Using Principal Components to Produce an Economic and Social Development Index: An Application to Latin America and the U.S

Article excerpt

NICOLAS SANCHEZ [*]

This paper presents a principal components methodology for determining the weights for a set of indicators in a composite index of development. The procedure is applied to a 36-variable data set consisting of 1990 data for 19 Latin American countries and corresponding 1960 and 1990 data for the individual U.S. states. This paper compares the results with other well-known indices and uses the combined data set to better understand the level and scope of development in each region and over time. The general results are that the level of development of Latin American countries in 1990 are roughly distributed over the U.S. states in 1960 (though with a larger range), and the structure of development in Latin America is similar to the U.S. (JEL O57)

Introduction

This paper will present an index of development used to compare the level of development of the U.S. states in 1960 and 1990 to 19 Latin American countries in 1990. The index methodology used contrasts with other well-known measures in that a statistical procedure is used to determine the weights for the variables in the index and to some extent the particular set of variables itself. This exercise is useful since it provides a way for those familiar with U.S. economic development to understand the development of Latin America and vice versa. The general results show that Latin American countries are roughly distributed around the U.S. states in 1960, though with a wider range.

An Overview of Development Indices

The field of development measurement is one of the most hotly contested subjects in economics. Until the 1970s, gross national product (or gross domestic product (GDP)) per capita was the main gauge of development. [1] These average income figures appear to be natural measures of welfare, but it became increasingly evident a few decades ago that they are inadequate for measuring human development, especially because they contain little information pertaining to the distribution of income. [2] As a result, in the 1970s, increasing attention was paid to other socioeconomic variables. [3] The general result of this literature is to question the validity of income per capita as the best measure of development. [4] Yet, a single yardstick is desirable for measuring development, and several composite welfare measures have been suggested (for example, see McGillivray [1991], Camp and Speidel [1987], Hicks and Streeten [1979], and Morris [1979]). The most well known is the human development index (HDI), constructed b y the United Nations Development Program [various].

Since its inception in 1990, the HDI has become the standard way to compare levels of development. [5] In its current form (last revised in 1999), it purports to be the unweighted average of three separate indices which are designed to measure health and longevity, knowledge and communication, and access to goods. To form the three component indices, the methodology rates each country from zero to 1, where zero is considered the minimum possible value, and 1 is considered the maximum possible value (Kakwani [1993] uses a similar methodology). The health and longevity index uses life expectancy at birth, the knowledge and communication index uses the literacy rate (with a weight of 2/3) and a schooling rate (with a weight of 1/3), and the access to goods index is the natural logarithm of the purchasing power parity (PPP) measure of GDP per capita. [6]

The often conflicting critiques of the HDI originate on several grounds. The harshest criticisms are summarized by Srinivasan [1994, p. 241] who condemns the EDI as being "conceptually weak and empirically unsound, involving serious problems of noncomparability over time and space, measurement errors, and biases." Several researchers have criticized the data used by the HDI. McGillivray [1991] suggests it is inappropriate to use variables that are so closely correlated. Srinivasan provides criticisms of the particular statistics used. …