Country size and welfare

We consider a set of independent variables similar to the set in Chapter 3, where we analyzed the determinants of government size (government consumption) to assess the influence of country size on welfare in mul-tiple regressions based on OLS, where standard errors are White hetero-scedasticity-consistent.¹⁰⁶The results are summarized in Table 5.2, and there is a reasonable case for rejecting the central theoretical predictions.

There does not seem to be any relationship between country size and welfare, even when we include a set of control variables which might help in explaining the sources of welfare. Again, analogously to Chapter 3, country size is measured by population figures and welfare by per capita GDP. We are, of course, aware of the shortcomings of these mea sures but have to stick to them, because they are available and harmo -nized for a sufficiently large sample of countries. Similar to Chapter 3, we use logs when distributions are very skewed. Descriptive statistics and the correlation matrix of all variables can be found in Tables 5.3 and A.16 in the Appendix.

The results in Table 5.2 are clearly not in line with theoretical pre-dictions. The log of population is not significant in any of the seven mo-dels chosen. Furthermore, with the exception of model (2) the coeffi-cients are far from being significant, and we therefore clearly do not ob-tain a result that supports theoretical expectations. The univariate re-gression with a comprehensive set of 158 countries under consideration fares poorly with regard to explaining welfare. Due to the high number of observations, this is a remarkable result. In model (2) we arrive at the expected result that trade openness plays an important role in determin -ing welfare. This is the only model where the log of population is not far from being significant, but one has to bear in mind that we have fewer observations than in model (1), and the fit of the model is rather poor, which is somehow surprising, because we expected the variable open ness to exhibit considerable explanatory power. Note that the bivariate cor-relation between openness and welfare is surprisingly low¹⁰⁷, but there is – as expected and in line with theories in international economics – a

106 For details on the method and on Whitecorrection see Section 3.2.3.

107 Pearson correlation coefficient: 0.220 (p = 0.029; two-sided).

high and significant (negative) correlation between the log of population and openness.¹⁰⁸Even though correlation between the log of population and the log of per capita GNP is low (even negative, but not significant-ly), it is correct to rule out the linear influence of size on the correlation between welfare and openness by running a partial correlation. Irrespec -tive of the openness measure chosen, the magnitude and the significance of the correlation remains, nevertheless, almost unchanged.

Models (3) – (7) show relatively high adjusted R², and most of the coefficients have the expected sign. Trade openness and the freedom in-dex are always significant. Note that higher figures mean lower freedom and that the negative sign of the coefficient is therefore perfectly in line with theoretical notions. The indices for vulnerability and transport costs are also significant, and their magnitude is not negligible, especially when we regard the vulnerability index. In other words, the remo te ness and vulnerability of an economy or a region are important deter -minants of welfare. Given the fact that at least all VSC in the Pacific re-gion have to be considered remote, having to bear high transport costs for imports and exports, and that most of island VSC in the Pacific and the Caribbean exhibit high ratings of vulnerability due to natural hazards and environmental problems, the disadvantage of those coun-tries becomes obvious. Although vulnerability indices also incorporate size as a source of vulnerability¹⁰⁹, it is not difficult to conclude that the countries in the Pacific and in the Caribbean would have lower growth rates, higher unemployment and a lower living standard even if they were not so small. Hence, it is difficult to compare their economic per-formance to that of European VSC due to the remarkable effects of transport costs and vulnerability. Note that because of the high correla-tion of the two variables resulting from their definicorrela-tion, their magnitude and significance decrease when both are incorporated in one model (see model (4) versus models (3) and (5)).

Geographic dummies do not pop up with surprising news. The OECD dummy is, as expected, always positive and significant, the Sub-Saharan Africa dummy is negative and significant. The only slightly

sur-108 Pearson correlation coefficient: –0.636 (p < 0.001; two-sided).

109 Note that vulnerability indices also take the level of transport costs into account. It is therefore not surprising that the correlation between those two independent varia-bles is very high (Pearson correlation coefficient: 0.630) and highly significant.

Table 5.2: OLS regressions for log per capita GDP and log population

Sources: logarithm of GNP, per capita income and population density from Baratta (1999); urbaniza -tion rate from World Bank; vulnerability, transport costs and trade openness from Briguglio (1995) based on IMF and UNCTAD statistics from 1991; freedom index from Freedom House; war dummy, war time and revolutions per year from Barro-Lee.

(4) (5) (6) (7)

aother source for trade openness: Penn World Tables (from 1985)

** significant at the 1% level; * significant at the 5% level; t statistics based on White heteroscedasti city-consistent standard errors in parentheses

prising result is the significant sign of the South East Asia dummy. More confounding is the insignificance of the three variables concerning war and revolutions. They are far from being significant, although we expec-ted them to have clear negative impact on welfare. There are several explanations for the insignificance of the three variables. The most promis -ing line of arguments is that many countries currently involved in an armed conflict with adjacent countries or in state of revolution fail to provide data, which means that they are ruled out a priori. The small number of observations in models (3) and (4) also points in this direc-tion. Another intuitive explanation is the fact that the variables might not be able to measure the central effects properly, because they cover a relatively long time period. Contrary to that, we would assume that only those conflicts, which have been taken place in recent years, determine welfare (with the exception of long-lasting conflicts, of course).

The effect of population density is rather ambiguous, since it is only significant in one out of three models. Contrary to that, the urbaniza tion ratio seems to be a stable determinant of wealth. The magnitude of its effect and its significance is very similar across different specifications. A higher urbanization ratio is associated with a higher per capita GNP.

Abbreviations: Abbr. = Abbreviations; Obs. = Number of Observations; St. dev. = Standard deviation.

Variable Abbr.

Trade openness 1991 open

Population density (pop/area) popdens

Log of population 1996 logpop

Urbanization ratio 1997 (in %) urbrat

Dummy for Latin American countries laamd

Dummy for OECD countries oecdd

Dummy for Sub-Saharan African countries africad

Dummy for Asian countries asiad

War dummy ward

War time wart

Revolutions per year revo

Index of freedom (1-7) freedom

Vulnerability index vul

Log per capita income 1996 lognpc

Transport cost index 1991 trans

Table 5.3: Variables, abbreviations, sources and standard statistics

Note that one has to exercise caution in interpreting this result causa -tion ally, since it is at least as plausible that welfare is a determinant of ur-banization.

The main purpose of model (6) is to test for the effect of country size by having as many observations in the regression as possible. We therefore had to exclude some variables which were only available for less than 100 countries, but the goodness of fit of the model is still re-markable. All variables in model (6) are highly significant with the ex-ception of the logarithm of population, although trade openness does not appear in the model. We take this result as a further confirmation of our argument concerning the irrelevance of size. The effect of trade openness may be replaced by two variables which are highly correlated with openness, i.e. population density and vulnerability.¹¹⁰Whereas the high correlation between vulnerability and trade openness is not surpris -ing due to the definition of the vulnerability index, there is no apparent

110 Pearson correlation coefficients: 0.582 (p < 0.001; two-sided) for openness and pop-dens; 0.690 (p < 0.001; two-sided) for openness and vulnerability.

Source Obs. Mean St. dev.

Briguglio based on IMF 114 0.39 0.27

Baratta (1999) 191 229.63 1234.71

Baratta (1999) 191 6.63 1.00

World Bank 147 54.17 23.02

Barro-Lee, own 195

Barro-Lee, own 195

Barro-Lee, own 195

Barro-Lee, own 195

Barro-Lee 118

Barro-Lee 118 0.08 0.18

Barro-Lee 133 0.18 0.28

Freedom House 189 3.53 2.01

Briguglio 114 0.45 0.14

Baratta (1999) 159 3.22 0.67

Briguglio based on UNCTAD 114 0.19 0.24

rationale – at least as far as we know – for the correlation between openness and population density. The only straightforward argument would run along the following lines: Smaller countries are more open and, generally, have higher population densities, so that there is a high correlation between openness and population density. A partial corre lation controlling for country size (logpop) does, however, not con -vincingly support this notion, and the bivariate correlation between the logarithm of population and population density is rather low. The reason for the high correlation is therefore not clear.

In model (7) we rely on another data source for trade openness in order to have more observations. It should not matter that data from the Penn World Tables for openness stem from 1985, because countries’

openness indices should not change dramatically over time, and if they do change, the developments should, on average, be rather parallel across countries (with the notable exception of former Eastern European coun-tries). Indeed, results change only slightly, which indicates that the ef-fects described above are sufficiently stable. With regard to country size, we can conclude that it does not determine welfare.¹¹¹