One can make hypotheses about the identity of the factors that are likely to influence spending in a given functional sector, and the significance of such factors can be empirically tested. Six groups of factors can be identified: (1) demographic influences, (2) sociological concerns, (3) the structure of the economy, (4) the level of economic development, (5) technological factors, and (6) environmental factors.

One can make hypotheses about the identity of the factors that are likely to influence spending in a given functional sector, and the significance of such factors can be empirically tested. Six groups of factors can be identified: (1) demographic influences, (2) sociological concerns, (3) the structure of the economy, (4) the level of economic development, (5) technological factors, and (6) environmental factors.

Demographic influences are likely to be principal underlying determinants of the demand for services. The larger the share of school-age groups in the population the greater the likely demand for education; the higher the percentage of elderly people in the population the greater the demand for medical care and perhaps more elaborate public mechanisms for old-age support. Other demographic variables, such as life expectancy, population growth, share of population in urban areas, and infant or child mortality rates, may imply the existence of a core underlying demand for certain types of services. Sociological concerns may explain whether there is a demand for the public sector to provide certain services; for example, the need for a social security mechanism is greater where extended family arrangements have broken down.

The sectoral structure of an economy may play a key role in shaping priorities for public expenditure. A dominant agricultural sector may require certain forms of public expenditure on agriculture to complement or service private sector activities. It might also be supposed that at low levels of development, the desire to change the structure of the economy may stimulate public expenditure in sectors that are not currently dominant elements in total output.

Technological factors influence the cost of realizing expenditure objectives. For example, the lower the desired pupil-to-teacher ratio the higher the cost of realizing a given percentage of enrollment for the population. Environmental factors may influence both the cost of providing services and the likely magnitude of the underlying demand; for example, poor access to clean water may imply a significant demand for investment to provide drinking water as well as the likely need for medical services because of the effects of contaminated water supplies.

These variables all focus on the major factors underlying the demand for public services. Yet, clearly, the level of real per capita income is the ultimate constraint on how much, in total, of that demand can be satisfied. In the typical low-income country, the recent high population growth rates have produced a population structure with a relatively high percentage in the school-age groups, which should imply a very high share of educational expenditure in total output. Yet the very low incomes may constrain government revenue so as to virtually preclude full enrollment even in primary schools, let alone in secondary schools. The quality of education will also fall short of that available in the higher-income countries. Thus, in analyzing the determinants of the share in gross domestic product (GDP) of public spending on a sector, the level of development (as measured by per capita GDP) seems to place a fundamental limit on possible spending in many sectors. It may also influence the likely balance in spending between the economic sectors, which are oriented toward stimulating current productivity and capital accumulation, and the social and administrative sectors, which aim at providing current consumption. It can be added, furthermore, that, while per capita income constitutes a constraint that limits the total provision of goods and services to satisfy a country’s need, it does not necessarily reflect the degree and the proportion in which these goods and services are provided by the public sector as opposed to the private sector. The division between the public and private sectors may reflect institutional and political considerations as well as the government’s capacity to obtain resources (through taxation) to finance these expenditures.

At a general level it may be hypothesized that the types of goods and services purchased by government expenditure—the so-called economic categories of expenditure—are significantly influenced by sectoral priorities. In any sector there may be a wide range of services, each potentially produced by a host of different technologies. Yet, on balance, the mix of labor, current consumption of other goods and services, and capital goods is likely to be different for each functional sector, so that the economic mix of expenditure will be largely determined by the functional mix. For example, a high share of spending on education is likely to imply a high share of spending on wages and salaries and perhaps on goods and services; similarly, a strong correlation might be expected between the share of spending on economic sectors and public capital formation. In developing predictive norms for appraising the share of spending on different economic categories of expenditure, the functional spending priorities are thus assumed to be the primary determinants.4

Sections III and IV describe the precise specifications used to explain the shares in total output associated with public expenditure on different sectors and different economic inputs. The equations are then used to predict a “norm” for spending on a sector or on a specific type of expenditure in a given country. The norm simply reflects what a country would be expected to spend on a sector, given the country’s economic, social, and demographic characteristics and given the actual expenditure of the large number of countries, both developed and developing, in the sample. In effect, the norm is defined according to how a large number of countries actually spend their funds, without regard to any external judgment about the optimality of this spending.

For any country, the ratio of actual to predicted expenditure ratios is computed and taken as an index for the purposes of international expenditure comparison—the IEC index. For example,

IEC health=(Actual health expenditure/GDPPredicted health expenditure/GDP)×100

A high value of the IEC index (e.g., above unity) for a functional expenditure category simply indicates that a country is spending more than would be predicted, given its economic and social characteristics (or in an IEC index for an economic input, given the structure of its functional expenditure). It does not indicate the actual share in GDP of a given category of expenditure; a country with a low IEC index (e.g., less than unity) may, nevertheless, be spending a higher share of GDP on a category of expenditure than a country with a high IEC index. For reference purposes, Appendix Tables 10 and 12 provide the actual functional and economic expenditure shares in GDP, and Appendix Tables 11 and 13 show the shares as a percentage of total government expenditure and net lending. By dividing the IEC indices for a country into these actual shares, the predicted shares may be calculated and expressed as a percentage.

The sources of the deviation of an IEC index from unity for a given country cannot be directly inferred from the results and may represent a conscious policy choice by the authorities to attach a different emphasis to a sector than is attached by its peer countries. An upper limit has been placed on the value of the IEC index. It is quite possible that the econometrically predicted values of the expenditure share in GDP may be a very small, or even a negative, number. As the IEC index equals the ratio of the actual to predicted shares, this ratio can lead to either a negative index value or to an exceptionally large value. Both simply indicate that a country is spending far more than would be expected. In both, a maximum value of 400 has been arbitrarily attached to the IEC index. Where an IEC index number is associated with a negative value and where the actual government expenditure is extremely small (under 0.1 per cent of GDP), the value assigned is shown as 400, although, in fact, it might be more appropriate to give a value of 100; after all, the actual expenditure is extremely small and the predicted expenditure is so small as to be negative, therefore, it could be maintained that actual is close to predicted, i.e., 100. However, in practice, in the seven cases where this happened (out of about 2,000 indices), the reader is signaled by an asterisk to treat the IEC number with care.

The data for the dependent variables for this cross-country study have been drawn from the most recent volume of the Government Finance Statistics Yearbook, published by the International Monetary Fund.5 Up to 93 countries have been included in this study, generally using 1977 as the base year for comparison. Expenditure has been disaggregated into the following functional and economic categories:

article image

The choice of independent variables was greatly influenced by the availability of data. The principal data sets were taken from the International Monetary Fund’s International Financial Statistics (IFS) and the World Bank’s World Tables.6 Several data problems were encountered. First, in calculating the share in GDP of expenditure, an adjustment in GDP was necessary where the fiscal year in the government accounts differed from the calendar year. For example, where the fiscal year 1977 ended on June 30, 1977, the use of GDP for calendar year 1977 could seriously underestimate the share of expenditure in total GDP, particularly if the country had experienced significant inflation during 1977. In such a case, a measure of the average of GDP in 1976 and 1977 was used. Second, in estimating per capita income, some obvious instances occurred where the use of a clearly overvalued nominal exchange rate yielded per capita income estimates that did not accurately reflect the relative income in a given country. As the IBRD World Tables also give estimates of per capita income for 1977, where these proved significantly different from the estimates derived from strict use of the nominal exchange rate, the IBRD estimates were used.

Third, the disaggregated public expenditure data in the GFS Yearbook relate to the consolidated central government accounts. In some countries the role of provincial and local governments is quite prominent, particularly in the provision of certain government services, notably education. Inclusion of central government spending alone would yield an excessively understated picture of the expenditure policies of such countries. In a recent study, the Organization for Economic Cooperation and Development (OECD) provided data on the share of total general government expenditure in GDP by functional categories for 1973–75. These shares have been used instead of the data in the GFS Yearbook for the following countries: Australia, the Federal Republic of Germany (hereinafter referred to as Germany), Canada, the United States, and Japan. On an economic classification, some data on the general government expenditure of these countries are available from the OECD.7 Other important countries (for example, India, Nigeria, and Brazil) were omitted because no comparable data were available.

Fourth, for some countries, the GFS Yearbook classification of expenditure obscures the ultimate intent of the expenditure. For example, block grants to localities in the United Kingdom are legally not earmarked for any particular sector and thus are included in the GFS Year-book under “other expenditure.” The OECD statistics indicate that much of this expenditure is, in fact, directed toward education, community services, roads, and housing.8 Similarly, it is often difficult to distinguish expenditure on health from expenditure on social security (as in Costa Rica). When such problems are obviously distortive, an attempt has been made to reclassify expenditure in the appropriate functional expenditure categories by using country or OECD sources of information. In specifying the model, one equation has also been estimated to predict the sum of health and social security expenditure in order to capture any obvious example of misclassification.

Finally, all the equations were estimated by using the least-squares method. Multicollinearity was tested in every case and variables exhibiting major multicollinearity were rejected. In specifying the equations, multiplicative dummies were used to test whether there might be discontinuities in the effects of individual independent variables according to per capita income. Alternative amounts of per capita income were tested as the breakpoint for such discontinuities, and it was observed that, where significant at all, a per capita income of US$1,750 seemed to yield the lowest sum of squared residuals for the equations. In general, such multiplicative dummies appeared statistically significant only in the equations explaining the functional expenditure shares.