STATISTICAL METHODOLOGY AND THE USE OF ECONOMIC AND SOCIAL INDICATORS IN THE ESTIMATION OF PER CAPITA GDP LEVELS FOR DEVELOPING COUNTRIES*
| Published date | 01 November 1980 |
| Author | C. L. GILBERT |
| Date | 01 November 1980 |
| DOI | http://doi.org/10.1111/j.1468-0084.1980.mp42004001.x |
OXFORD BULLETIN
of
ECONOMICS and STATISTICS
STATISTICAL METHODOLOGY AND THE USE
OF ECONOMIC AND SOCIAL INDICATORS IN
THE ESTIMATION OF PER CAPITA GDP
LEVELS FOR DEVELOPING COUNTRIES*
By C. L. GILBERT
1. INTRODUCTION
There is a large number of purposes for which per capita national income
figures for different countries are required on a comparable basis. They include the
determination of aid priorities and of income-related contributions to international
or supra-national organizations. Despite this there is a tendency for some
economists to argue that because such comparisons can never be precise, they
should be completely avoided, but this objection would prevent virtually all
quantitative work in economics. A more realistic position is that to recognize that,
provided the inevitable imprecision in these estimates is recognized, they can be
valuable if used sensitively. A second objection to comparison exercises of this
form is based on the argument that per capita GDP is not always a very satisfactory
measure of welfare. This may be conceded, but since there is likely to be little
consensus on the form that a universal social welfare function should take, in the
end one is obliged to rely on per capita income or product estimates. These may of
course be supplemented where necessary by figures relating to other aspects of
welfare (for example, income inequality), and in the end it is probably more useful
for studies of this sort to provide the basic positive statistics from which welfare
measures, of whatever sort, may be calculated, than to impose a particular welfare
function which is unlikely to command wide assent.
In principle, the construction of per capita national income aggregates
presupposes the existence of a common set of prices in which to value identical or
very similar commodities in different countries, and it will only be possible to
construct such a set of international prices if relative prices between pairs of goods
do not differ between countries. This is far from being the case, but in the absence
of this condition, the almost universal practice of using market exchange rates to
convert own currency national income estimates to a common currency is not
* This paper draws upon a previous paper presented to the European Meeting of the Econometric
Society held in Geneva in September 1978. I am grateful to Bob Bacon for comments on this and on
other work incorporated in the present paper. The responsibility for errors, misconceptions etc. remains,
however, entirely my own.
281
Volume 42 November 1980 No. 4
282 BULLETIN
strictly valid. In practice. this objection may not be considered very important in
relation to other objections to the same practice, and in particular to the objection
that market exchange rates may not reflect purchasing power parities. But for all
these reasons there is a p7imafacie case for considering other methods of generating
national income comparisons.
There are three groups of alternative methods that we might consider: (i) One
might attempt to generate a set of international prices with which national income
in the various countries may be revalued. The most comprehensive realization of
this proposal is the International Comparison Project of the United Nations,
reported in Kravis et al. (1975, 1 978a). The drawback of this approach is that it is
time-consuming and very expensive, so that at least in the near future only a
relatively small number of countries may be expected to be covered. (ii) 'Short-cut'
methods have been proposed to circumvent the difficulties associated with the ICP
approach. In particular, if one could discover a robust relationship between
nominal per capita GDP (i.e. the own currency estimate converted using the market
exchange rate) and the purchasing power parity method of (j), one could use this to
interpolate the unknown purchasing power parity figures. This approach was
adopted by Kravis et al. (1 978b). (iii) There is by now a long tradition of using
economic and social indicators to bypass conventional national income estimates.
This approach is particularly attractive in relation to the least developed nations,
where there is reason to suppose that own currency national income estimates may
be even less reliable than elsewhere.
A major virtue of indicator methods for obtaining per capita GDP estimates is
that economic and social indicators are comparable in a fairly straight-forward
way between countries, and, relative to national accounting variables, they are
usually fairly accurately measured. Examples of indicators that are frequently
used are per capita energy consumption (an economic indicator) and percentage
literacy (a social indicator). Indicator methods for obtaining per capita GDP
estimates are surveyed in Heston (1973) and del Drago (1975).
Indicator procedures almost invariably estimate per capita GDP as a weighted
sum of the indicator values for each country (i.e. they almost invariably utilize a
linear framework). If therefore a full matrix of indicator values is available, the
problem becomes simply one of choosing the weights. The procedures adopted for
doing this may be classified into two groups depending on whether or not they
utilize information other than that in the indicator matrix for deriving the weights:
we shall classify techniques that do use such information as 'interpolative' or as
'calibrated procedures' and those that do not as 'free-standing'.
The best known free-standing technique is the principal component procedure
in which the indicator weights are taken as proportional to the leading principal
component of the indicator correlation matrix. There are two, well known,
objections to this procedure; (i) it is ad hoc and in no way related to national
accounting concepts; and (ii) there is no statistical model which rationalizes this
procedure.
The best known interpolative indicator-based technique is the Beckerman-
Bacon procedure (Beckerman (1966), Beckerman and Bacon (1966, 1970)). The
To continue reading
Request your trial