Counting the Poor
ABSTRACT
It has been noted in the literature that failure to meet the target set by government for reducing
the headcount ratio of poverty in Britain is partly due to the success of government policy in
generating economic growth, However there is another way of evaluating government policy. If
the purpose of poverty reduction is to reduce the incidence of social exclusion faced by
identifiable groups, then other aspects of policy have to be examined for internal consistency.
Measures of poverty which better capture the sense of exclusion are needed, and an exclusive
focus on the target headcount ratio is not helpful.
JEL CLASSIFICATION: D31, I32, I38
A. N. Angeriz*1
S. P. Chakravarty+2
*Welsh Economy and Labour Market Research (WELMERC)
Department Of Economics
Singleton Park
University of Wales
Swansea, SA2 8PP
UK
1
[email protected]
2
[email protected]
+
WELMERC and The Business School, University of Wales, Bangor, Gwynedd, LL57 2DG,
UK
Helpful discussions with Vani Borooah, Chris Galbraith, David Hojman and constant advice and
encouragement from John Treble are acknowledged. They are, however, not responsible for any
remaining errors. The work was partially funded by a grant from >>>> to John Treble.
August 2003
Counting the Poor
I. Introduction
This paper is concerned with identifying issues that require examination for evaluating
poverty reduction policies. An exclusive focus in policy evaluation on the target headcount
ratio either to criticise or to justify the policy is unwarranted. The underlying rationale for
poverty reduction policy need to be examined even if the goals of the policy are expressed in
the form of crude targets about the head count ratio of the poor. Thus the argument that the
"main reason why it has proved so hard for the Government to reduce the child poverty
count" is the "focus on relative rather than absolute income" (Brewer et al 2003) is not a
sufficient defence of government policy.1
In November 1998, the Statistical Programme Committee of the European Union agreed on a
poverty line based on the median income. In the EU countries, anyone having an income
below 60 per cent of the median income is defined to be poor. Thus the poverty datum line
for income changes over time. When governments set targets about reducing the percentage
of those who are poor, the targets are set by reference to the above changing line. In setting
these targets, no explicit indication may be given about how the median income is expected
to change over time. There may be no explicit statement about acceptable changes in income
inequality. No explicit target, for example, is set for the rate of change in the median income
with respect to the mean;2 and a degree of ambiguity is apparent about the expected changes
in income distribution in the context of which targets for the headcount ratio of poverty are
set. This ambiguity cannot be resolved by re-interpreting the targets, ex post, by reference to
some absolute poverty line that was not contemplated when the goals for poverty reduction
were announced.
1
If we use the particular absolute poverty line that is given in Brewer et al (2003), the head count ratio of
poverty would decrease faster than it would if we kept the government's definition of 60 per cent of the median
income. The potential success of government policy is shown up in a good light. But if we use a mean based
relative measure of the poverty line, the government policy of poverty reduction is shown to fail miserably. If
we are to evaluate government policy, we must stick to the government's own definition of the poverty line.
2
Poverty line in Britain is set at 60 per cent of median income. If we set it at 60 per cent of the median income,
reducing the number of the poor becomes a more difficult task . Mean income has increased by 25 per cent as
against a rise of 21 per cent for the median income between 1997 and 2002.
The question of the efficacy of government action is not a numbers game, which can be
settled by reference to a single number, the target head count ratio. Instead, the views of
social exclusion that might have informed the relevant aspects of policy need to be examined.
It is reasonable to surmise that the purpose of poverty reduction policy in Britain in the
context of the rhetoric about child poverty is to reduce the incidence of social exclusion by an
identifiable group, households containing children. The success of government in the pursuit
of the reduction of child poverty as a goal then has to be measured by the ability of policy to
reduce the incidence of social exclusion by this group. The potential for continuing success of
the policy has to be gauged by examining the characteristics of those who remain poor.
A measure of the incidence of exclusion experienced by a particular group -- for example,
households where children are present -- is the headcount ratio of poverty in this group.
Another measure might be the contribution to aggregate poverty by this particular group of
households. One might probe further into the experience of poverty in those households
where children are present.3 One might go even further and ask if those households with
children who have remained poor have different characteristics from those who have been
pulled out of poverty by policies that have been applied. The question concerns not just
whether a policy was effective, but whether it would remain effective.
The object is not necessarily to arrive at a summative judgement on policy, but it is to
examine the issues that might have informed policy and to press the argument for vigilance
about internal consistency of policies that are adopted. This approach to evaluating policy
entails a move away from the exclusive focus on the head count ratio of child poverty.
Instead, an examination of the trends of a more general index, the FGT(α) index, of poverty is
needed. The headcount ratio is a special case of this index. The rationale for the choice of
FGT(α) is discussed in greater detail in Section III below.
The paper is organised as follows. Section II discusses conceptual issues underlying measures
of poverty and examines the link between poverty and income distribution. Section III
describes how this link might be found in the distribution of income amongst the poor
3
The contribution of a group, say households containing children, to the total index of poverty has two aspects.
The contribution depends on both the proportion of the poor who belong to households containing children and
the experience of poverty (the poverty index) of those households who are poor and where children are present.
The FGT(α) index of poverty, for α >1, allows for above examinations to be conducted. We examine both
FGT(2) and FGT(3).
themselves by examining the FGT(α) index of poverty. A property of this index is that it can
be additively decomposed to calculate the share of contribution to poverty by mutually
exclusive and exhaustive groups -- for example, single parent households, households with
children and more than one adult, pensioner households, and others -- whose income lie
below the poverty datum line. Section IV describes the Family Resources Survey data and
Section V applies the ideas discussed in Section III to this data set for the years 1996/97 to
2001/02. An estimate of the taxes that are paid and benefits that are received as transfer
payments is also provided. Section VI concludes and suggests how the findings reported in
this paper may be further examined.
II. Measurement of Poverty
The definition of the poor as those whose income falls below some poverty datum line raises
the question of how the poverty line is to be delineated. There are two ways that this problem
is generally approached.
i.
The first is to define some minimum level by reference to physical requirements -- for
example, nutritional requirements -- for survival.
ii.
The second is "to endeavour to define the style of living which is generally shared or
approved in each society and find whether there is ... a point in the scale of the
distribution of resources below which as resources diminish families find it
particularly difficult to share in the customs activities and diets comprising their
society's style of living" (Townsend 1979).
Both the above approaches raise difficult conceptual issues. How do we define the physical
requirements for survival?4 How do we define the "style of living" approved by society? 5
4
The setting of the British supplementary benefits levels after the war was informed by the Beveridge Report of
1942. It provided an estimation of the subsistence level of income needed for survival. The problem with this
approach is that people can remain alive for quite a number of years even with incredibly low levels of nutrition.
The Physician Task Force on Hunger in America had issued a report in 1987, "Hunger Reaches Blue Collar
America", as a sequel to two previous annual studies on the spread of malnutrition in the US. (Christopher
Reed, "US doctors discover starvation on increase", The Guardian, Dec 7 1987, p.9) warning that millions of
homeless and jobless were suffering from an "epidemic" of malnutrition. Yet we have not heard of widespread
deaths from hunger in America. Malnutrition and hunger impacts on the quality of life in other ways than by
causing immediate death. For example, Atkinson (1983) points out that there is no unique level of food intake
defining the subsistence level of nutrition. Instead, physical efficiency declines in a number of ways due to
malnutrition of different kind.
5
Desai and Shah (1988) attempt to resolve this problem, raised in the debate between Piachaud (1981) and
Townsend (1981), by re-defining the “style of living approved by society” as the “modal behaviour”(p.518). By
doing so, “we make the sociological view of poverty empirically measurable”(Desai and Shah, loc cit). A
On reflection, it appears that the distinction between relative and absolute poverty is not as
sharp as it might seem at first sight. Relative prices are not independent of the distribution of
income. As more people acquire cars and buses run with empty seats, those who have to
depend on buses for transport have to carry a greater fraction of the fixed cost of bus service.
Changes in income distribution may lead to changes in relative prices. This, in turn, may lead
to a change in what and how much the poor can buy with a fixed sum of money. Thus the
subsistence level of income, often thought of as some absolute level, is itself a relative
concept.6
Another reason for introducing the distribution of income into poverty measures is that goods
in themselves do not provide utility; they empower an individual with the capabilities for
securing utility. For example, a bicycle is a good. Being able to go from A to B is a
capability. The capability derived from a good depends on the distribution of income. If
poverty is measured not in terms of the lack of ability to buy certain goods but in terms of the
lack of capability to do certain things, then relative deprivation in terms of goods could
sometimes result in absolute deprivation in terms of capabilities (Sen, 1983). A simple if
concocted example might be as follows: Suppose the purpose of acquiring a good (say a car)
is to enable one to visit friends. If most people do not have cars, friendship is generally made
amongst those who live within walking distance of each other. If, instead, most people have
different approach to understanding the minimum requirements for survival is to examine what people buy. If
the demand function is characterised by a linear expenditure system of demand functions, then the minimum
survival requirement may be defined as the sum of the constant terms as follows. Suppose that there are n goods
and xi amount of good i is demanded, when income is M. The utility function which leads to a linear demand
function is it is maximised subject to the budget constraint as follows:.
i n
i
U   xi  si 
i 1
i n
s.to :   pi  xi   M
i 1
The demand function resulting from the above maximisation exercise turns out to be a linear function:
i n


pi  xi  pi  si   i   M    pi  si  , and
i 1


in
p
i 1
i
 s i could now be regarded as the minimum survival
bundle of goods which are bought before additional amount of any good is purchased from the money that is left
over (Green 1976, pp 142-143). Theil and Clements (1987, p10) call si the “subsistence consumption” of the ith
commodity.
6
See Kenneth Arrow (1982): " As the still very large pool of unemployed workers in 1940 was absorbed into
industrial work by the rising demand for war goods, the demand for meat rose sharply. The meat supply did not
fall; it actually rose, but the rising demand still drove the prices up. From the viewpoint of the newly employed,
there was indeed a shortage."
cars, social customs might change and contact with neighbours might become less important.
Now the few who cannot afford cars could suffer a special disadvantage due to their inability
to afford cars. They cannot visit friends.7
Nowadays, most governments in OECD countries use a measure of poverty related to the
mean or the median income of the population as a whole. The United States remains an
exception, where the US Census Bureau continues to calculate an absolute measure
notwithstanding recommendations to the contrary by a panel of the American Academy of
Sciences. The methodology is informed by the ideas of Mollie Orshansky, who developed a
technique for calculating the subsistence budget by combining data on household 'choice'
(Household Consumption Survey) with some bureaucratically-defined level of minimum
food requirement (Orshansky, 1966).8 The British government’s position is that the absolute
standard -- the backbone of the Beveridge approach characterising much of post-war social
security policy -- has been superseded by "a notion of a relative minimum with all groups in
society having a share in the long run increase in national prosperity."9
There are basically two aspects of the distribution of income which enter into the calculation
of poverty indices. The first is the distribution of income in the population as a whole and the
second is the distribution of income amongst those who are poor. Governments in most
OECD countries do not over-concern themselves with changes in the right hand tail of the
income distribution in deciding on the poverty line. This line is set by reference to the median
rather than the mean income.
The distribution of income enters into measures of poverty also in another way. Agreement
about the poverty datum line only allows for the headcount ratio, the proportion amongst the
population of those who fall below the poverty line. If it is accepted that the measure must
reflect the difference in how poverty is experienced by those who fall much below the
poverty line compared to those who are just under that line, the headcount ratio needs to be
7
It should not be concluded, on the basis of the argument presented here about commodities versus capabilities
that the distinction between relative and absolute poverty is can be entirely erased. See the debate between Sen
and Townsend (Oxford Economic Papers, vol. 37, Dec 1985).
8
The US Census Bureau revises the datum line by mainly revising the largest basket of consumption
expenditure, food, for the poor. Thus the money needed to purchase the minimum survival level, as determined
by the Census Bureau, is proxied by food prices. The payments of state benefits to the poor are partly made in
food stamps, indicating the stronger grip of the farm lobby on government in the US. There are also other
institutional differences and it is beyond the scope of this note to evaluate the relative merits of the US and UK
policies.
9
HMG, 1985. p. 16.
enriched with a welfare function-based measure. The welfare function is needed to capture
the normative value that is placed by society on the distribution of income amongst the poor.
III. A Decomposable Index of Poverty
The FGT(α) index of poverty is a candidate for consideration. This index is chosen here also
because it is decomposable. Suppose that there are n number of units (say individuals in
households) in society of whom m have income below the poverty line Z. 10 Suppose that
these households are divided into k distinct (ie.: mutually exclusive but exhaustive)
subgroups. FGT(α) can be additively decomposed to isolate the experience of the depth of
poverty by different groups – eg. single parents couples with children pensioners living alone,
etc.
There are also other properties of FGT(α) that make it particularly suitable for
examining the question of social exclusion suffered by a particular group of the poor, for
example, households containing children.
The attraction of this index becomes apparent by following the literature on the development
of poverty indices. Once the poverty datum line is agreed, then the next question arises: how
is poverty to be measured? The simplest approach is to count the ratio of people whose
income falls below the poverty line. This measure is provided by the Head Count Ratio (H).
This ratio tells us something about the extent of poverty prevalent in that society. But to
develop a better understanding of the extent of poverty we need to know, also the distribution
of income of those who fall below that line. The simplest approach would be to construct an
index by adding up the feeling of deprivation, measured along a scale that makes possible
inter-personal comparison of those who are poor. The Poverty Income Gap, I, is a candidate
for this index. It attempts to capture the intensity of deprivation by adding up the amount of
income needed to be transferred to the poor in order to bring all of them up to the datum line
level of income (Beckerman and Clark 1982). In order to make the measure independent of
the number of the poor and the currency in which poverty is income is recorded, this index is
commonly normalized, producing the Poverty Income Gap Ratio, P.
10
The household is taken as the reference unit for discussion in this section. But the results reported in the paper
are for individuals. We use household equivalised BHC (Before Housing Cost) incomes provided by the HBAI
(Households Below Average Income) database, which is produced by the DWP (Department of Work and
Pensions). Complementary information is derived from the Family Resources Survey. The indices reported in
this paper, therefore, are based on individuals that are poor, unless otherwise indicated.
PI
mZ
i m
I 
i 1
(Z  yi )
mZ
where m denotes the number of units (say, households) enjoying an income below the datum
line, Z. The income for this set of units is the set {y1 ... ym}, where yi < Z for all values of i =
1,…,m.
The problem with P is that it does not satisfy the Transfer Axiom, which is a desirable
property of any poverty index. This axiom entails that "a pure transfer of income from a poor
[household] to any other [household] that is richer must increase the poverty measure"
(Foster et al 1984 p.762). We note that, as long as both the households are below the poverty
line of income to begin with and neither crosses that threshold due to the transfer, then P does
not increase if income is transferred from the poor to the less poor. This inadequacy is
addressed by Sen (1976), who provides an index that combines the head count ratio with the
Gini coefficient of distribution to obtain a measure of the depth of poverty. For large values
of m, the Sen index, S, is defined as follows:
S  H  P  1  P  G, where G is the Gini coefficient for the poor ; and it is defined for
income {y1 ... ym}.
A problem with the Sen index is that a transfer from a poor household to a less poor one
could decrease the poverty measure if the second household crossed the poverty datum line as
a consequence of that transfer. Whereas this property of index S might be tolerable if both the
households initially were close to each other in income —for instance, if they were hovering
just below the poverty line and a small amount of transfer were contemplated. This property
is especially questionable if the household which loses out suffers significantly as a result of
the transfer (Thon, 1983).
A partial remedy to these problems is offered by Foster, Greer and Thorbecke (1984). Their
index FGT(α) has the added advantage of being decomposable by mutually exclusive and
exhaustive groups.
i m
 Z  yi 
FGT ( )  1   
n i 1  Z 

 

where n is the total population, but the summation is only over the poor,. ie all those whose
income fall below the poverty line. The parameter α is a special feature of this index
encapsulating an implicit weight placed on inequality aversion. The FGT(α) index for α = 0
is the head count ratio but H. For α = 1 the index is H multiplied by P where P stands for
Poverty Income Gap Ratio. But the FGT index becomes more interesting for α >1 which we
consider here. When α >1 the FGT index introduces distributional consideration amongst the
poor (p. 762, Foster et al op. cit.). For example, when α =2:


FGT (2)  H P 2  1  P   C ,
2
where C is the coefficient of variation in the income of the poor {y1 ... ym}. Inequality
amongst the poor increases the experience of poverty, as it is measured here.
Therefore poverty, as measured by this index, shows an increase even if the head count ratio
has not changed, when there is an increase in the dispersion of income amongst the poor.
More precisely, when α > 1 the above index satisfies the transfer axiom described earlier. A
stronger condition, called the Transfer Sensitivity Axiom, is satisfied if α > 2. This axiom is
explained below.
Suppose that persons A, B, C, and D are all poor. B has an income greater than A by an
amount q. D has an income greater than C by the above amount q. Person C is richer than A,
and by implication D is richer than B. The transfer sensitivity axiom is satisfied if, for any set
of the poor {A, B, C, D} described as above, an increase in the poverty index due to a
transfer from A to B is greater than the increase recorded due to a transfer of the same
amount of income from C to D.11 An implication of this axiom is that an increase in the
proportion of the poor who are further down the poverty datum line implies, ceteris paribus,
an increase in a poverty index satisfying this axiom. The FGT(α) index is reported here for
three values of the parameter  to chart the changing nature of poverty from 1995 to 2002.
An implication of this property is that the index (for α > 1) increases when there is an increase in destitution,
but the mean income of the poor remains unaltered.
11
Poorer units are given greater weight in the above index and "a larger α gives greater
emphasis to the poorest poor" (Foster et al, op. cit.). The FGT(α) index can also be
interpreted as a measure of the depth of poverty. It can also be decomposed to isolate and
measure the depth of poverty experienced by different groups – e.g. single parents couples
with children pensioners living alone, etc.
Suppose that there are k distinct –i.e. mutually exclusive and exhaustive-- subgroups of the
sample population, each containing nj units. Therefore, its sum over all the categories
j k
comprise the total sample of n households:
n
j 1
j
n.
Out of a population of nj in the jth group mj fall below the poverty line, so the total number of
j k
units m whose incomes fall below the poverty line in the whole sample is:  m j  m .
j 1
Thus, the aggregate FGT(α) index can now be regarded as the weighted sum of the index
computed for each of the considered sub-groups.
j k
n 
FGT ( )    j   FGT j ( ) ,
n
j 1 
where the summation runs over j = 1... k and the index for the subgroup j is:

i m
 j  Z  yij  

 1  
FGT j ( )  
 

,

Z  
 nj  
i 1 


where mj being the number of poor households in the jth subgroup. The poverty line income
is Z and yij is the income of the ith household in the jth group whose income falls below Z.
The percentage of the contribution to the total aggregate poverty index of the j th group is,
thus:
nj 
 n   FGT j ( )

PCNT j ( )  
 100
FGT ( )
These measures when they are combined with additional information contained in the Family
Resources Survey allow us to engage in informed discussion beyond the confines of a single
index of headcount measure of child poverty about the changing nature of poverty.
IV. Family Resources Survey
We use the Family Resources Survey to calculate indices of poverty for the years 1995 to
2002. Poverty "is measured on the basis of household disposable income adjusted for
household size (or 'equivalised' income)" in common with practice in the literature. 12 The
survey data contained in the FRS allow for the computation of taxes paid by the poor and
benefits received back in transfer payment from government.
The Family Resources Survey consists on a set of cross-sections providing information about
incomes employment demographic and other individual circumstances of about 25.000
households in Britain.
In order to calculate individual incomes, taxes, and benefits, two sets of data are used -- the
first set is the Family Resources Survey (FRS) and data compiled by government for
Households Below Average Income (HBAI). The HBAI dataset reports variables computed
by the Department of Works and Pensions (DWP), using the FRS data. The income recipient
unit is the individual to whom the per capita net income of the household is assigned. 13 The
net household income is computed by aggregating all household members’ total incomes and
subtracting direct tax and national insurance contributions. These results, in turn, are then
netted off the contributions to pensions the maintenance expenses to support children not
living in the household and the council tax contributions. Finally, the per capita net income is
calculated by equivalising the household’s income by the members in the McClements Scale.
The procedure conforms to the methods in HBAI statistics reported by government.
12
Piachaud and Sutherland, 2002.
This approach to allocating income within households is also used in the report on Households Below
Average Income (HBAI), which is based on the Family Resources Survey.
13
Figure 4.1 depicts the composition of individuals living in different type of households to the
entire sample.
Figure 4.1: Demographic Composition of the Sample
Family Type Composition
Percentage
40
35
30
25
20
15
10
5
0
1997
1998
1999
2000
2001
2002
Year
pensioner couple
pensioner single
couple with children
couple without children
lone parent
single without children
There is a reasonably stable demographic composition of the population during the period
examined here. However there is a slight decrease in the share of the dominant groups. The
‘couple with children’ and ‘couple without children’ categories lose around 2 points each in
the whole interval and this loss is compensated by a sustained increase in the percentage of
‘single without children’ and more modest increases in ‘lone parents’ and ‘pensioner
couples’.
The average per capita weekly disposable income net of taxes and equivalised as described
above is given in Table 4.1. The population is grouped into the six mutually exclusive and
exhaustive categories.14
14
(DWP, 2003). The data are explained in an internet publication (www.dwp.gov.uk/asd/frs) by the Department
of Work and Pensions.
Table 4.1. Demographic family type groups as accounted for in the FRS
Group 0: All households
Group 1: Pensioner couple (Benefit units headed by a couple, where the Head of the Benefit Unit is over the
state pension age)
Group 2: Pensioner single (Benefit units headed by a single adult, who is over the state pension age).
Group 3: Couple with children (Benefit units headed by a couple, below the age of eligibility of state pensions,
with dependent children).
Group 4: Couple without children (Benefit units headed by a couple, below the age of eligibility of state
pensions, with no dependent children).
Group 5: Single with children (Benefit units headed by a single adult, below the age of eligibility of state
pensions, with dependent children).
Group 6: Single without children (Benefit units headed by a couple, below the age of eligibility of state
pensions, with no dependent children).
The average per capita weekly income as calculated above for those who live below 60 per
cent of the median income are given in Table 4.2. The poverty line is also indicated in that
table. These can be compared with the average income of the total population, as reported in
Table 4.2 below.
Table 4.2: Average Per Capita Weekly Disposable Income (£)
Group
1995
1996
1997
1998
1999
All households
278
288
307
318
334
Pensioner couple
239
241
269
274
284
Pensioner single
209
217
232
240
252
Couple with children
271
285
302
313
329
Couple, no children
353
366
391
405
425
Single with children
178
188
189
202
212
Single no children
291
298
321
334
349
Note: Income data are equivalised and and deflated within each year prices.
2000
349
296
267
345
443
216
372
2001
363
310
279
363
449
229
384
2002
384
327
287
383
485
244
406
Table 4.3: Average Per Capita Weekly Disposable Income of the Poor (£)
Group
1995
1996
1997
1998
1999
All households
104
107
117
117
121
Pensioner couple
115
119
128
129
134
Pensioner single
111
114
122
123
127
Couple with children
103
102
115
114
117
Couple, no children
89
99
100
105
112
Single with children
118
120
127
131
134
Single no children
98
100
109
108
111
The poverty datum line (60% of the median income of sample
population)
All households
138
143
154
157
162
Note: Income data are equivalised and deflated within each year prices.
2000
127
140
135
123
110
142
118
2001
127
144
138
125
109
140
114
2002
136
151
146
133
123
151
123
171
176
187
Note that the mean income of the sample population is given in Table 1 and the median
income for this population can be obtained by dividing the last row in Table 2 by a factor of
0.6. A comparison of the trend of the mean and median income suggests that the disparity
between the mean and median has widened in favour of the mean especially since 1997.
V. Poverty Indices
Three measures of poverty are shown in Table 5.1. The first index, the Head Count Ratio, is
also the FGT(0) index. The poverty indices are calculated in Table 5.2 for each of the six
mutually exclusive and exhaustive groups described above. The separate indices for the
above groups are weighted by their respective population shares to obtain their percentage
contribution to total poverty. These contributions are shown in Table 5.3.
Table 5.1: Poverty Indices
Year
N
Head Count (%)
FGT(2)
FGT(3)
1995
62394
17.6
2.064
1.404
1996
62037
16.7
2.073
1.449
1997
60618
18.3
1.938
1.236
1998
55865
18.2
2.226
1.501
1999
53973
18.0
2.128
1.411
2000
58898
17.7
2.239
1.529
2001
55729
17.0
2.482
1.777
2002
59392
17.0
2.354
1.669
Table 5.2: Poverty Indices Decomposed by Population Groups
Pensioner couple
Year
N
Head Count (%)
FGT(2)
FGT(3)
1995
1996
1997
1998
1999
2000
2001
2002
6248
6036
5825
5491
5417
5957
5761
5972
18.97
20.56
19.77
20.35
21.92
19.77
20.31
20.89
0.899
1.069
0.853
1.150
1.166
1.667
1.122
1.323
0.407
0.556
0.319
0.557
0.515
0.549
0.515
0.626
Pensioner single
Year
N
Head Count (%)
FGT(2)
FGT(3)
1995
1996
1997
1998
1999
2000
2001
2002
5086
5028
4777
4325
4306
4547
4420
4470
22.08
20.98
21.57
21.96
21.31
20.28
20.17
21.26
1.386
1.419
1.383
1.653
1.604
1.458
1.437
1.661
0.644
0.703
0.57
0.771
0.788
0.674
0.663
0.819
Head Count (%)
FGT(2)
FGT(3)
17.61
16.03
17.23
17.01
16.61
15.83
14.46
14.22
2.214
2.485
1.890
2.392
2.338
2.229
2.248
2.246
1.515
1.832
1.197
1.672
1.596
1.525
1.604
1.649
Head Count (%)
FGT(2)
FGT(3)
11.39
10.93
10.98
10.76
11.28
11.31
11.28
10.56
2.419
1.848
2.285
2.188
1.913
2.487
2.773
2.192
1.855
1.349
1.723
1.651
1.405
1.939
2.254
1.692
Couple with Children
Year
N
24799
1995
24769
1996
23928
1997
21999
1998
20797
1999
22699
2000
21195
2001
22556
2002
Couple without Children
Year
N
1995
1996
1997
1998
1999
2000
2001
2002
12046
11918
11592
10900
10524
11460
10872
11816
Lone Parent
Year
N
Head Count (%)
FGT(2)
FGT(3)
1995
1996
1997
1998
1999
2000
2001
2002
Single
Year
5528
5558
5820
5270
5192
5924
5631
5960
25.96
23.44
31.25
32.86
31.19
29.83
26.72
26.91
1.163
1.362
1.823
1.685
1.798
1.720
2.227
2.013
0.568
0.786
0.945
0.768
0.939
0.894
1.313
1.138
N
Head Count (%)
FGT(2)
FGT(3)
1995
1996
1997
1998
1999
2000
2001
2002
8687
8728
8676
7880
7737
8211
7850
8618
17.96
17.06
18.17
18.43
17.44
19.37
20.59
20.67
2.952
2.737
2.711
3.247
3.042
3.508
4.487
4.169
2.153
1.970
1.869
2.365
2.212
2.618
3.468
3.223
The head count ratio of individuals living in poor households belonging to Group 3, Couples
with Children, has gone down substantially in recent years, and certainly between 1997 and
2002. Although the head count ratio of poverty amongst individuals living in Group 5, Lone
Parent Families, has gone up, most of the children live in households containing more than
one adult. Thus, the number of children living in poverty has declined.
In fact, the head count ratio has declined faster for this group than it has for the population as
a whole. For example, this ratio has declined from 18.3 to only 17.0 between 1997 and 2002
for the population as a whole. But the decline for Group 3 has been faster, from 17.23 to
14.22, during the same period. A consequence of the above trends is that the proportion of the
poor who belong to Group 3 has declined from 40 per cent to 38 per cent between 1997 and
2002. Hence the contribution of this group to the aggregate head count ratio has declined
from 37.16 to 31.76 per cent between the years 1997 and 2002. Thus, households comprising
couples with children have been more successful in escaping poverty, if we measure poverty
by the head count ratio, FGT(0). Brewer et al (2003) concentrate on the FGT(0) measure, and
rightly point out that the decline in poverty would be even greater if the poverty datum line
were set at a lower level. But that is not the entire story. There is another way of reading the
data.
Those who have been left behind now suffer greater intensity of poverty than their
counterparts earlier. Whilst the percentage contributions to all the FGT measures of poverty
by Group 3 have declined, both FGT(2) and FGT(3) measures themselves have gone up. For
example, FGT(2) has gone up from 1.89 to 2.25 between 1997 and 2002. FGT(3) has
increased from 1.2 to 1.65 during the same period. As we noted earlier, "a larger α gives
greater emphasis to the poorest poor" (Foster, Greer, Thorbecke 1984 p.763). If we consider,
especially, the increase in the FGT(3) index, we find that a greater fraction of the poor living
in Group 3 households are further away from the contemporary poverty datum line in 2002
than was the case in 1997.15 What can definitely be said is that those children who live in
Group 3 households experience a greater heterogeneity in income than their counterparts in
1997. Since the FGT(2) and FGT(3) indices for Group 5, Lone Parent Households, have also
increased between 1997 and 2002, we can make a stronger statement. Whilst the number of
children living in poverty may have fallen, there is greater heterogeneity in the income
distribution amongst those who now live in poverty.
An investigation into the nature of the heterogeneity is required in order both to analyse the
effectiveness of past policies and to consider whether these policies need to be changed in
order to address the changing circumstances, problem. It may have been the case that the
previous policies addressed only those who were just below the poverty line. As Brewer et al
(2003) explain, children in the third and fourth deciles amongst the poor experienced much
higher income increases than any other subgroup amongst the poor. Further attempts to
reduce poverty may entail attention to those at the very bottom of the distribution.
VI. Conclusions
Policy evaluation is not a numbers game but numbers can provide insight into how well
different aspects of policy are joined up. In this paper we examine one set of numbers, the
FGT(α) index of poverty for different values of the poverty aversion parameter α, to examine
one aspect of the success of policy in reducing child poverty in Britain. We come to a less
15
It is difficult to be more precise in the interpretation of poverty indices between time periods because the
datum line is not fixed (Foster and Shorrocks 1988). Fixing the datum line required arbitrary assumption about
the nature of absolute poverty, an little will be gained by attempting to obtain a precise interpretation of poverty
measures by fixing the poverty line.
sanguine view of the efficacy of government policy than the one arrived at by examining the
head count ratio alone.
There is more to be done if we are to comprehend the task involved in policy evaluation.
Whilst additional investigation is outside the scope of this paper, salient features of some of
these other questions that might be raised about poverty are suggested below.
If economic growth is not accompanied by increased transfer payment to the poor, then the
reduction of social exclusion entails greater inclusion of the poor into the labour market by
providing employment and making the reward from employment match the increasing
prosperity in society. If the growth rate of net market income16 for the poor is greater than the
growth rate in the median income in society then ceteris paribus there may be a decrease in
poverty. A system of taxation, which redistributes between the poor, makes it more difficult
to address the problem of reduction in poverty through the provision of greater market
opportunities for the poor. At present (see Table 6.4), the percentage of income tax and
National Insurance collected from the poor is of the same order of magnitude as the transfer
payments that are made to the poor.17 It is necessary to investigate if this particular aspect of
fiscal policy in Britain, which has developed over the years as the increase in the income tax
threshold has lagged behind the increase in national prosperity, may be responsible for the
fact the reduction in poverty for one group appear to be accompanied by increases in poverty
for other groups. For example, the contribution to the poverty of households containing
children has declined between 1997 and 2002, but the contribution to poverty of single
people below retirement age has increased.18 The index of poverty has also increased from
this group during this period (Table 5.2). Before a view can be taken about the efficacy of
poverty reduction policy for households with children, it is necessary to establish whether this
reduction is obtained at the expense of other groups amongst the poor.
One particularly striking feature of Group 5, Lone Parents, that has been noted in the
literature is that the rate of return to work for single parents is low. This may partly be due to
16
Net of taxes and lost state benefits due to higher income.
In the 1950s, a family on an income at average male wage level did not pay any income tax. Now income tax
is payable at a level less than half of the average male wage rate. As tax thresholds have declined, more benefits
have become means tested. The problem of incentives for taking employment at the lower end of the income
group has become less tractable. The mitigation of the incentive problem by raising the threshold at which
income tax becomes payable appears to be rejected by governments fearful of increasing the marginal rate of
tax.
17
the fact that benefits while not at work are more generous for this group.19 However, the
benefits are more generous if child-care costs are ignored. The coherence of poverty
reduction policy cannot be judged without reference to aspects of taxation and childcare
policies that have an impact on the decision to work.
The FGT(α) index to give us an insight about poverty that might be missed if discussion
remains confined to the head count ratio. Consider Group 3, households containing more than
one adult and at least one child. The head count ratio has declined between 1997 and 2002,
but the FGT(α) index tells a more complex story. The characteristics of those who have
remained poor have changed (see Table 6.1), and policies which may have worked to reduce
the head count ratio in the past may not be sufficient to bring the poor amongst this group out
of poverty now.
Table 6.1: Composition of Group 3 Households: Percentage in Each Income Range
Income over 75% of z
Income below 75% of z and over 50%
Income below 50% of z and over 25%
Income below 25% of z
1997
64.5
24.5
6.0
5.0
2002
60.6
20.8
9.8
9.0
Change
3.9
3.7
-3.8
-4.0
A final point needs to be made about analysing government policy. Governments account for
a large share of the gross domestic product (GDP), and it is necessary to examine the share of
the expenditure that benefits the poor. The discussion above makes no distinction between
income as measured using survey-based data, generally on expenditure, as in the FRS and
income-based data that might be inferred from trends in the gross domestic product GDP.
The survey-based approach ignores the goods and services that are provided by government.
Government expenditure forms a large share of the GDP especially in the richer countries of
the West. Even in the UK where government's share of the GDP is low in comparison to the
richer countries in the EU it has fluctuated in the region of 40 per cent of GDP for almost
three decades now. If the poor are assumed to benefit from government expenditure on
services such as the defence of the realm law and order and education the poverty count
would change. For example, Xavier Sala-I-Martin (2002) estimates that the number of people
18
See Annex.
Dickens and Ellwood (2003). The direct monetary outlay out of earnings needed for child care may,
paradoxically, increase as the employment rate amongst non-single parents increase. The extent of child care
covered within families containing unemployed members is not adequately reflected in models examining the
incentives of return to work by single parents.
19
in the world living below US$1 a day has fallen to 286 million whilst the World Bank
estimates put the figure closer to 2.2 billion.20 The World Bank uses survey data but Sala-iMartin attributes to each household in the world a proportionate share of government
expenditure in the country. Bhalla (2003) provides a comparison of these different methods
of counting poverty. Economists generally do not place such inordinate faith in governments'
benevolent intention towards the poor and rely on survey data alone to measure poverty.
However the conceptual problem remains about how to apportion the benefits of government
expenditure amongst different segments of society to obtain a measure of poverty. This
expenditure is significant (see tables 6.2 and 6.3).
Table 6.2: Total government spending per person in ECU in 1995
EU
9089
Luxembourg
15377
France
10221
Denmark
14742
Netherlands
10000
Sweden
12849
Italy
6859
Austria
10853
UK
5908
Germany
10773
Ireland
5195
Finland
10633
Spain
4781
Belgium
10269
Portugal
3268
Source: EU News Release No 9097 - 19 December 1997
Table 6.3: Total Government Spending as % of GDP
1970
1995
1970
1995
EU
35.6
48.5
Sweden
57.5 f
65.1
Luxembourg
29.9
47.6
Denmark
41.7
58.2
Germany
37.1
47.7
Finland
36.8 c
56.8
Italy
29.9
48.2
Netherlands
41.3
51.1
Portugal
30.2 d
46.1 e
Austria
44.9 c
50.7
Spain a
31.5
46.0
Belgium
40.4
50.6
UK
34.4
41.2
France
36.4
50.5
Ireland
37.1
40.9 b
Source: EU News Release No 9097 - 19 December 1997
Key: a: 1980&1984; b: 1994; c: 1980; d: 1977; e: 1993; f: 1989
20
Chen and Ravallion (2002).
The poor pay a large amount in tax in Britain certainly compared to the benefits they get.
(Check Table 6.4). For example, the amount of direct taxes and national insurance
contributions made by the poor is about 20 per cent of the amount of benefit income received
by them. The figures are more striking if the poor are divided into two groups, Group A and
Group B. Group A comprises those whose income lies between the poverty line and 75 per
cent of the poverty line. Group B comprise the rest. They are poorer. The ratio of N.I and
direct taxes paid to the Exchequer and the benefits received from government by members of
Group B, the poorer of the two groups, varies between 40 to 47 per cent. Suppose that the
taxes the poor pay are returned as services to them. Further suppose that the poor benefit
additionally from the contribution made by the non-poor towards the provision of services by
government then the measures of poverty would substantially decline. To evaluate
government policy towards the poor it is necessary to define the concept of what constitutes
pro-poor government expenditure. Then we can measure the extent to which this expenditure
is pro-poor.21 These are ideas that need to be looked at if we are to understand the role of
government policy in poverty alleviation.
Table 6.4: Direct Taxes and National Insurance Contributions as a Proportion of Benefits
[(Direct Taxes+NI)/Benefits]*100
1995
1996
1997
1998
1999
2000
2001
2002
All below the poverty line
20.8
27.2
21.2
23.6
19.9
18.4
20.8
21.1
Group A
14.7
16.0
15.9
17.6
12.2
11.0
12.0
11.7
Group B
42.9
66.8
38.3
41.2
40.3
38.9
43.7
47.5
In this paper we address a limited question and focus on the FGT(α) index to give us an
insight into poverty that might be missed if discussion remains confined to where the poverty
datum line should be set in computing the head count ratio of child poverty. The object is not
to pronounce on the success or failure of policy, but to point to aspects of policy that needs to
be examined to see if the purpose and outcome of policy are internally consistent.
21
Since 1999, the IMF began requiring poverty impact of government expenditure as a condition for assistance
to low income developing countries. Impact assessment methodologies are still at their infancy.
References:
K. J. Arrow (1982), "Why People Go Hungry", New York Review of Books, Vol. 29,
No. 12.
A. B. Atkinson (1983), The Economics of Inequality. 2nd edition Oxford: OUP.
M. Brewer, T. Clark and A. Goodman (2003) “What happened to Child Poverty in the
UK under Labour’s First Term?”, The Economic Journal, Vol. 113, No.488, pp.
F219-F239.
W. Beckerman and S. Clark (1982), Poverty and Social Security in Britain Since
1961, Oxford: Clarendon Press.
S. S. Bhalla (2003), "Crying Wolf on Poverty", Economic and Political Weekly, July
5th, pp. 2843-2856.
B. Bradbury and M. Jntti (1999), "Child Poverty Across Industrialised Nations",
UNICEF: Innocenti Occasional Papers Economic and Social Policy Series, No 71.
R. Dickens and D.T. Ellwood (2003), “Child Poverty in Britain and the United
States”, The Economic Journal, Vol. 113, No.488, pp. F240-F257.
H. A. John Green (1976), Consumer Theory. Basingstoke: Macmillan.
S. Chen and M. Ravallion (2002), "How Did the World's Poorest Fare in the 1990s?"
World Bank Working Paper.
S. Chen and M. Ravallion (2001), "How Did the World's Poorest Fare in the 1990s?"
Review of Income & Wealth, Vol. 47, Issue 3, pp. 283-301.
Department of Work and Pensions (2003), Family Resources Survey Great Britain
2001-02, www.dwp.gov.ac.uk/asd/frs
M. J. Desai and A. Shah (1988), "An Econometric Approach to the Measurement of
Poverty", Oxford Economic Papers, vol. 40, pp. 505-522
J. E. Foster J. Greer and E. Thorbecke (1984), "A Class of Decomposable Poverty
Measures" Econometrica vol 52, pp 761-66
J. E. Foster and A. Shorrocks (1988), "Poverty Orderings", Econometrica, vol 56, No
1, pp. 173-177.
HMG (1985), Green Paper on the Reform of Social Security, June 1985 vol. 3.
R. Ross Mackay and J. Williams (2003), "Beginning to Think About Need: Public
Spending in the Regions", Bangor: typescript.
M. Orshansky (1966), "Counting the Poor: Another Look at the Poverty Profile" in L.
A. Ferman et al (eds) Poverty in America, Ann Arbor: University of Michigan Press
1966)
D. Piachaud (1981), “Peter Townsend and the Holy Grail”, New society, 10
September.
X. Sala-i-Martin (2002), “The Disturbing ‘Rise’ in Global Inequality”, NBER
A. K. Sen (1983), "Poor Relatively Speaking" Oxford Economic Papers, Vol. 35,
July 1983, pp. 153-69; and reprinted in Amartya Sen, Resources Values and
Development Oxford: Blackwell, 1984.
A.K. Sen (1976), "Poverty: An Ordinal Approach to Measurement", Econometrica
Vol. 44 , No. 2, pp. 219-231.
T. N. Srinivasan (2003), "Globalization and the Poor", Paper presented at the
Conference on the Wealth of Nations Erasmus University at Rotterdam April 9-11.
H. Sutherland and D. Piachaud (2002), "Changing Poverty Post-1997", London
School of Economics CASE paper 63.
H. Theil and K. W. Clements (1987), Applied Demand Analysis, Cambridge, Mass:
Ballinger.
D. Thon (1983), "A Note on a Troublesome Axiom for Poverty Indices", Economic
Journal, Vol. 93, pp. 199-200.
P. Townsend (1979), Poverty in the United Kingdom, Harmondsworth: Penguin.
P. Townsend (1981), “Reply to Piachaud”, New Statesman, 17 September.
P. Townsend (1985), “A Sociological Approach to the measurement of Poverty – A
Rejoinder to Professor Amartya Sen”, Oxford Economic Papers, Vol. 37, No. 4, pp.
659-668.
ANNEX
Contribution to Poverty Indices Decomposed by Population Groups
(in percentages)
Pensioner couple
Year
N
6248
1995
6036
1996
5825
1997
5491
1998
5417
1999
5957
2000
5761
2001
5972
2002
Pensioner single
Year
N
Head Count (%)
10.8
12.0
10.4
11.0
12.2
11.3
12.4
12.4
FGT(2)
4.36
5.02
4.23
5.08
5.50
5.27
4.67
5.65
FGT(3)
2.90
3.73
2.48
3.65
3.67
3.63
3.00
3.77
Head Count (%)
FGT(2)
FGT(3)
5.47
5.55
5.62
5.75
6.02
5.14
4.59
5.31
3.74
3.93
3.64
3.98
4.46
3.48
2.96
3.69
FGT(2)
42.64
47.85
38.49
42.32
42.34
38.37
34.45
36.24
FGT(3)
42.90
50.47
38.25
43.87
43.59
38.45
34.33
37.52
FGT(2)
22.63
17.12
22.55
19.18
17.53
21.62
21.80
18.52
FGT(3)
25.51
17.89
26.66
21.46
19.41
24.68
24.75
20.16
5086
10.2
1995
5028
10.2
1996
4777
9.3
1997
4325
9.3
1998
4306
9.4
1999
4547
9
2000
4420
9.4
2001
4470
9.4
2002
Couple with Children
Year
N
Head Count (%)
24799
39.8
1995
24769
38.3
1996
23928
37.2
1997
21999
36.8
1998
20797
35.6
1999
22699
34.5
2000
21195
32.4
2001
22556
31.8
2002
Couple without Children
Year
N
Head Count (%)
12046
12.5
1995
11918
12.6
1996
11592
11.5
1997
10900
11.5
1998
10524
12.2
1999
11460
12.4
2000
10872
12.9
2001
11816
12.4
2002
Lone Parent
Year
1995
1996
1997
1998
1999
2000
2001
2002
Single
Year
1995
1996
1997
1998
1999
2000
2001
2002
N
5528
5558
5820
5270
5192
5924
5631
5960
Head Count (%)
13.1
12.6
16.4
17.0
16.7
17.0
15.9
15.9
FGT(2)
4.99
5.89
9.03
7.14
8.13
7.73
9.07
8.58
FGT(3)
3.59
4.86
7.34
4.83
6.40
5.88
7.47
6.84
N
8687
8728
8676
7880
7737
8211
7850
8618
Head Count (%)
14.2
14.4
14.2
14.3
13.9
15.3
17.1
17.7
FGT(2)
19.91
18.57
20.02
20.58
20.49
21.84
25.46
25.70
FGT(3)
21.36
19.13
21.64
22.22
22.48
23.87
27.49
28.02