The cliometrics of academic chairs. Scientific knowledge and economic growth: the evidence across the Italian Regions 1900–1959

Abstract

The paper elaborates and tests two hypotheses. First, that knowledge is not a homogeneous activity, but rather a bundle of highly differentiated disciplines that have different characteristics, both in terms of generation and exploitation, that bear a differentiated impact on economic growth. Advances in scientific knowledge that can be converted into technological knowledge with high levels of fungibility, appropriability, cumulability and complementarity have a higher chance to affect economic growth. Second, that academic chairs are a reliable indicator of the amount and types of knowledge being generated by the academic system. Hence the analysis of the evolution of the academic chairs of an academic system is a promising area of investigation. In this paper the exploration of the evolution of the size and the disciplinary composition of the stock of academic chairs in five Italian macro-regions in the years 1900–1959 provides an opportunity to understand the contribution of scientific knowledge to economic growth in each regional system. The econometric analysis confirms that advances in engineering and chemistry, as proxied by the number of chairs, had much a stronger effect on the regional economic growth than advances in other scientific fields. These results have important implications for research policy, as they highlight the differences in the economic effects of academic disciplines, and for the economics of science, as they support the hypothesis that academic chairs can be used as reliable indicators of on-going research activities in the different types of scientific knowledge.

Appendices

Appendix 1

1.1 1A. Computation of the regional levels of income per capita in each of the five identified macro-regions

The first step in order to compute the level of income per capita in each macro-region consists in computing the levels income per capita for each of the Italian regions: we multiply the regional differential (rdiff _jt ) provided on a yearly bases by Daniele and Malanima (2007, 2011) with the time series of Italian national GDP per capita (using the latest series in costant prices provided by Malanima), we are then able to obtain the regional levels of GDP per capita (r_prod _jt ):

$$ r\_prod_{jt} = rdiff_{jt} *GDPpc_{t} $$

(7)

where j denotes the region and t denotes time

Then we aggregate these series in order to obtain the level of GDP per capita in each of the five macro regions identified (Y _it /P _it ), by using a weighted average of the regional levels of income per capita; w _jt denotes the share of population of region j within the macro region i, as provided by the population census every ten years.

$$ {{Y_{it} } \mathord{\left/ {\vphantom {{Y_{it} } {P_{it} }}} \right. \kern-0pt} {P_{it} }} = \sum\limits_{j} {w_{jt} r\_prod_{jt} } $$

(8)

where j denotes the single regions, i denotes the macro-region and t denotes time.

1.2 1B. The reconstruction of the regional level of public expenditures in Italy

In order to compute the level of public expenditures in public works in each of the five Italian macro-regions we take advantage of the data by Picci (2002) who provides the shares of the total expenditures in public works at the national level for each region. Specifically Picci provides the shares corresponding to different subperiods, drawing on different data sources (De Stefani 1925; Montanaro 2002). The subperiods that cover the time-span we are interested in (1901–1959) are respectively: 1895–1905, 1906–1915, 1915–1924, 1928–1932, 1933–1937, 1938–1942, 1943–1947, 1948–1953, 1954–1959. For the years from 1925 up to 1927, for which we lack data, we assumed a constant rate of increase between the regional shares of 1924 and those of 1928. In order to obtain the share of national public expenditures for each macro-region we simply summed the share of the individual regions as follows:

$$ mr\_ps_{it} = \sum\limits_{j} {r\_ps_{jt} } $$

(9)

where r_ps denotes the share of expenditures in public works in each region and mr_ps denotes the share in each macro-region (j denotes the regions and i denotes the macro-region). Finally we multiplied the shares obtained through this procedure by the expenditures in public works (in costant prices) made available by the recent work of Baffigi and the Bank of Italy on the Italian national accounts, which constitutes the most reliable source of data on Italian public accounts at the national lavel Hence:

$$ mr\_public_{it} = mr\_ps_{it} *nat\_public_{t} $$

(10)

in which mr_public denotes the expenditures in public works in each of the macro-region, while nat_public indicates the time series of national expenditures in public works at the country level as provided by Baffigi (2011). At the end of this procedure we computed the level per capita of expenditures in public works at the macro-regional level in the following way:

$$ {{PUB\_INV_{it} } \mathord{\left/ {\vphantom {{PUB\_INV_{it} } P}} \right. \kern-0pt} P}_{it} = \frac{{mr\_public_{it} }}{{P_{it} }} $$

(11)

where P indicates the population of each macro-region in each single year of observation.

Appendix 2: Testing for panel stationarity and co-integration

In order to test for the presence of a cointegrating relationship between the variables of interest in Eq. (3) our methodology will follow the path usually adopted in the literature: the first step consists in the examination of the time series properties of the data, checking for the presence of unit-roots, then if the variables appear to be non-stationary we will check for the existence of a cointegrating relationship among our variables. If a cointegrating relationship exists, it will be possible to estimate Eq. (3) adopting panel cointegration techniques.

In order to check for the stationarity of the variables we use the panel unit root test of Im, Pesaran and Shin (2003) (IPS) that allows for a heterogeneous autoregressive unit root process across cross-sections and is based on an augmented Dickey-Fuller (ADF) regression for each cross-sectional unit:

$$ \Updelta x_{it} = z_{it}^{\prime } \phi + \rho x_{it - 1} + \sum\limits_{j = 1}^{k} {\gamma \Updelta x_{it - j} + \varepsilon_{it} } $$

(12)

where z _it is a vector including deterministic terms such as the constant and the time trend, while k is the lag order. The test statistic is built as the average of the individual (i.e., for each cross-sectional unit) ADF statistics. The null hypothesis is that all units contain a unit root (i.e. ρ _i  = 0) against the alternative that at least some units display a stationary process.¹⁴

If, as expected, the variables are non-stationary (because of the presence of a unit root), the next step is to test for cointegration. We will test for the presence of cointegration by taking advantage of the tests devised by Kao (1999) and Pedroni (1999, 2001, 2004) and usually implemented in the literature. Both tests check for the presence of a cointegrating relationship among the variables, however Pedroni test is more complete, since it introduces seven types of tests allowing to test for the stationarity of the residuals obtained from the equation of interest under many different hypothesis about the properties of the relationship in terms of heterogeneous slopes, fixed effects and individual trends.

Table 7 reports the results of the IPS (Im, Pesaran and Shin 2003) test about the presence of unit roots in the variables of our model. The null hypothesis of the test is that all units display unitary roots, while the alternative is that only some of them are non-stationary. In the left panel are presented the results of the IPS test on the variables with and without the inclusion of a trend. The results confirm that, disregarding the inclusion of a time trend, all variables contain a unit root.¹⁵ This suggests the impossibility to apply the usual panel-data estimation procedures on the levels of these variables. In the right panel of Table 7 we run the same test on the variables in differences, the results confirm the I(1) nature for the great majority of the variables, since their first differences are stationary.

Table 7 Tests for stationarity

In Table 8 instead we present the results of the residuals-based test on the existence of a cointegration relationship among these variables. As previously said we rely on the usual tests on cointegration introduced by Kao (1999) and Pedroni (1999). As it is evident in the Table the Kao test confirms the presence of cointegration: the null hypothesis of no-cointegration is rejected at the 95 % level. The strategy proposed by Pedroni instead introduces seven different tests about the hypothesis of stationarity of the residuals: our results show that in almost all of the tests the null hypothesis of no-cointegration is again rejected at the 95 % level. All these results together confirm that there actually is a cointegrating relationship between the variables of Eq. (3) and hence drives us towards the estimation of Eq. (6). It must be stressed that, since the Pedroni test does not allow more than seven variables to be tested for cointegration, when we included also the per capita expenditures in public works we had to aggregate together the chairs in medical sciences (MS) and in other natural sciences (ONS).

Table 8 Cointegration tests