 92-42-35714936, 38474385
Fax: 92-42-36560905
Main Campus
Intersection Main Boulevard Phase VI
Burki Road
Lahore, Pakistan.
Phones: 36560935, 36560939