Statistical relationships between growth and social structures

One may think of literally thousands of aggregate characteristics of societies showing extremely high degrees of correlation with indicators of economic development, either when comparing different countries at different levels of development or when analyz­ing the evolution of a single country over time.

As an example of the correlation approach to the consequences of growth, Table 1 shows the relationship between the level of economic development and a few indicators that very roughly describe changes in societies’ economic and social structure generally associated with economic growth. As it will be seen later in this chapter, these indicators describe important channels through which growth and development may modify social structures. They include the size of the government, the level of urbanization, education, health, demographic patterns, labor force participation, gender differences and income inequality. The first three columns of the table report the results of a simple regression of these indicators on GDP per capita expressed in 2000 US dollars after correction for purchasing power parity.

It can be checked there that, with two exceptions, all indicators appear to be sig­nificantly and strongly correlated with economic growth. For instance, focusing on the pooled regression, the GDP share of public expenditures is shown to increase by 0.5 percentage point when GDP increases by $1,000 (thus confirming ‘Wagner’s law’) although this coefficient is not statistically significant for the 1970 cross-section. Like­wise, the urbanization rate is shown to increase by 0.3 percent and the literacy rates by 3-4 percentage points in presence of the same increase in income per capita, whereas fertility decreases by 0.15 children; the 1970 cross-section shows slightly different re­sults in all of these cases. As a final example, income inequality and female gender bias appear to be significantly and negatively correlated with growth. In the case of the for­mer, a parabolic regression exhibits the familiar inverted-U shape introduced by Kuznets some 50 years ago^[432]- with a nonsignificant result occurring with the 1990 cross-section.

All these results are interesting. Yet, there are various reasons to think that simple regression on a cross-section of countries or even on a pool of cross-section time-series observations is a very crude approach to identifying the consequences of growth. On the one hand, the existence of a correlation does not say much about the causality link between two variables. Causality may be direct in either one direction, or possibly in both. It may also be indirect and simply reflect the fact that the two variables under scrutiny are both related to a common set of other variables. On the other hand, GDP per capita tends to increase more or less regularly over time so that there may be a confusion between its effect on socio-economic indicators and that of other variables with a comparable time trend.

Alternative econometric specifications permit taking into account some of the pre­ceding points. At the same time, however, they often modify the order of magnitude and the significance of the preceding relationships. In a few instances, they even modify their direction.

The next four columns of Table 1 show estimates of the growth sensitivity of socio­economic indicators obtained with alternative econometric specifications. In all cases, the sample is obtained by pooling country data over various years in the periods 1960-2002. In column (4), a set of year dummy variables is added to the regression. This accounts for the fact that socio-economic indicators might evolve over time un­der the influence of some common factors independent of national economic growth. In column (5), it is a set of country dummy variables that is introduced so as to con­trol for ‘fixed effects’ or, in other words, the effect of largely unobserved fixed country characteristics that might affect both the original level of GDP per capita and that of the indicator under scrutiny. The corresponding estimate of growth sensitivities thus ab­stracts from differences in country means and takes into account only differences in the average time behavior of GDP per capita and socio-economic indicators across coun­tries. Column (6) combines both approaches by abstracting from differences in country means as well as from an exogenous nonlinear time trend common to all countries. Finally the estimates in column (7) are obtained by running the simple regressions of socio-economic indicators on GDP per capita in decadal differences.

Adding a common nonlinear time trend to the original simple model does not mod­ify the growth sensitivity of the socio-economic indicators in a significant way. More substantial changes are obtained when fixed country effects are introduced. As could be expected, growth sensitivity generally falls when cross-country differences are ignored, or more exactly when cross-country differences are attributed to fixed characteristics that include, inter alia, initial development levels.

Table 1

Estimated growth elasticity of selected economic and socio-demographic indicators

1708 F. Bourguignon

Table 1

(Continued)

Ch.

Source: Author’s calculation on WDI data. Notes. Robust T-statistics in brackets. Cox transformation of variable x is x/(1 — x). Labor force participation in %. Significant at 5%.

** significant at 1%.

cases - as for instance with fertility and gender life expectancy differential - the sign of growth sensitivity is even reversed.

Of course, such a correction of the original estimates may well be excessive. Adding a time trend is certainly bound to reduce growth sensitivity estimates, especially when estimation abstracts from cross-country differences. The time trend is likely to pick up those changes in the indicators which are independent from country specific economic growth. Yet, results obtained with that method are not always very convincing. In partic­ular, that fertility would significantly increase as a response to growth once independent forces are taken into account, as shown in the last two columns of Table 1, seems to be in contradiction with the intuition of most demographers.^[433] The results shown in Ta­ble 1 are likely to mask some heterogeneity among countries with respect to the drop in fertility that is not dependent on economic growth.

The estimates reported in the last column of Table 1 confirm the preceding results. Restricting the analysis to correlation in decadal differences overall shows the same order of magnitude for the growth sensitivity of socio-economic indicators, but also makes those sensitivities often nonsignificant.^[434] In comparison with simple regressions and correlations, estimates based on differences or on fixed effects thus suggest that the evolution of the few general socio-economic indicators considered in Table 1 proba­bly obeys other forces in addition to economic growth, or that the effect of growth is less simple than implicitly assumed in these statistical models. In particular, it is quite possible that the effects of growth on socio-economic indicators are strongly heteroge­neous across countries.

Taking into account this heterogeneity across countries and going beyond the simple statistical techniques used to produce the results of Table 1 meets a fundamental con­straint: the limited number of observations. Estimating the preceding model country by country on a time series basis would certainly permit to fully account for country speci­ficity. But it would only inform us about the consequences of growth at a particular stage of the development process of a given country. On the other hand, taking into account observed heterogeneity by interacting growth rates or development levels with a host of country characteristics and policy variables and by introducing nonlinearity is also bound to run into too few degrees of freedom. Social consequences of growth take time to show up, so that the informational content of annual time series is not proportional to the length of these series. In Table 1, one can see that there is little difference between the last two columns even though column (6) relies on full annual series whereas col­umn (7) uses only decadal differences. This means that the information that matters is not year-to-year fluctuation but ‘episodes’ of growth characterized by uniformly high, moderate or low growth rates. If this is the case, then available data may make it difficult to estimate with satisfactory precision the observed heterogeneity in the consequences of growth.^[435]

In summary, the analysis in this section suggests that other forces than economic growth are behind the time evolution of most socio-economic indicators, even though the absolute value of simple correlation coefficients is often very high. It is also possi­ble that the effect of economic growth on these indicators and the social features they describe is too complex to be described by simple regression analysis. In particular, the relationship may be nonlinear - as hypothesized for instance by Kuznets for income inequality - or it may depend on specific country characteristics, including policy and institutional variables. There is no simple statistical method to identify a priori these other forces or these interactions, and limited observations may also be a serious hin­drance for this identification.

Under these conditions, it is likely that the only methodological approach able to help identify and understand the consequences of economic growth is of a ‘structural’ nature. In other words, what is required is to establish hypotheses on the phenomena that guide the overall evolution of the socio-economic indicators under analysis and to test the corresponding model. This is in stark contrast with the reduced form approach so often found in the literature and illustrated by Table 1. Within a structural approach, a theo­retical model of the behavior of the socio-economic indicator being studied must first be established on the basis of economic theory. This is what the next section attempts to do for some possible social consequences of growth.

2.