The story of two transitions: Unified Growth Theory and the European growth experience, 1300-1870 Alexandra M. de Pleijt and Jan Luiten van Zanden Utrecht University E-mail corresponding author: A.M.dePleijt@uu.nl Abstract: This paper sets out the firstly establish the ‘stylized facts’ about the transition from an economy dominated by Malthusian forces to one characterized by ‘modern economic growth’, as it occurred in Western Europe between 1300 and 1850. Thanks to the new research in the evolution of population, GDP and its components, we can present a much clearer picture about patterns of long-term economic growth in North-Western Europe, in particular England and Holland. This new research has not been used in recent contributions to ‘unified growth theory’, which we consider a missed opportunity. We will try to explain these ‘stylized facts’ testing various insights from ‘unified growth theory’, in particular about the links between demographic change, human capital formation, and economic growth. 1. Introduction We aim at integrating new empirical results into debate on transition to modern growth regime. To be written. 2. The stylized facts of European economic growth, 1300-1850 All ‘facts’ are interpretations, so if we want to establish the ‘stylized facts’ about European economic growth in the centuries before the Industrial Revolution, we have to be careful which period and territorial entity we focus on. Can we explain the Industrial Revolution by looking at growth in the 18th century only, or is a much longer time-span required? What is the relevant entity we should look at: is growth indeed a sub-national/regional phenomenon, or is the ‘nation’ state the appropriate unit of analysis, or is growth not limited by ‘national’ borders? As we will demonstrate in this section, these are relevant issues for deciding what the stylized facts are. Let’s begin with the issue: where is economic growth located? The concept of the ‘Little Divergence’ is relevant here: a substantial body of evidence – starting with the real wage estimates by Allen (2001), and including the new generation of GDP estimates already introduced – points to the fact that there was a divergence in levels of economic performance within Europe between 1500 and 1800. North-Western Europe – basically the Low Countries and England – showed more or less stable real wages (after the increase in real wages in the 14th century following the Black Death), whereas real wages in Eastern and Southern Europe declined sharply in the long run. In terms of GDP per capita there is a similar divergence between the North-Sea area and the rest of the continent: in the latter real incomes stagnate or even decline somewhat, whereas Holland and England show a lot of progress: they are substantially richer in 1750 than in 1500 (Broadberry et.al.; Prados de la Escosura; Bolt and van Zanden 2013). This is mirrored in patterns of urbanization (rising in the North Sea area, stabilizing in the rest of Europe), in the development of political institutions (parliaments decline almost everywhere, except for England and the Netherlands) (Van Zanden, Buringh and Bosker 2012), and in indices linked to human capital formation (such as book production and consumption) (Baten and Van Zanden 2008). So growth seems to be an international 1990 GK dollars phenomenon, originating in the region bordering the North Sea: the Low Countries and England formed the core of this wider region, but parts of Belgium, Northern France, NorthWest Germany, and Scotland, were probably also participating in it (but we do not have detailed statistics and estimates to test this, and it may be that the region experienced long term growth was also changing over time – for example, Scotland maybe only joined in the 17th or even 18th century). This was a strongly integrated region, in which, for example, in the late Middle Ages, England supplied the wool for the textile industries of the Low Countries, the main source of employment of the large Flemish cities. In the 14th-15th centuries Flanders formed the urban core of this economic system – and England its ‘periphery’. In the 16th century Brabant (and in particular Antwerp) took over the role of being the core. After 1585 the urban center moved to Holland, a switch that resulted in the Dutch ‘Golden Age’ of the 17th century. After about 1650 London gradually replaced Amsterdam as the central hub in the commercial network of North-Western Europe, and he urban core switched to England, as a result of which the Netherlands in the 18th century started to specialize on livestock products for the British market – confirming its role as ‘new’ periphery. Our working hypothesis (to be further tested below) is therefore that growth was a regional phenomenon, not (really) limited by international boundaries, and concentrated in the North western part of Europe, or more specifically, in the North Sea region. The next question concerns the beginning and the periodization of the process of economic growth in this region. Here we suggest a division into three periods: before the Black Death (‘classic Malthusian economy’), between 1347 and 1820 (slow but consistent growth), and after about 1820 (rapid, post-Industrial Revolution growth). This is based on the recent work on estimating GDP in the long run mentioned in the introduction. Figure 1 presents the big picture: estimates of GDP per capita of England (annual estimates, 12701700), Great Britain (annual, 1700-1870), Holland (annual, 1347-1807), the Netherlands (annual, 1807-1870) and Belgium (point estimates, 1500-1850). GB England NL Holland Belgium Figure 1. Per capita GDP growth in England, Holland and Belgium, 1270-1870 Sources: Broadberry et.al.2012; Van Zanden and Van Leeuwen 2012; Buyst 2011. 2 For the period before 1347 we only have detailed estimates for England, starting in 1252 (Figure 2). This series shows no growth before 1348; the trend is indeed slightly negative, mainly due to serious crises in the 1280s and the 1310s. Population growth seems to matter a lot: the population grows from 4.2 million in 1250 to 4.8 million in 1290, when GDP per capita registers a declining trend; after the crisis of the 1280s, the population starts to fall (to about 4.6 million in 1340), which results in a stabilization of real income (albeit with huge fluctuations, such as the ‘Great Famine’ of the 1320s. In short, this economy shows distinct Malthusian features, with a decline of real GDP during the (last stage of the) great Medieval boom (1080-1280) during which the population increased a lot, and stability after the 1280s because also population growth came to a halt. The level of GDP per capita is, however, much above subsistence levels, especially at the start of this century; during the 1250s the English GDP per head of 800-850 dollars is between at least twice and almost thrice the subsistence level (of 300-350 dollars), and quite high by international standards – only the more wealthy ancient economies reached similar levels, if we are to believe recent estimates of pre-1000 GDP. The earliest estimates for Holland (in 1347) also point to similar high GDP (almost 900 dollar). 950 5.000 900 4.800 850 4.600 800 4.400 750 700 4.200 650 4.000 600 3.800 550 500 1252 3.600 1262 1272 1282 1292 1302 GDP per capita (left-hand) 1312 1322 1332 1342 Population (right-hand) Figure 2. English per capita GDP and population growth, 1252-1347 Sources: Broadberry et.al. 2012 The starting point is therefore an already relatively prosperous pre-industrial economy, which however showed certain Malthusian traits. The flat trend between 1250 and 1347 is broken by a large decline of the population after 1347, resulting in a sudden increase in real per capita GDP – both in England and in Holland. From the Malthusian perspective this is what is to be expected, so it seems to confirm the Malthusian character of the Late Medieval Economy. In other parts of the world (Spain, Egypt), the collapse of population did not have similar positive effect on real incomes – so it is not inevitable that the Black Death has such a result and it probably tells us something about the quality of institutions in the North Sea area: a large exogenous shock such as the Black Death did not result in collapse of the economy (as appears to have happened in Egypt), but in a sudden increase in real incomes and real wages. One could argue that this is a typical ‘Malthusian’ response, but it is also exactly what one 3 would expect to happen in a market economy: one factor of production – human labour – suddenly because much more scarce, as a result of which its remuneration increases. What is even more important, is that after this exogenous shock real incomes did not return anymore to the pre-shock level, but remained much (20-40%) higher than before, and, especially in the case of Holland, began to show a consistent rate of growth (of about 0.19% per year), which resulted in a doubling of GDP per capita in the next 250 years (not taking into account the effect of the 1347/48 shock). The English growth experience was somewhat different: between 1400 and 1600 real incomes fluctuated around a plateau of about 1000 dollars (against 700-800 dollars before 1347), and only in the second half of the 17th century growth really took off. We know less about growth in Belgium, but the few point estimates that have been made (by Erik Buyst) suggest that it followed a third pattern: its level was quite high at the start of the 16th century, there was some per capita growth until the 1560s or 1570s, but the late 16th and 17th centuries saw a decline of GDP per capita, due to the loss of industrial and tertiary activities to the Northern Netherlands (Figure 1). The decline of Belgium and the rise of the Netherlands were part of the same process, in which the urban center of the North Sea area moved from Antwerp to Amsterdam. The next change in the urban system, from Amsterdam to London, began in earnest after about 1670, and had similar consequences (England again began to grow, Holland stagnated at a high level). The main point about growth in this second phase – beginning in 1347 and ending in the early 19th century – is related to this. Superficially, each country knew its separate growth cycles, characterized by period of rapid growth followed by (and/or preceded by) long periods of stagnation. Flanders boomed in the late Middle Ages, but declined after about 1560; Holland had its most spectacular growth spurt during its Golden Age (1585-1670), but stagnated in the 18th century; conversely, English real incomes did not increase between 1400 and 1600, but rose rapidly after 1650. Scholars have concluded from this that growth was sporadic, intermittent, non-consistent – that growth spurts were always followed by stagnation and/or decline. Perhaps Jack Goldstone’s concept of periodic ‘efflorescences’ best captures this idea. But what seems a reasonable picture of growth before 1800 if one looks at individual regions within the wider North Sea region, is perhaps plain wrong when looked at the region as a whole. Figure 3 attempts to do this by merging the two (England and Holland) or three (also including Flanders – but only after 1500) into one index of real GDP per capita of the North Sea region as a whole. What then appears is a remarkable stable path of economic growth, in which stagnation in one part of the region (in England between 1400 and 1600, or Holland after 1670) is compensated for by growth in the rest of the region (in Flanders and Holland when England stagnates, and in England when Holland’s growth decelerates). But the region as a whole grows at a stable rate of 0.18 percent per year, almost without real interruption, between 1347 and 1820 (after which growth accelerates). The phases of growth and stagnation within certain sub-regions are linked to changes in the international division of labour: when a certain sub-region (Flanders first, followed by Holland, and finally England) manages to attract the high-value added activities in industry and services, and develops into the urban core of the North Sea, its growth accelerates. Similarly, when regions lose their edge in manufacturing or international trade and shipping, they stagnate – or even see their GDP per capita decline (as happened in Flanders after 1560). But these shifts within the larger regional unit of the North Sea region can to some extent be separated from the process of economic transformation that is happening in the region as a whole, which lead to a consistent growth of GDP per capita from the 1350s to the 1820s (as Figure 3 shows). 4 10.000 y = 921,03e0,0017x R² = 0,8737 1.000 100 England and Holland Three countries Exponentieel (England and Holland) Figure 3. Economic growth in the North Sea area, 1348-1820 Sources: see Figure 1. What is more, as indicated already, the relation between per capita GDP and population levels is also changing in time. This link is given in Figures 4 (England 12521870), 5 (Holland: 1348-1870) and 6 (the North Sea region as a whole: 1348-1870), which present, on the vertical axis, the log of GDP per capita, and on the horizontal axis, the population. In a Malthusian regime one expects a negative link between the two, which is evident in Figure 5 between 1252 and 1400. The escape from the Malthusian regime is in both cases very clear: there is a negative relation between levels of per capita GDP and population growth prior to 1400, whereas this disappears in England and the North Sea region, and becomes positive in the case of Holland. In England the shift towards a positive relation becomes visible after the 1650s – exactly when the centre of economic activity shifted from Holland to England. Our findings contradict research that used real wages as an indicator of income levels (e.g. Clark and Mills 2010, Nicolini 2007). Clark and Mills do, however, date the escape from the Malthusian confines at about 1650, but they found a negative relation before 1650. We conclude that the Malthusian link disappears in the 14th century, and that in particular when the North Sea region is seen as a whole, consistent growth began in the same period. 5 Figure 4. Per capita GDP and population: England, 1252 – 1870. Sources: Broadberry et al 2012. Figure 5. Per capita GDP and population: Holland, 1348 – 1870. 6 Sources: van Zanden and van Leeuwen 2012. Figure 6. Per capita GDP and population: North Sea region, 1348 – 1870 Sources: see figure 4 and 5 above. 3. Human capital formation North Sea region, 1300-1807/1900 In other papers we have explained how it was possible to estimate the long-term evolution of educational attainment in Holland (Van Zanden and Van Leeuwen 2012), and England (De Pleijt 2013). To summarize the latter paper: thanks to research carried out by Stone (1964, 1969), Cressy (1980), Stephens (1987), Moran (1985), and others that considered the development of English schooling and literacy, it is possible to quantify human capital formation: primary education (based on literacy rates) and secondary schooling (grammar and university enrolment) are combined and adjusted to population levels and rates of life expectancy to estimate annual education levels of men, women and the total population between 1300 and 1900. Combining the evidence offers a bridge between the different views with some interesting findings as shown in figure 7 and 8. Increases in average years of education were gradual: the basis of elementary schooling was led in the late 14th century, which paved the way for an ‘educational revolution’ in higher education after the 1530s (Stone 1964). Average English men spend about 2 years in education at the end of the 17th century. The ‘revolution’ was followed by a severe ‘depression’ that started in the 18th century. Many secondary schools disappeared and strong population growth reduced average years of higher education (it fell from 1 year in 1700 to 0.2 years by the 1870s). Women were not admitted to formal secondary schooling, but their basic skills increased remarkable after 1700, such that it alleviated for an overall decline in education levels during the Industrial Revolution. 7 Figure 7. Growth in human capital formation, England: 1307-1900. Source: De Pleijt 2013. Figure 8. Growth in primary and secondary schooling, England: 1307-1850. Notes and Source: De Pleijt 2013. Women were not admitted to formal secondary schooling. Series capturing the average years of education of women is therefore solely based on the development of literacy. The English economy displayed substantial dynamism and growth before the Industrial Revolution, which makes it appealing to examine the relationship between human 8 capital formation and economic growth more in-depth. To test this relationship, it was first of all necessary to run the Quandt-Andrews unknown breakpoint test (Quandt 1960, Andrews 1993), since the relation may change over time. The obtained regression results for per capita GDP show a break around 1808 (QLR statistic of 5.46), whereas the results for average years of education report on two breaks at about 1563/64 and 1806-1811 (QLR statistic of respectively 3.91 and 3.93).1 We thus identified two breakpoints, such that the regressions should be performed for each sub-period: i.e. 1307-1563, 1565-1806 and 1811-1900. Both breakpoints can be explained as growth starts to accelerate around 1810 and average years of education starts to grow exponentially due to increases in the attendance of elementary schooling (see figure 8). The break of the 1560s might be attributed to the growth of secondary schooling that occurs between 1530 and 1600, which causes a sharp increase in the average years of schooling of men as is shown in figure 8. Secondly, as depicted in figure 7 and 1, human capital formation and real per capita GDP exhibit an upward trend. This evolution of time-series is in line with the possibility that our variables are non-stationary and co-integrated, an observation that is confirmed by several tests (the appendix to this paper reports on the results of the augmented Dickey-Fuller tests and Johansen tests for cointegration). We estimated single equation Error Correction Models (ECMs) as proposed by Bardsen (1989) to explore the relationship between per capita income (lnYt) and human capital formation (lnEt): ∆lnYt = + 0∆lnEt - 1(lnYt-1 - 2 lnEt-1), where is the intercept, 0 captures the short run effect of human capital formation on economic growth, 2 estimates the long term effect that is dispersed across future time periods at a rate equal to 1. Regression Dep. Var. Sub-period 1 ∆lnYt 1307-1563 2 ∆lnYt 1565-1806 3 ∆lnYt 1811-1900 ∆lnEt 0.0928 (0.4247) -0.1783*** (0.0437) 0.1189*** (0.0052) 1.2697*** (0.3129) 0.1216 (0.2550) -0.2988*** (0.0478) 0.9537*** (0.0165) 2.1165*** (0.3376) -0.0379 (0.4107) -0.3707*** (0.0951) 0.6906*** (0.0072) 2.6514*** (0.6823) 255 0.0920 241 0.1578 89 0.2164 lnYt-1 lnEt-1 N R2 Table 1. Per capita GDP growth England, 1307-1900 Notes: Robust standard error in parentheses. *** Coefficient is significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level. Standard error of the long run multiplier is calculated according to the transformation proposed by Bewley (1979). The short run effects are insignificant in all sub-periods, indicating that there is no direct effect of human capital formation on per capita GDP growth. There are, however, long 1 The critical level used in the Quandt-Andrews unknown breakpoint tests is 3.66. 9 run effects (estimated by lnEt-1) of which the magnitude increases between 1307-1563 and 1565-1806, whilst the rate at which it is dispersed over future time periods increases (given by lnYt-1).2 Our results also indicate that the long effect becomes somewhat weaker when growth starts to accelerate after 1811. We applied Granger causality tests, since interest lies in the direction of causality as well. Table 2 reports on the results. Our findings confirm that human capital contributed to early modern growth in the period 1307-1806, whereas an endogenous system is found in the last sub-period (1811-1900). P-value 1307-1563 - Education causes growth - Growth causes education Yes No 0.000 0.594 1565-1806 - Education causes growth - Growth causes education Yes No 0.000 0.980 1811-1900 - Education causes growth - Growth causes education Yes Yes 0.000 0.000 Table 2. Granger causality test England, 1307-1900 Notes: Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). As clear from figures 7 and 8, it is possible to differentiate between levels of education as we have series covering average years of primary schooling (EducPrim) and average years of secondary and tertiary schooling (EducHigh). A further distinction can be made according to gender (women and men), although it should be stressed that the series of women is solely based on the development of literacy (women were not admitted to formal secondary schooling before the late 19th century) such that it only measures reading and writing abilities. Several Quandt-Andrews unknown breakpoint tests have been conducted that indicate a structural break in the dataset of women around 1418 (QLR statistic of 5.168) and 1810 (4.561), whereas the breakpoints in the series capturing primary skills of men are similar to those reported above – i.e. 1563/64 (QLR statistic of 3.704) and 1810 (QLR statistic of 5.234). With respect to average years of higher education, we find two different breakpoints: 1562/73 (probably caused by the sharp increases in grammar schooling) and 1700 (decay of higher education) (QLR statistics of respectively 4.721 and 17.01). The variables are again non-stationary and co-integrated (results reported in the appendix), which makes it necessary to estimate our regressions with single equation ECMs of which the results are given in tables 3 (women) and 4 (men). Regression Dep. Var. Sub-period Proxy 4 ∆lnYt 1418-1806 Women 5 ∆lnYt 1811-1900 Women ∆lnYt 1565-1806 Women ∆lnEt 0.1044 (0.1721) -0.0926*** (0.0226) 0.1012 (0.2109) -0.2262** (0.0810) -0.0129 (0.1628) -0.2091*** (0.0406) lnYt-1 2 6 Our results remain equal when we standardize the coefficients. 10 lnEt-1 N R2 0.1862*** (0.0030) 0.6888*** (0.1691) 0.5805*** (0.0065) 1.6688*** (0.5973) 0.4021*** (0.0073) 1.5845*** (0.3069) 387 0.0476 89 0.1498 241 0.1108 Table 3. Per capita GDP growth England, 1307-1900 Notes: Robust standard error in parentheses. *** Coefficient is significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level. Standard error of the long run multiplier is calculated according to the transformation proposed by Bewley (1979). Regression Dep. Var. Sub-period Proxy 7 ∆lnYt 1307-1563 MenPrim 8 ∆lnYt 1565-1806 MenPrim 9 ∆lnYt 1811-1900 MenPrim 10 ∆lnYt 1307-1562 MenHigh 11 ∆lnYt 1564-1699 MenHigh ∆lnEt -0.1245 (0.3084) -0.1621** (0.0413) 0.1157*** (0.0053) 1.1554*** (0.2975) 0.0876 (0.1816) -0.2596*** (0.0444) 0.6889*** (0.0123) 1.8761*** (0.3201) -0.1152 (0.4350) -0.3623*** (0.0803) 0.7569*** (0.0079) 2.5543*** (0.5681) -0.0153 (0.2843) -0.2030*** (0.0470) 0.1247*** (0.0048) 1.4498*** (0.3381) 0.9416 (0.6488) -0.4476*** (0.0707) 1.7179*** (0.0962) 3.2617*** (0.5138) 255 0.0827 241 0.1324 89 0.2174 254 0.1021 134 0.2430 lnYt-1 lnEt-1 N R2 Table 4. Per capita GDP growth England, 1307-1900 Notes: Robust standard error in parentheses. *** Coefficient is significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level. Standard error of the long run multiplier is calculated according to the transformation proposed by Bewley (1979). Equation 6 is added to table 3 to capture the sub-period 1565-1806. There is no breakpoint in the series of women around 1564, but this makes it possible to compare the regression results with that of men (equations 7 – 9). For all measures of human capital and sub-periods, there are no short run effects. Long run effects are significant and increase in magnitude over time, whilst the effect of human capital formation of women is slightly below that of men as shown in equations 5 and 6 of table 3 and equations 7-9 of Table 4. The obtained results indicate a stronger long-term effect of advanced skills on per capita GDP than do basic skills when differentiation between average years of primary education (MenPrim) and higher education (MenHigh) (equation 11). Interestingly, the long-run impact of higher education disappears after 1700 due to the movement away from secondary schooling, which points to a process of deskilling during the first phase of the industrial revolution that is mentioned in the literature (Mitch 1992 and Mokyr 2001). Granger causality tests for the sub-periods and different indicators of human capital formation are given in Table 5. The results are equivalent to those reported above: education granger causes per capita income – it does not work in the opposite direction before 1811; the process becomes endogenous after 1811, confirming established views. P-value Women 1418-1806 11 - Education causes growth - Growth causes education 1811-1900 - Education causes growth - Growth causes education MenPrim 1307-1563 - Education causes growth - Growth causes education 1565-1806 - Education causes growth - Growth causes education 1811-1900 - Education causes growth - Growth causes education MenHigh 1307-1562 - Education causes growth - Growth causes education 1354-1699 - Education causes growth - Growth causes education Yes No 0.000 0.581 Yes Yes 0.000 0.002 Yes No 0.001 0.251 Yes No 0.000 0.379 Yes Yes 0.000 0.011 Yes No 0.000 0.327 Yes No 0.000 0.572 Table 5. Granger causality test England, 1307-1900 Notes: Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). Figure 9 depicts the growth in human capital for Holland and the North Sea region. The human capital series of Holland only goes back to the 1560s (and we are working on further refining the results and projecting the series back into the Middle Ages). For Holland and the North Sea region it is therefore only possible to perform regressions for the subperiod 1564-1807). No structural breaks were found in this GDP series (see also Van Zanden and Van Leeuwen 2012), although our test results show that our variables are non-stationary and co-integrated (results reported in the appendix). We therefore estimated ECMs of which the results are shown in table 3. 12 Figure 9. Growth in human capital formation, Holland and North Sea region: 1565-1806. Notes and Source: De Pleijt 2013. Women were not admitted to formal secondary schooling. Average years of education of women are therefore solely based on the development of literacy. Regression Dep. Var. Sub-period Region 12 ∆lnYt 1565-1806 Holland 13 ∆lnYt 1565-1806 North Sea ∆lnEt -2.5083* (1.2941) -0.3778*** (0.0582) 0.1931*** (0.0212) 2.1923*** (0.4495) 0.1555 (0.2249) -0.3990*** (0.0594) 0.7131*** (0.0121) 2.8835*** (0.4284) 243 0.1923 243 0.2104 lnYt-1 lnEt-1 N R2 Table 6. Error Correction Models for per capita GDP growth Holland and the North Sea area, 1565-1806 Notes: Robust standard error in parentheses. *** Coefficient is significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level. Standard error of the long run multiplier is calculated according to the transformation proposed by Bewley (1979). Our regression results report again on a long run effect of human capital formation on growth, but the coefficient is much smaller than in the English case (Table 3). Finally we tested for Granger causality for Holland and the North Sea region. In Holland we find that education causes growth and growth causes education; this pattern of endogenous growth may be linked to Holland as being ‘the first modern economy’ (Vries and van der Woude 1997). 13 The North Sea region as a whole only shows education causing growth (as in the English case) (Table 4). P-value Holland: - Education causes growth - Growth causes education Yes Yes 0.000 0.000 North Sea: - Education causes growth - Growth causes education Yes No 0.000 0.296 Table 7. Granger causality test Holland and North Sea region, 1564-1807 Notes: Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). Our regression results on long-term per capita income growth and human capital formation contradict the earlier conclusion of Mitch (1993), Allen (2003) and Mokyr (2001) that points towards an insignificant relationship between education and early modern growth. These studies focus on literacy rates as a proxy of human capital formation that suffers from the disadvantage of measuring only very basic skills (reading and writing abilities), where these analyses overlooked the growth in (higher) education before 1750. The regression results reported in this section are similar to recent studies that use estimates for more advanced skills, as these find a positive relation between development and human capital formation (book production and consumption: Baten and van Zanden 2008; secondary schooling: Boucekkine et al 2007). We furthermore find that basic skills affected preindustrial growth, which opposes the study of Goldin and Katz (1998) as they find that basic skills such as reading and writing abilities were irrelevant before the 20th century. A second finding is that progressive skills, as measured by average years of higher education, contributed even more to economic growth than basic skills, although there is a clear movement away from higher education on the eve of the Industrial Revolution (setting in after 1700): literacy rates of men stagnated and there was a remarkable decline in average years of higher education. It was not until the late 19th century that English education levels started to rise again, which, against the view of Unified Growth Theory, was mainly the result of government polices implemented in the mid-19th century. A third finding, aligned with Modern Growth Theory (e.g. Lucas 1989, Romer 1990, Galor and Moav 2002), is that growth becomes endogenous during the second Industrial Revolution since our Granger causality tests suggests that education contributed to technological progress, whereas per capita GDP growth fostered further growth in human capital formation. In contradiction to these theories, however, and as stressed above, our results show that human capital formation already contributed to the process of development long before the second Industrial Revolution. 4. Conclusion This paper considered the ‘stylized facts’ about patterns of long-term economic growth in the North Sea region between 1300 and 1900, thereby focussing on demographic change, human capital and economic growth. The evidence presented in this paper suggests that Holland and England managed to escape the Malthusian confines long before the epoch of Modern Growth. The negative relationship between population levels and per capita GDP disappears after the Black Death, and is transformed in a positive link, in particular in Holland and the North Sea area as a whole. In explaining the transition from a Malthusian economy to modern growth we therefore have to focus on two transformations: the late Middle Ages, and the early 14 19th century. The first break saw the emergence of sustained by slow growth in a small part of Western Europe (England, Holland and Flanders), the second break – the ‘traditional’ Industrial Revolution – saw the transition towards rapid growth in a much larger (and rapidly expanding) part of the Continent. We have studied the links between these various regimes of growth and human capital formation. Several regression analyses have been performed that measure the magnitude of human capital formation (as measured by average years of education) on economic growth in the North Sea region. With respect to England, it was required to divide our analysis in three sub-periods (i.e. 1307-1563, 1565-1806 and 1811-1900) since the relation between human capital formation and growth changes over time. The obtained regression results show that there was a long run effect of human capital formation on growth that increases in magnitude between 1307 and 1811. This is an important result: education did already matter for growth before the Industrial Revolution (contra: Mokyr and Allen), and we cannot explain the continuous growth that occurred in this region since the 14th century without taking this effect into account. In Holland we already see the emergence of endogenous growth (which is a feature of the post 1810 period). For England and the North Sea region as a whole it was only possible to estimate the relationship for the sub-period 1565-1806, of which our regression analysis indicate a long run effect of human capital formation on economic growth. Our evidence suggests that growth was endogenous in Holland between 1565 and 1806 but not in the region as a whole. APPENDIX In order to examine the time-series properties of the data, we performed several augmented Dickey-Fuller and Johansen tests for co-integration. The results of these tests are reported in this section. lnYt lnEt ∆lnYt ∆lnEt 1307-1563 -2.254 (-3.430) 0.338 (-3.430) -10.074 (-3.430) -6.886 (-3.430) 1565-1806 -0.667 (-3.431) -3.035 (-3.431) -11.711 (-3.431) -10.091 (-3.431) 1811-1900 1.024 (-3.461) -0.887 (-3.461) -8.417 (-3.461) -4.925 (-3.461) Table 8. Dickey-Fuller test results equations 1-3 (England) Notes: Critical values (at 5% level) between parentheses. Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). Trace statistic 1307-1563 15.66 1565-1806 41.68 1811-1900 33.55 Table 9. Johansen tests for co-integration, equations 1-3 (England) Notes: Critical value is 15.41. Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). lnYt lnEt ∆lnYt ∆lnEt 15 Women: 1418-1806 1564-1806 -2.593 (-3.425) - 1811-1900 - MenPrim: 1307-1563 - 1564-1806 - 1811-1900 - MenHigh: 1307-1562 1564-1699 -2.214 (-3.430) -3.377 (-3.445) -2.018 (-3.425) -3.179 (-3.431) 0.194 (-3.461) -15.847 (-3.425) 0.317 (-3.430) -2.333 (-3.431) -0.262 (-3.461) - -1.093 (-3.430) -0.942 (-3.445) -9.890 (-3.430) -15.255 (-3.445) - - -11.447 (-3.425) -10.701 (-3.431) -7.562 (-3.461) -4.668 (-3.430) -8.920 (-3.431) -9.241 (-3.461) -5.957 (-3.430) -12.268 (-3.445) Table 10. Dickey-Fuller test results equations 4-11 (England) Notes: Critical values (at 5% level) between parentheses. Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). Test results of per capita income only reported when sub-period under consideration deviates from those given in table 8. Trace statistic Women 1418-1806 1564-1806 1811-1900 19.07 28.95 23.56 MenPrim 1307-1563 1565-1806 1811-1900 21.33 35.88 28.33 MenHigh 1307-1562 1564-1699 29.24 35.38 Table 11. Johansen tests for co-integration, equations 4-11 (England) Notes: Critical value is 15.41. Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). lnYt lnEt ∆lnYt ∆lnEt Holland -3.368 (-3.431) -1.252 (-3.431) -9.503 (-3.431) -5.809 (-3.431) North Sea -2.449 (-3.431) -2.001 (-3.431) -10.413 (-3.431) -9.614 (-3.431) Table 12. Dickey-Fuller test results equations 12 and 13 (Holland and North Sea region) 16 Notes: Critical values (at 5% level) between parentheses. Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). Trace statistic Holland 37.08 North Sea 57.51 Table 13. Johansen tests for co-integration, equations 12 and 13 (Holland and North Sea region) Notes: Critical value is 15.41. Lag-order determined with the Schwarz’s Bayesian information criterion (SBIC) and the Hannan and Quinn information criterion (HQIC). 17