The Human Development Index:
"Handle with Care"
ALLEN C. KELLEY
ECONOMISTS AND POLICYMAKERS have long questioned the emphasis on
growth of gross national product per capita (GNP/N) as the singular goal
and measure of national development. GNP/N fails to capture the distribution
of the benefits of economic progress-in particular, the number and condition
of persons living in poverty; and it abstracts from a multitude of specific
factors that relate directly to human welfare-for example, the benefits of
health, education, and political and social freedoms.
The arguments advanced in favor of using the GNP/N goal and measure
have been its simplicity, the assertion that it represents a reasonable proxy
for several dimensions of human welfare, and most importantly, the absence
of an alternative single measure that better approximates human develop-
ment in the aggregate. Such an alternative measure has recently been offered
by the United Nations Development Programme (UNDP) in its Human De-
velopment Report 1990 (HDR) (New York: Oxford University Press), which
unveils a "Human Development Index" (HDI). The UNDP appropriately
recognizes the considerable difficulty in conceptualizing and measuring the
somewhat nebulous condition of human development, and thus notes that
the HDI "opens the debate" (p. iii) that will result in refinements of both
the analytical framework and the empirical inputs over time.
I propose to participate in that debate. I will challenge the usefulness
of the conceptual framework of "human development" as specifically rep-
resented in the HDI, illustrate the sensitivity of this measure to plausible
refinements, and argue that it offers only limited insights beyond those ob-
tained by small modifications to simple measures of economic output. Until
the conceptual underpinnings of the HDI are more firmly established, analysts
and policymakers are better served by using much simpler measures and
POPULATION AND DEVELOPMENT REVIEW 17, NO. 2 (JUNE 1991) 315
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
316 THE HUMAN DEVELOPMENT INDEX
methods for evaluating human development. In the meantime, it is important
that the nature and pitfalls of the HDI be understood. I contend that the HDI
is a tool of limited value, and my goal is to stamp this new tool with a
precautionary label: "Handle with Care."
Definition, measurement, and assessment
of human development
Definition
Human development is defined in the HDR as "a process of enlarging people's
choices" (p. 1). According to the HDR, this is accomplished most funda-
mentally by living a long and healthy life, being educated, and having a
decent standard of living; and it is both augmented and facilitated by political
freedom, guaranteed human rights, and personal self-respect. While this
represents a defendable list of components (or indicators) of individual wel-
fare, the key to appraising this list is whether the indicators can be aggregated
in a way that permits assessment, and that exposes the impacts of policy
manipulation. Such aggregation forces the specification of tradeoffs between
the various indicators of human development. It also results in an index that
is potentially capable of answering the central questions: when has human
development occurred; to what level or extent; and what has caused it?
Measurement
To this end, the HDR unveils a "Human Development Index," a composite
of life expectancy at birth, adult literacy, and real gross domestic product
per capita (GDP/N). Because the HDI is the centerpiece of the HDR's con-
ceptual framework, it is important that this index be understood and assessed.
In particular, one seeks to know how the HDI differs from the usual measure
of development (per capita GNP), what are its strengths and weaknesses,
and in what ways it can be improved.
The most commonly used indexes of welfare and poverty are based on
an absolute standard defined in terms of a specified bundle of goods and
services, often tied to minimum caloric needs. (For example, the World De-
velopment Report 1990 [WDR-1990] uses US$370 GDP/N in 1985 purchasing-
power-parity dollars. This represents the upper range of the poverty lines in
a number of low-income countries. In contrast, the HDI is based on a relative
standard and represents an assessment of the extent to which each country
is successful in attaining the maximum value (described as "adequate" or
"desirable") within a specified range of values for each human development
indicator. (A calculation of the extent to which a country falls short of the
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
ALLEN C.KELLEY 317
maximum value yields a "deprivation index.") In the case of life expectancy
and literacy, this range is simply the lowest and highest values for the 130
countries in the UN data sample. For life expectancy, the range of values is
36 years (the difference between 78 years for Japan and 42 years for Af-
ghanistan, Ethiopia, and Sierra Leone in 1987); for literacy, the range is 88
percent (the difference between 100 percent in several developed countries
and 12 percent in Somalia in 1985).
The range in the GDP/N indicator is $4,641 (the difference between
Zaire's GDP/N of $220 and $4,861, the average of the poverty-line measures
in several developed countries, denoted as the "North minimum purchasing
power"). The GDP/N measure is constructed by using a "purchasing power
parity index" to better reflect consumer buying power across countries; and
it is transformed into logarithms to provide considerably less weight to im-
provements in spending power at higher levels of income. (Hereafter
GDP/N is taken to be in purchasing-power-parity dollars.) This mathematical
transformation captures the widely embraced and important economic hy-
pothesis of diminishing marginal utility of income-the notion that the "hap-
piness" or "utility" gained from an additional dollar of income is greater at
lower than at higher levels of income. Both adjustments are analytically
appropriate, although many of the purchasing-power-parity index figures
are still of dubious quality, and the specific logarithmic transformation is
necessarily arbitrary.'
The calculation of a country's HDI starts with the computation of a
"deprivation index" for each indicator of human development. These cal-
culations are best explained with an example. Kenya's life expectancy dep-
rivation index is .53 = (78 - 59)/(78 - 42). This value indicates that Kenya,
with a life expectancy of 59, is approximately half-way in the life expectancy
range between the worst (42) and best performance (78) worldwide. Similar
indexes for adult literacy and GDP/N are compiled. A "composite deprivation
index" of the three indicators is obtained by simple averaging. (Technically,
GDP/N deprivation is zero for countries whose GDP/N exceeds the assumed
poverty line.) Since the lowest attainable value of zero represents no depri-
vation (and correspondingly the highest level of human development), the
HDI is obtained by subtracting the composite deprivation index from unity.
Japan gets highest honors with an HDI value of .996; Niger is lowest
with a value of .116; and the United States scores .961. Some 44 countries
with a combined population of 1.47 billion are classified as having "low"
human development (HDI values less than .5); 40 countries with a combined
population of 2.06 billion are classified as having "medium" human devel-
opment (HDI values between .5 and .79); and 46 countries with a combined
population of 1.47 billion are classified as having "high" human development
(.8 or higher). No defense is offered for these specific cutoffs, although they
provide roughly equal numbers of countries in each category.
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
318 THE HUMAN DEVELOPMENT INDEX
Assessing the meaning and usefulness of the HDI can be made on two
grounds: the extent to which it is based on an appropriate conceptual frame-
work and is properly measured, and the extent to which it provides new or
modified insights into development.
Assessment I: The conceptual and measurement
framework
An examination of some of the underlying assumptions and features of the
HDI at once illustrates the difficulty in constructing such an index, and the
caution that must be exercised in using it.
First, the HDI is based on a country's position along a range of maximum
and minimum values for each indicator of human development. The specific
weight of that indicator in the HDI can be sensitive to the choice of these
endpoints, for which the HDR tends to select exceptional values. For example,
an adequate/desired life expectancy is chosen to be 78, a value attained by
only one country (Japan). A less exceptional, but reasonable value might
have been 73 (the average of the developed countries in 1987), or even 71,
the value for these countries in 1975. (Adopting the latter value would raise
China's HDI from .72 to .78, placing it close to the cutoff for "high" human
development.) It would have been useful had the HDR included tests of the
sensitivity of HDI rankings to alternative endpoints.
In fact, the HDI turns out to be quite sensitive to the choice of maximum
life expectancy. As an experiment, I set the adequate/desirable life expectancy
at age 73 and evaluated the impact. This raised 22 countries from "low" to
"imedium" human development and raised another ten countries from "me-
dium" to "high" human development.
Second, as implemented, the HDI effectively assumes that little or no
progress in human development can be made by the developed countries.
They all are close to the maximum values in literacy and life expectancy,
and most exceed the "North minimum purchasing power." The HDI values
for the developed countries vary from .96 to 1.00. As a result, the HDI has
operational meaning only for the developing countries. (The HDI values for
those countries vary from .12 to .95.) This is a disappointing feature that
might be overcome in the future by the use of improved indicators of human
development. For example, surely the attainment of education beyond the
literacy level has a favorable impact on "enlarging people's choices." Possibly
years of educational attainment, or a proxy thereof (e.g., enrollment rates),
would better capture the benefits of expanded education on human choice.2
Third, the various indicators of human development are given equal
weight. While a priori it is difficult to justify any set of weights, testing the
sensitivity of the HDI to alternative weights would have been useful. The
greatest problem occurs with the relatively low weight accorded to GDP/N
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
ALLEN C.KELLEY 319
in moderate- to high-income countries since the variable measured by this
particular indicator (income) can be used to acquire and/or produce either
of the other two indicators (improved health or education). Possibly a weight
roughly reflecting the acquisition/production-transformation would have
been appropriate under the premise that in some countries individuals may
well have elected to use their income to expand choices in ways that do not
result in, say, improved education or health. Indeed, it might be argued that
the capacity to choose among many dimensions of human development
accorded by expanded income in particular merits giving a relatively higher
weight to this indicator.
Alternatively, the production-transformation between income per cap-
ita and other human development indicators may be nonlinear, and thus
might justify unequal or even variable weights by income level. For example,
relatively small expenditures on immunization of children likely convey
much greater impact on life expectancy at birth (and thus human devel-
opment a' la HDI) than relatively large expenditures on life-support systems
to sustain individuals in old age. The HDR neither justifies the use of loga-
rithms to transform income per capita on these grounds, nor provides any
empirical evidence to reveal the nature of this production-transformation.
Indeed, the HDR effectively treats the three indicators as independent of one
another. Given the likelihood of strong dependence as noted in the above
example, unintended biases (e.g., unintended weights accorded to the three
indicators) can be introduced that merit detailed scrutiny in subsequent
revisions of the HDI.
Fourth, there is asymmetry in the choice of the GDP/N endpoint value
(the developed countries' poverty line) and the endpoints for literacy and
life expectancy (health), taken as the highest country value worldwide. Com-
putationally, this implicitly provides greater weight to literacy and longevity
than to GDP/N. If, in fact, the logarithmic transformation of GDP/N adequately
captures diminishing marginal utility of income, then why not use the max-
imum value for this indicator, or alternatively, the "average" GDP/N of the
high-income countries? Even this procedure does not equally weight the
three indicators in a welfare sense because there is plausibly diminishing
marginal utility to health and education expenditures as well. A consideration
of the complexity of these fundamental conceptual issues, and the arbitrar-
iness of any set of weights, likely accounts for the lack of any defense offered
in the HDR for the equal-weight assumption.
To test the sensitivity of the HDI to the specific logarithmic transfor-
mation of GDP/N, I recalculated the HDI using $12,952 as the adequate/
desirable endpoint value, which corresponds to the average of the developed
countries' GDP/N. This twofold increase in income per capita had a much
smaller impact than expected. For example, only four countries were re-
classified from "medium" to "low"; and only five countries were reclassified
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
320 THE HUMAN DEVELOPMENT INDEX
from "high" to "medium." The HDI does not appear to be particularly
sensitive to the poverty-line cutoff, a disquieting finding. It is difficult to
believe that such a large increase in income per capita, from $4,861 to
$12,952, would have only a small impact on "enlarging people's choices."
Fifth, overall human development is advanced not only by improve-
ments in the various human development indicators, but also by their more
equal distribution. This is a common property of welfare indexes based on
relative criteria. An example of this property is useful. Sierra Leone's life
expectancy at birth increased significantly between 1960 and 1987-by some
31 percent, from 32 to 42 years. Surprisingly, no human development would
be attributed to this accomplishment by the HDI since Sierra Leone's relative
position did not change. Apparently, living much longer is not sufficient to
register an improvement in human development.
These examples expose vividly the strong weight accorded to the dis-
tribution and the relative ranking of countries by human development in-
dicators in the HDI. This specific weighting requires detailed justification,
about which the HDI is substantially silent. Such considerations led the World
Bank (WDR-1990) to select an absolute measure of "poverty," since
[p]overty is not the same as inequality. . . At maximum inequality one
person has everything and, clearly, poverty is high. But minimum inequality
(where all are equal) is possible with zero poverty (where no one is poor) as
well as with maximum poverty (where all are poor). (p. 26)
And such considerations have led the International Labour Organization to
emphasize an absolute measure of "basic human needs." Of course, even
using an absolute measure does not sidestep the conceptual problem of
reconciling the impact of improvements in absolute and relative changes on
welfare unless relative changes do not count at all-an assumption repre-
senting the opposite extreme from the HDI. However, since the concept of
relative deprivation-the essence of the HDR's framework-constitutes the
distinguishing feature of the HDR vis-'a-vis the poverty/human-development
assessments of the World Bank, the International Labour Organization, and
most national governments, this concept merits much more detailed and
convincing defense than is provided.
Other qualifications of the HDI can be made, but the above should be
sufficient to recommend caution in using the present index. The HDR iden-
tifies several areas in which the index can be improved, most notably by
including indicators of political freedom and human rights, and by taking
into account the distribution of indicators of human development within
countries. Based on additional data collection, the UNDP plans to introduce
modified indicators of human deprivation in subsequent issues of the HDR.
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
ALLEN C.KELLEY 321
Hopefully there will be improvements in other dimensions of the index as
well-for example, a better proxy for educational attainments than adult
literacy (a rather crude measure) and a better proxy for health than life
expectancy at birth. While these modifications can be useful, they will not
resolve the thorny conceptual difficulties relating to the selection of appro-
priate adequate/desirable standards for each index, the combining of indi-
cators with weights that are both interpretable and defendable in a meaningful
welfare sense, and especially the assessment of human development based
on the concept of relative deprivation.
Having said this, one should not downplay the bold attempt of the
UNDP, with its proposed HDI, to broaden our view of development beyond
the usual GNP/N measure. However, given the high costs and monumental
difficulties in devising an improved measure of human development, it would
be appropriate at this stage to assess the benefits of such an effort. In particular,
what new and/or modified insights have we obtained from use of the HDI?
Assessment II: New or modified insights
The main and likely the most robust insight that follows from the HDI is
displayed here as Figure 1 (based on Figure 1.2 in the HDR), which compares
a ranking of countries by the HDI and the usual measure of economic de-
velopment-gross national product per capita (GNP/N), converted into US
dollars using exchange rates. As noted by the UNDP, substantial disparity
FIGURE 1 Ranking of countries by HDI and GNP/N
0.9
0.4-
20
-18
0.8 - - 16
0.7- HDI - 14 m
GNP/N
0.6 - - 12 z
0.5 1 01l
8
0.3 -4
0.2 -2
0.1 I - 0
0 10 20 30 40 50 60 70 80 90 100 110 120 130
Country rank
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
6
322 THE HUMAN DEVELOPMENT INDEX
exists between the two measures, especially at low rankings where GNP/N
rises only slowly while the HDI traces out a strong upward trend. This leads
to the major conclusion that "There is no automatic link between the level
of per capita income in a country and the level of its human development"
(p. 15).
Rephrasing this conclusion in a dynamic context, and based on other
data presented in the report, we may say that human development has
progressed much more rapidly in the developing countries than economic
development, as proxied by the growth of GNP/N. While this is a useful
insight, it is not particularly novel, at least to specialists in the field. Indeed,
it is a stylized fact of recent history (sunumarized in the HDR, but standard
in development textbooks as well) that improvements in education and life
expectancy have far outstripped the growth of GNP/N in the third world,
and that improvements in these social indicators have proceeded faster and
at an earlier stage of development than in the now-developed world.
Possibly more interesting to our appraisal of the HDI is the frequent
allegation that the "trickle down" of GNP/N growth to the masses of the
third world has been disappointing. Thus, not only is the preoccupation with
GNP/N as a measure of economic development inappropriate, but we also
need a broader measure of human development (like the HDI?) to rectify
this deficiency. Ironically, while the HDI-GNP/N comparisons seem to cast
doubt on GNP/N as a measure of human development, it now appears that
trickle down is dramatically understated by GNP/N if the HDI is taken as the
norm. Improvement in human development, according to the HDI, is much
better than GNP/N growth would suggest. Or is it?
Another way of evaluating the contribution of the HDI to measuring
welfare and human development is to compare it with the logarithm of
GNP/N. After all, the hypothesis of diminishing marginal utility of income
is widely accepted by all economists. While nonspecialists might well use
GNP/N as a linear approximation to welfare, teachers of development and
development textbooks would caution students against interpreting the "wel-
fare" or "happiness" of the Swiss, with a 1987 GNP/N of $21,330, as being
165 times that of the average Ethiopian, with a GNP/N of $130.
In Figure 2, therefore, I compare HDI with the logarithm of GNP/N.
The notable disparity between HDI and GNP/N, as highlighted in the HDR,
vanishes. Indeed, log GNP/N appears to represent a reasonable overall ap-
proximation to the HDI.3 Of course, there are conspicuous exceptions for
specific countries-mainly the oil-exporters with high GNP/N whose HDI is
overstated, some socialist countries with high literacy and/or low mortality
rates whose HDI is understated, and a few others. However, in the devel-
opment textbooks, and in most research based on cross-country comparisons,
these countries are almost always viewed as outliers or special cases. The
HDI mainly confirms this classification.
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
ALLEN C.KELLEY 323
FIGURE 2 Ranking of countries by HDI and log GNP/N
4.2
4
- 0.6 t / Log GNP/N 3.4
3.2Z
0.9
0.8 - HDI 3.8
0.7 -3.6
0.5 -3
0.4 -2.8
0.3-
0.2 2.2
0 .1 r i I | i a l l I I | X 2
0.0 10 20 30 40 50 60 70 80 90 100 110 120 130
Country rank
The bottom line
The HDI contributes only modestly to providing new or modified insights
compared with the usual measure of economic development (GNP/N) if the
latter is transformed into logarithms to capture, for welfare purposes, the
widely embraced concept of diminishing marginal utility of income. (The
HDI does indeed underscore the importance of this hypothesis when making
welfare assessments.) Moreover, most of the "exceptional" countries (e.g.,
the OPEC states) are usually treated as outliers without the use of the HDI.
As a result, if the HDI is to make a major contribution, it must do so by
providing a useful quantitative assessment of absolute changes in human
development. However, considerable caution must be exercised in inter-
preting such absolute changes, given the questionable conceptual foundation
of a relative welfare index, the arbitrary assumptions used to establish the
endpoints of the indicators and the specific manner in which they are proxied,
and the lack of an analytical basis for selecting specific weights to combine
the indicators. Additionally, the present index provides only limited infor-
mation on the distribution of indicators within countries and fails to include
systematic information on, and an analysis of, political freedoms and human
rights.4 Plausibly it is in these latter two areas where notable improvements
over the existing and very simple measures of human development (e.g., log
GDP/N) are likely to be made. I therefore conclude that at the present time,5
the human development index contributes little to the assessment of human
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
2.6
2.4
324 THE HUMAN DEVELOPMENT INDEX
development, and may well distract from the newly launched and otherwise
valuable Human Development Reports, which represent a useful addition to
the growing list of annual monitoring exercises on the status of world de-
velopment.
Notes
1 Using the purchasing-power-parity in-
dex rather than official exchange rates can
have a sizable impact on the ranking of specific
countries. For example, based on official ex-
change rates, GNP/N is $130 and $150 (in
1987) for Zaire and Ethiopia, respectively;
based on the purchasing-power-parity index,
GDP/N is $454 and $220, respectively.
2 This is recognized by the World Bank
in its measurements of "poverty." For ex-
ample, WDR-1990 (p. 29) uses enrollment
rates, life expectancy, and under-five mortal-
ity as social indicators qualifying the simple
poverty-line measures based on GDP/N in
1986 purchasing-power-parity dollars.
3 To evaluate this assessment quantita-
tively, HDI was taken as a linear function of
GNP/N, and log GNP/N, for the sample of 130
countries, yielding r2s of .40 and .71, respec-
tively. The log transform yields a substantially
improved and reasonable statistical fit.
4 The UNDP plans to incorporate these
additions in future modifications of the HDI.
5 As an interim measure, I recommend
using either (1) GDP/N in purchasing-power-
parity dollars, with countries classified into
high to low human development levels by
income cutoffs that provide less weight to
higher incomes, or (2) GDP/N in purchasing-
power-parity dollars, transformed into loga-
rithms. This single measure should then be
qualified by measures of intra-country abso-
lute poverty, plus a few selected indicators
relating to health, education, life expectancy,
political freedom, and the like. The World
Bank uses the first procedure in its analysis of
poverty in the WDR-1990.
This content downloaded from 210.212.129.125 on Tue, 23 Jan 2018 10:16:49 UTC
All use subject to http://about.jstor.org/terms
ليست هناك تعليقات:
إرسال تعليق