Time for Growth

Severgnini, Battista; Boerner, Lars Working Paper Time for Growth Working paper, No. 4-2019 Provided in Cooperation with: Department of Economics, Copenhagen Business School (CBS) Suggested Citation: Severgnini, Battista; Boerner, Lars (2019) : Time for Growth, Working paper, No. 4-2019, Copenhagen Business School (CBS), Department of Economics, Frederiksberg, https://hdl.handle.net/10398/9730 This Version is available at: https://hdl.handle.net/10419/208595 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/ Department of Economics Copenhagen Business School Working paper 04-2019 Department of Economics – Porcelænshaven 16A, 1. DK-2000 Frederiksberg Time for Growth Lars Boerner Battista Severgnini Time for Growth∗ Lars Boerner Martin Luther University of Halle-Wittenberg DAFM, King’s College London Battista Severgnini Copenhagen Business School March 28, 2019 Abstract This paper studies the impact of the early adoption of one of the most important high-technology machines in history, the public mechanical clock, on long-run growth in Europe. We avoid endogeneity by considering the relationship between the adoption of clocks with an instrument based on the appearance of repeated solar eclipses. This is motivated by the predecessor technologies of mechanical clocks, astronomic instruments that measured the course of heavenly bodies. We find a significant increase in growth rates between 1500 and 1700 in the range of 30 percentage points in early adopter cities and areas. Finally, additional quantitative analysis suggests a positive relationship between mechanical clocks and contemporary long-term orientation nowadays. Keywords: technological adoption, cities, mechanical clocks, information technology, long-term orientation JEL classification: 033, N13, N93. ∗We would like to thank Marcus Asplund, Zachary Bleemer, Steve Broadberry, Michael C. Burda, Francesco Caselli, Alessandra Cassar, Carl-Johan Dalgaard, Paul David, Brad DeLong, Barry Eichengreen, James Fenske, Oded Galor, Regina Grafe, Avner Greif, Andrea Ichino, Andreas Irmen, Saumitra Jha, Robert Margo, Kris Mitchener, Joel Mokyr, Enrico Moretti, Nathan Nunn, Vanessa Ogle, Martha Olney, Albrecht Ritschl, Jared Rubin, Peter Sandholt Jensen, Paul Sharp, Chad Syverson, Otto Toivanen, Harald Uhlig, Juuso V¨alim¨aki, David Weil, Mirko Wiederholt, Niko Wolf, Cristiano Zanetti as well as seminar participants at UC Berkeley, Stanford University, University of San Francisco, King’s College London, London School of Economics, European University Institute, Humboldt University Berlin, Copenhagen University, Copenhagen Business School, HECER Helsinki, University of Cambridge, AEA Meeting in San Francisco, RES Meeting in Sussex, EEA Congress in Geneva, the 6th Annual Workshop on Growth, History, and Development in Odense, EHES in Pisa, the SFB Conference in Berlin, the 2nd CREA Workshop on Aging, Culture and Comparative Development in Luxembourg, and the Fifth CEPR Economic History Symposium in London. A special thank goes to Gerhard Dohrn-van Rossum for sharing his data and expertise. This project was partially supported by Otto Mønsted Foundation. Battista Severgnini thanks the Department of Economics at UC Berkeley for the hospitality during a revision of the paper. All errors are our own. Lars Boerner: DAFM, King’s Business School, King’s College London. Martin-Luther University Halle-Wittenberg. Address: Bush House 30 Aldwych London WC2B 4BG UK. Email: [email protected]. Battista Severgnini: Department of Economics. Copenhagen Business School. Address: Porcelænshaven 16A. DK - 2000 Frederiksberg, Denmark. Email: [email protected]. 1 1 Introduction This paper investigates the impact of the early adoption of high-technology machines on long-run growth. Various studies have widely discussed the possible benefits and drawbacks of the role of hightechnology innovation on firms and nations economic success. On the one hand, a well-established literature by various scholars (e.g., Ricardo (1821), Leontieff (1983), Samuelson (1988), and Zeira (1998)) has claimed that the impact is negative because advanced machines lower wages, which in turn reduce population and income growth. On the other hand, in the past twenty years, which coincide with the introduction of new innovations in information technology (henceforth, IT), new research has found a more differentiated picture of this relationship. In a 1987 article published in the New York Times, Robert Solow underlined a productivity paradox (also known as the Solow Paradox), which highlights that the American productivity slowdown in the 1970s concurs with the adoption of computers (”You can see the computer age everywhere but in the productivity statistics.”). However, other scholars have found the advantage and the positive effects of the spread of computers on society: Bresnahan et al. (2002) underline the positive role of high tech capital and the complementarity with skills and innovations at the firm level; Caselli and Coleman (2001) use country data and find a strong and positive relationship between human capital, computers and productivity; Andersen et al. (2012), examine the negative role played by lightning in IT diffusion and explain the higher economic growth across American states due to digital technologies. In contrast, more recently Acemoglu et al. (2014) confirm a Solow Paradox in IT-intensive sectors, where an increase in labor productivity is associated with a decline in employment. Some main problems with these types of studies is that they have to address several empirical challenges. First, it is difficult to identify the adoption of IT at the micro level and to create a representative aggregate picture at the macro level. Second, the identification of adoption does not necessarily guarantee the accurate use of the new technology. Finally, the time series for potentially identifying growth are relatively short. To find an answer to the question concerning the relationship between technology and economic performances, case studies based on the introduction of innovative machines can be useful. In an early reply to Solow, the economist and economic historian Paul David (1990 and 1991) suggests resolving the study of the Solow Paradox from a historical perspective. Examining the innovation of the dynamo in the late 19th century he argues that it simply takes time until the use of such 2 a general purpose technology (GPT, henceforth)1affects economic growth rates. As described by several studies,2gains in efficiency due to new technologies can increase over time because of the gradual replacement of the old technology with the new one and the intergenerational learning-by- doing process of the workers. Crafts (2002), among other scholars, took up this line of argumentation and compared the impact of different GPTs, such as electricity and computers, on long-run economic growth. He finds comparably strong evidence for the effect of IT, but admits that there are problems in measuring and comparing such effects adequately. A more recent study related to the different impact of technologies on American growth dynamics after the Civil War has been done by Gordon (2016). Going further back in time, Dittmar (2011) finds that the invention and diffusion of the printing press during the 15th century had a long-run growth effect (between 1500 and 1800) at the city level, since the possibility to print media, which increased the dissemination of ideas, facilitated human capital accumulation and helped business practices. These findings by Dittmar (2011) can be related to a broader process of social and cultural change observed by Crosby (1997), who relates medieval and premodern technological innovations to a change in the perception of the world from a qualitative to a quantitative perception. Such a claim can already be found in Le Goff (1971), who identifies a changing environment during the late Middle Ages in ”atmosphere of calculation”. However whereas Le Goff attributes such a change already to the late Middle Ages, Crosby emphasizes the long-run process starting during the late Middle Ages and only fully accelerated by the end of the 16th century. Important is that the technological change triggers and coevolves with cultural and social change. Only the development of both might enfold growth and development of a society. In this paper, we attempt to shed light on potential productivity paradoxes from a new perspective. We study the impact of one of the most important technologies ever invented in history, i.e., the public mechanical clock, on economic growth. This technology was first introduced in Europe at the end of the 13th century, and it spread across Europe during the subsequent two centuries. Mechanical clocks have been identified as one of the greatest technological inventions of the last millennium.3 1Bresnahan and Trajtenberg (1995) define GPTs according to three characteristics: first, they should be pervasive in most economic sectors; second, they should improve over time with lower costs for the consumers; finally, they should incentivize new products and processes. 2See, for example, Devine (1983), Atkeson and Kehoe (2007), and, for a summary, Weil (2013)’s textbook on economic growth. 3The public mechanical clock fulfills many of the attributes described by Bresnahan and Trajtenberg (1995) as previously mentioned. For an identification of the clocks as a GPT see Comin et al. (2010) and Weil (2013). However, 3 The importance of mechanical clocks has been discussed by several scholars in different fields. Landes (1983) claims that clocks were the technological sensation of the 14th century, which is similar to computers today. Furthermore, he argues that the clock had a strong impact on productivity: it enabled increases in organizational skills in terms of the coordination and division of labor and the monitoring of production processes. Very much in line with Landes, Mokyr (1992) argues that the mechanical clocks was one of the most important technology inventions of the last millennium. Moreover, Thompson (1967) highlights that the mechanical clock overlapped with a changed work culture and increased work discipline. Le Goff (1982) claims that the introduction of the public mechanical clock was a turning point for the Western society. It helped create a new epoch, ”the time of the merchants”, because it enabled business people to better frame and measure all types of economic activities in a timely manner. In addition to the already discussed points, Dohrn-van Rossum (1996) finds evidence for the improvement of various coordination activities in premodern towns such as market times, administrative meetings of the town governments, and school and university lecturing times. Other economic historians with a greater focus on the transition to modernity, e.g., Mumford (1934), Rosenberg and Birdzell (2008) and Voth (2001), argue that the clock had a profound impact on the processes of the Industrial Revolution. Mumford even describes the mechanical clock, and not the steam machine, as the key machine of the the modern industrial age because the knowledge accumulated from the mechanical clock had a positive spillover during the Industrial Revolution. Furthermore and more generally, prominent social scientists such as Marx (1863), Weber (1905) and Sombart (1921) claim that clocks had a fundamental impact on the evolution of capitalism and the rationality of societies. Most of these studies have in common that they emphasize the strongly time-lagged impact of the clock, i.e., the technology took several centuries to develop any economic (or cultural) influence on societies. They claim that the learning and use of new forms organization and coordination of production related to the clock could only develop embedded in a change in work culture, for instance, in the form of punctuality and discipline. Furthermore, the introduction of clocks was not related to any economic needs but was a sign of prestige and progressiveness, thus showing one to be at the technological frontier. Therefore, the clock indeed would be an excellent investigation subject the clock differs since it misses the technical complementarities to other technologies, but remained a stand-alone technique. However, it did have the important effect of creating a new high-skill and precision industry (Mokyr (1992)). 4 for testing the Solow Paradox. In addition, we would not have any issues with endogeneity when considering the direction of causality, i.e., that economic growth would not spur the introduction of public mechanical clocks. Both claims have been derived based on qualitative studies and thus need rigorous quantitative analysis. Thus, examining this enumeration of findings and claims made by this large number of scholars we should find some long-run growth effects based on changes in organization, production, and work culture. This means that the introduction of clocks in medieval cities should have localized spillover effects in these towns and further affect growth at a more aggregate level. To test the impact of clocks on economic growth, we construct a unique dataset collected from several historical sources. To study economic growth, we use the dataset of European city sizes collected by Bairoch et al. (1988). We use population size over time as a proxy of cities’ welfare to study the trajectories of premodern economic growth (Acemoglu et al. (2005b)). Our dataset contains all cities for which population was recorded from 1000 to 1850. Our central explanatory variable is the information we have collected on the construction of public mechanical clocks in these cities. We identify a group of early adopting cities from the first adopters in 1283 until 1450. The end point is calculated based on the inflection point and hazard rate of the S-shaped diffusion curve of mechanical clocks. Based on this identification we measure the impact of the implementation of clocks by early adopters compared to other cities. We study population size from 1000 until 1850 and control for a broad set of variables, including the institutional and geographical characteristics of cities and regions. Our dataset has several favorable characteristics and is not affected by sample selection because the construction of clocks is well documented in the available source material. Therefore, we have a representative and rather complete sample on the date of the adoption and geographical location of this new technology. Clocks were affordable and had relatively low maintenance costs (as we will discuss later). Thus, once installed, clocks were used and maintained by the city population over many centuries. Furthermore, clocks were nonexclusive public goods that were easy to understand and use by an entire city’s population. Finally, our study allows us to have a very long-run perspective on the effect of this technology on economic growth. Our empirical strategy includes various standard approaches used in the related empirical literature. We consider the estimation of a standard differences-in-difference regression as benchmark. We also address endogeneity problems between the size of the city and the adoption of clocks, missing explanatory variables, and measurement errors of the main explanatory variables by employing an 5 instrumental variable approach based on the presence of total and annular eclipses. We exploit data constructed by the National Aeronautics and Space Administration (NASA, henceforth) for detecting populated geographical area intensively covered by solar eclipses before the adoption of the first clocks. The use of solar eclipses as an instrument for clocks is motivated by two types of observations by science and technology historians: first, eclipses and other astronomic movements created curiosity in societies, influenced the study of astronomy and triggered the construction of mechanical devices that aided in measuring these astronomic events, such as astrolabes and specially designed water and sun clocks (Turner (1911) and Dohrn-van Rossum (1996)); second, these machines have been identified as the predecessor technologies of mechanical clocks. Finally, we consider different robustness of our results. In particular, placebo tests based on different sets of past (i.e., from 2000 B.C. to 800 A.D.) and future (i.e., from 1450 A.D. to 3000 A.D.) eclipses provide additional support on the connection between the eclipses and the adoption of the clocks. Following the proposed methodologies, we find that earlier adopters, compared to other cities, displayed a significant increase in population growth during the period of 1500-1700. These robust results indicate that the new emerging technology of public mechanical clocks indeed localized spillover effects on various social and economy-supporting activities and led to higher city growth rates in the long-run until new substitute technologies (such as the portable watch and the pendulum) were introduced in the 18th century. To further substantiate our findings, we make some more extensive robustness checks. First, we investigate whether the introduction of a subsequent technology in the form of the printing press affects our results. However, both technologies have a strong effect and our previous results do not change. Second, using an alternative instrument, in the form of the distance to the first innovators, provides similar results. Third, we test the relationship between the adoption of the public mechanical clock and the population exploiting a propensity score matching technique without any alterations of the results. Fourth, we test whether our results are affected by particular subsets of the dataset considered. We run an entropy test (Hainmueller and Xu (2013)), and in this, the results are also consistent. Finally, we exploit the time of adoption as additional information for our estimation. We run an event-study regression analysis (Autor (2003)) and find persistent effects over several centuries after the introduction of the clock. Complementary to this city-level analysis we do a robustness check on the macro level. We use the penetration rate of the new technology on the country-level to estimate the GDP-growth rate of 6 a country. As an instrument, we use the share of the population covered by solar eclipses. Econometrically, we follow the methodology by Czernich et al. (2011) and find it again very similar to the microlevel. Finally, we also compare our data on the diffusion of public mechanical clocks on contemporary cultural norms related to people’s attitudes toward time. Studies in psychology (e.g., Levine (1998) and Levine and Norenzayan (1999)) underline how different paces of life are intertwined with biological and city’s rhythm (Bettencourt et al. (2007)). Furthermore, we consider whether public mechanical clocks can influence long-term orientation, a value which is a relevant driver for conducting business and for the social life in general (Hofstede et al. (2010)). Simple correlations based on country-level data suggests a positive link between the early adoption of mechanical clocks and proxies for the pace of life. Very similar results are obtained once long-term orientation is studied. We also consider the regression framework introduced by Galor and ¨ Ozak (2016) and exploit a dataset based on secondgeneration migrants from the European Social Survey, finding that mechanical clocks can have a positive long-run legacy for explaining long-term orientation. Our findings contribute to the literature in the following ways. First, public mechanical clocks had a positive impact on economic growth and development. This is in line with the previously mentioned large body of literature that derives this conclusion based on qualitative studies on changes in economic institutions and organizations. For instance, Mokyr (1992) writes that mechanical clocks created order, organization, and a shared set of objective information. This improved the measurement of productivity, increased the efficiency among workers, and greatly affected other sectors. Our estimates indeed give evidence for higher economic output based on such changes. Our empirical results suggest a causal relationship between 1500-1700 when controlling for potential econometric estimation biases. During earlier periods closer to the first implementation of clocks, no such causal relationship can be found. This supports the claims have been made by the scholars who determined that the construction of clocks was not motivated by any economic needs. Second, our results shed light on the role of technology for economic development and growth before the Industrial Revolution. Our results provide evidence on the quantitative impact of technological change that was triggered by the upper-tail of human capital (Mokyr (2002 and 2005)) and Squicciarini and Voigtl¨ander (2015)) that occurred well before the scientific revolution (Long (2011), Zilsel (2011), and Zanetti (2017)). The first clocks had been constructed by the so-called Vitruvian artisans, who had various backgrounds 7 Figure 3: City size and the adoption of the public mechanical clock 8.5 9 9.5 10 10.5 City size 1000 1200 1400 1600 1800 year Unweighted average of cities’ population with 95% confidence interval. The red line represents the adopters, the blue displays the non adopters of the mechanical clocks. Source: Authors’ calculations based on the authors’ dataset of clocks. Cities available in the Bairoch et al. (1988)’s dataset and with at least three consecutive observations during the adoption period. made the clock valuable and could result in higher economic growth rates. Answering the first question is rather straightforward. The clock offered an accessible and audible signal that divided the day into equally long units (Landes (1983)). To answer the second question a more sophisticated answer is needed. The concept of dividing the day into measurable subunits existed before the clock (Lippincott et al. (1999)). The division of the day into twelve parts dates back to ancient times. However, the length of the hour depended on the length of the day and was a fixed proportion of the sunlight hours (the so-called temporal hours). Thus, an hour could vary during the summer and winter periods. In line with this concept, the hour could be measured by sun clocks. However, because this measurement technique depends on time of the year and the weather conditions, the length of the hour varied and the technology was less reliable. Therefore, it was not intensively used in societies. Rather, people followed simpler indications such as the position of the 14 sun, i.e., the sunrise, noon, and sunset, as guidelines. The concept of the division into twenty-four equally long hours (the so-called equinoctial hours) also dates back to ancient times. However, it was rather complicated to measure and could not be directly derived from the constellation of the sun, as it had to be derived by calculations. This division of time was mainly used to follow the course of the heavenly bodies. Astronomic instruments, so-called astrolabes, or specially calibrated water clocks were employed to measure these activities. The use of astrolabes to measure daily time was overly complicated, and the use of water clocks required additional calculations and a very precise calibration of the clock.10 Therefore, the introduction of the mechanical clock improved the quality of time keeping dramatically.11 The use of bells as signals existed before the introduction of mechanical clocks (Dohrn-van Rossum (1996), chapter 7). In late medieval cities, it became popular to indicate and coordinate all types of social and economic activities with various bells, fanfares or flags. These signals were approved by the city government and were used for specific tasks and groups of people. Therefore, what was new with respect to the public mechanical clock was the introduction of a regular, repetitive, precise, and common signal for urban society that could be used for all types of signaling purposes. In this way, the multiplicity of signals, which in some cities reached their limits by the late Middle Ages and created chaos rather than order, could be replaced by one abstract signal. Finally, the remaining question that needs to be answered is how the clock affected the daily life of the population and was transformed into higher economic output. Clocks had an effect on the organization and coordination of daily life activities with respect to economic, administrative and educational tasks. There exists evidence from the 15th century onwards that the public clocks were used to coordinate such activities in many cities (Dohrn-van Rossum (1996)).12 The organization of markets neatly documents this change. Whereas prior to public clocks, the market time typically started with sunrise and ended at noon, with the introduction of clocks, market times were determined by the stroke of the hour. Furthermore, market time was shortened and market access was granted to dif- 10The use of water clocks can only be documented in a few sources in ancient Europe. There are references for ancient Rome, but the clock was likely calibrated based on the length of the day. Furthermore, we have some evidence that such mechanisms potentially existed in the neighboring Arabic world in the form of water clocks, which produced regular repeating sounds. However, according to the source material available, these machines were rather automates for entertainment and admiration. Finally, it is documented that medieval monasteries used water clocks. 11For a more detailed discussion of the quantification of this improvement, see Cipolla (1967). 12H¨arter (2007), in an extensive case study on on regulations in German Imperial cities and territories, finds a massive increase in regulations on market starting time, in particular, during the 16th century. 15 ferent groups of people at different times. For instance, market regulations offered time-differentiated access to consumers, retailers, and wholesalers, and in some cases, a differentiation between foreigners and locals or religious groups was made. Furthermore, we find evidence for the tight organization of administrative meetings of town officials following the signals of the public clock. Finally, schools and universities began to use the public clocks to determine the starting and ending times of lectures.13 There were several potential economic benefits of such improvements of organization and coordination. First, the precise public indication of time reduced the urban populations search, match, and waiting time. This enabled people to better plan activities and have more time to do other tasks. In this sense, the clock is an information technology that improves coordination and reduces transaction time, as discussed by Hayek (1945). However, it could also improve the coordinating task itself. The concentration of the market time created thicker markets and could improve the allocation quality between the demand and supply sides. The precise separation of different groups of buyers and sellers allowed towns to create a more powerful market policy. Towns particularly intended to avoid commodity hoarding and speculation. A precise meeting time in markets, town halls, or educational institutions could create better human capital spillover effects within meeting groups. The division of education into single hours allowed for the creation of schedules with alternating easier and more difficult subjects, making the learning experience more productive.14 These effects on coordination and organization should be considered in a long-term perspective. If it is true that other types of time measurement devices (e.g., sun and water clocks) can be used for similar purpose of coordination, then the mechanical clock helped to quantify the strikes. According to the historical evidence related to the city of Paris at the beginning of the 14th century collected by Crombie (1961), the mechanical clock allowed for thinking about the time as an abstraction to be quantified during the contemporary life. This effect is also confirmed by studies in physics (Einstein and Infeld (1938)) and neurobiology (Gibbon et al. (1997)), which show that the presence of the mechanical device gradually helps individuals in conforming their personal subjective time with a standard one (”government time”). 13As remarked by Thompson (1967), during the appearance of the mechanical clocks, it was possible to observe a contemporaneous internalization of time discipline during the working hours as opposed by the rhythm of nature (”task time”). However, this thesis has been recently challenged by recent historiographical research. See Ogle (2015) for a review. 14The relevance of time coordination and synchronization in working places has been highlighted from a theoretical point of view by Weiss (1996), while Hamermesh et al. (2008) show that timing altered by daylight, different television schedules and time zones affect temporal working coordination and sleeping time with effects on economic activities. 16 Furthermore, clocks enabled the coordination and monitoring of production activities (Landes (1983) and Dohrn-van Rossum (1996)). The public clock created an ”objective” measurement for the employer and employee or any cooperative group of productive agents. Whereas the use of church clocks and special work clocks as signaling devices had previously been used to determine the starting and ending time of the day for specific working groups, the public mechanical clock could now precisely measure the working time and breaks and enable payment by the hour (and, often, ”wage punishment by the hour” when workers did not show up on time) and payment for overtime hours.15 The public mechanical clock was introduced particularly for simple tasks, for which monitoring and payment by the hour were meaningful. For instance, Landes argues that it must have been particularly useful in the booming textile ”industry” of the time. Another well-documented sector was large construction sites, such as those for domes or cathedrals, where many workers had to be coordinated and monitored at the same time (Dohrn-van Rossum (1996)). More differentiated uses of time to precisely define and synchronize work tasks evolved over time.16 Impressive evidence of the perfection to synchronize work tasks can be found fin the late 17th century law book of the ”Crowley Ironworks”, the biggest ironwork in Europe at that time, which further illustrates this development (Thompson (1967)). However, in this case, the use of the clock as a control function did not automatically translate into measures to increase productivity. In a dispute between different guilds in late 14th century France, the public clock was used to coordinate working activities to restrict working time to limit the amount of output produced and create less competition (Fagniez (1877)). Similar evidence can be found for 15th century North Germany, where the restriction of working time ordered by the guilds limited the amount of output and increased the price for products (Wulf (1991)). The implementation of organizational regulations in the form of monitoring and the coordination of the labor force to increase productivity only appeared only over time. These guild examples show that the use of the clock as a productivity supporting device is related to 15To the best of our knowledge, we do not have time series that allow us to test the impact of the public mechanical clocks on wages and, as underlined by Munro (2005), the information whether wages were remunerated either by piecework or based on the time worked is missing. A careful reader might find a potential contradiction with both theoretical and empirical studies in personnel economics, where the switch from hourly wage to a piece rate compensation increases productivity. However, as highlighted by Lazear (1996), this relationship holds if the group of workers paid by piece-rates are a self-selected group that is likely to have a higher productivity than the average worker. Given the historical evidence from Landes (1983), this does not seem the case. He suggested that the introduction of the time signal of the clock did select skilled workers and increased the pressure of unskilled workers, being one of the causes of the historical episode of the Revolt of Ciompi (1378-82) in Florence. 16 A detailed analysis of the construction sector for instance documents changes in regulations in North German cities starting during the late 15th and the 16th century (Wulf (1991)). 17 the work culture of a society. Thompson (1967) documents how the work culture indeed changed over time. Based on case studies, he shows that after the implementation of the clock, a new perception of work discipline evolved slowly and gradually. Building on Thompson’s insights, Glennie and Thrift (1996, 2009) develop this perspective further and argued that along with the implementation and the further development of the clock, a new culture of work coordination, regularity, and repetition evolved. In a different strand of research McClelland (1961) finds empirical evidence for a developing ”achieving society” during the early modern times. H¨arter (2007), using a large sample of German cities and territories, finds an increase in regulations on social orders related to time such as curfew, closing time of taverns, etc., during the early modern times. Further supporting evidence for such a change can be found in the cultural movements of the 16th and 17th centuries (Macey (1979) and Wendorff (1980)). Protestantism of the 16th century identified time as a scarce product that had to be used wisely to achieve moral values and goals during the individuals’ worldly life (Engamarre (2009) and Boerner et al. (2019)). Seventeenth century scientists and philosophers, such as Robert Boyle and Thomas Hobbes, used the clock as a metaphor for the functioning of the world and to explain how institutions, such as the state, should work. Finally, this broad penetration can also be reflected in wealthy peoples acquisition of home clocks and watches during the 16th and 17th centuries, which was triggered by the early implementation of public clocks during the 14th and 15th centuries (Cipolla (1967)). Finally, the invention of the mechanical clock marked the beginning of a new phase of technological innovations where fine mechanics, automation and applied mathematics were combined. Craftsmen were becoming Vitruvian artisans by expanding their knowledge to other fields, early university graduates with craftsmen talents applied their knowledge to creating new machines, and learningby-doing engineers invented new mechanical tools. This movement has been recently identified as a prephase of the ”Scientific Revolution”, which took off only during the 17th century (Zilsel (2011), Long (2011), and Zanetti (2017)). It also has been seen as part of a change from a qualitative to a quantitative perception of the world. (Crosby (1997)) Thus this process of technological change created new scale and scope for the further development of science and technology, and as with all the other qualitative findings, it must be analyzed whether it translated into quantitatively measurable numbers in economic growth and development. 18 2.4 Solar eclipses as instruments Our econometric analysis can be potentially biased for three reasons, which are a common problem in econometric estimations. First, reverse causality can be a central issue since wealthier cities might have been more likely to adopt a public mechanical clock. Second, although we used several regressors that have been used in studies that analyzed the long-run growth of cities, our estimated equation can suffer misspecification. Finally, our historical data might be affected by potential measurement errors. In our empirical analysis, we will use an instrument based on solar eclipses.17 In this section, we aim to support this claim by historical narratives and stylized facts. Further quantitative evidence will be provided later. The use of solar eclipses as an instrument for the implementation of public mechanical clocks requires a more detailed two-step analysis. In the first step, we present the link between the appearance of solar eclipses, the curiosity of the western society to understand these phenomena, and the creation of astronomic instruments to measure and understand these events. In the second step, we describe the connection between astronomic instruments and the first public mechanical clocks. The observation and documentation of the course of the celestial bodies and specific astronomic events date back to ancient times (Lindsay (1858) and Steele (2000)). Solar eclipses have elicited a special fascination. They could be observed by everyone, and due to their rare appearance, they were perceived as sudden, irregular, and often supernatural events. This was, in particular, perceived this way by the European medieval society where in contrast with the Arab and Chinese societies, hardly any recently compiled astronomic knowledge existed (the ancient Greek knowledge was almost forgotten).18 These celestial movements created much curiosity and speculation, and left room for interpretation in European societies. Whereas there was an interest in understanding, learning, and catching up on astronomic knowledge from the Arabic world (Chabas and Goldstein (2012)), the movement and constellation of the celestial bodies was very much understood as God’s work. Therefore to understand and predict the future constellation of the stars meant to be closer to God’s plan and revelation (Borst (1989)). Coincidental political and religious events during solar eclipses supported such causal reasoning further. For instance, in one of the Gospels, the evangelist Luke 17In Section 6, we consider distance from the first adopters as an alternative instrument. 18Stephenson (1997) reports a large set of sources that document the different perception of solar eclipses by varies societies. 19 reports a total solar eclipse during the crucifixion of Jesus. In other examples the death of the son of Mohamed in 632 and the death of the emperor Louis and the Treaty of Verdun in 840 coincided with solar eclipses. In addition, beyond this connection between astronomy and religiosity, the field of astrology also developed (which was not clearly separated from astronomy), in which the prediction and understanding of the constellation of the stars was related to personal future, destiny and the fortune of whole societies (Blake (2014)).19 For instance the appearance of a solar eclipse exactly 1000 years after the crucifixion of Jesus created speculations about the end of the world (Borst (1989)). Furthermore, there is also ample evidence of the existence of astronomical and astrological advisors at the European royal courts and aristocratic houses starting during the 12th and 13th century (Mentgen (2005) and Deimann and Juste (2015)). Again, European rulers were particularly receptive to the advice. For example, Guido Bonatti who was an Italian mathematician, astronomer and astrologer served Frederik II during the second half of the 13th century. Bonatti advised Frederik II on many political decisions (for instance, the optimal constellation for going to war) and became one of the most important astrologers and public figures of the time. The study of these heavenly bodies and astronomic events required the development and the application of various instruments. In particular, so-called astrolabes were developed, which date back to ancient times and were transmitted from the Arabs to medieval Europe (Turner (1911)). An astrolabe was able to measure and simulate astronomic constellations and to measure time in equinoctial hours.20 Astrolabes became essential instruments for astronomers and astrologers to measure constellations of heavenly bodies. The use of the time function was in particularly important in European astrolabes (McCluskey (2000)). Furthermore, also sun and water clocks were also used to measure astronomic activities; however, they were not as versatile and easy to handle as astrolabes. In Europe, water clocks can be observed in particular in medieval monasteries, where they were also used to study astronomy. The construction of several very advanced astronomic water clocks to study astronomy can be documented in the medieval Islamic world and in medieval China (Cipolla (1967) and al Jazari (1973)) The connection between astronomic events and the development of machines measuring astronomic movement is not an exclusively medieval phenomena. Looking at some arche- 19The European astrological tradition was very much based on Islamic astrology which had its peak much earlier on. However the main books had only been translated by the 12th century into Latin. 20The places where these astrolabes were found in Europe (King (2011)) seem to overlap with areas where solar eclipses frequently appeared. However, due to the fragmented nature of the source material, further quantification is not possible 20 ological discoveries, we can find some suggestive links between astronomic events and prototypical machines for measuring time.21 Therefore, we can establish a link between the observation of astronomic events and the creation of instruments and basic machines to measure these events. The use of solar eclipses not only appears to be a strong motivation for the development of intellectual curiosity and astronomic instruments but also enables us to separate Europe into both areas with and without eclipses and consequently areas with stronger or weaker motivation to study and understand astronomy. The second link is between astronomic instruments and the development of public mechanical clocks. Price (1955) and White (1978) stress that mechanical clocks are not originated by previous forms of clocks but by planetaria and geared astrolabes. Dohrn-van Rossum (1996) states that medieval scholars were interested only in the development of machines that were related to astronomy. Cipolla takes the clock as a prime example of such a machine. Whereas the precise sequence and evolution from earlier clocks and astronomic instruments to the creation of public mechanical clock have been widely debated, there are no doubts that a clear correlation can be established, which was outlined in Section 2.1. Consequently, we can use the appearance of solar eclipses through the curiosity, invention and application of related astronomic machines as an instrument for the implementation of public mechanical clocks. More precisely, we consider regions and cities where solar eclipses appeared as places with a higher likelihood of building clocks.22 21As matter of example, juxtaposing the number of total solar eclipses during the ancient times with the location of the astrological heritage provided by the United Nations Educational Scientific and Cultural Organization (UNESCO) and available at https://www3.astronomicalheritage.net/, we can observe several cases where ancient and more recent astronomical sites are located next to intense solar eclipse activities. Keeping in mind that this sample is not representative, we would like to refer to the examples of Stonehenge (Hawkings (1988)), the Stone Circle of Odry in Poland (Sadowski et al. (1993)), and the Navajo star ceiling in the US (Williamson (1984)). Two case studies are particularly striking. The first one is related to the archeological site located at Deir el-Bahri in Egypt. According to the archeological interpretation of the discoveries of the Tomb of Senenmut at Western Thebes (Neugebauer and Parker (1988)), and dated 1470 B.C., the ceiling of the heritage depicts one of the oldest representation of the celestial firmament, which was inspired by the direct vision of the celestial movements and very likely influenced by two total solar eclipses during the years 1522 B.C. and 1477 B.C., respectively. In addition, the problem of measuring the astronomical time justifies the presence in the tomb of a monumental guideline for the construction of a water clock (clepsydra), which would be introduced later by the Egyptians in the Roman Empire (Sloley (1931)). The second example is related instead to the so-called Antikythera mechanism, dated to 205 BC and discovered from a shipwreck on the omonimous Greek island of Antikythera. Defined by Price (1959) as an ”ancient Greek computer”, it is a primordial clocklike mechanism for calculating the motion of stars and planets and for predicting eclipses during the eclipse cycle of 223 lunar months. 22 The reader might wonder why we use cities and regions rather than the location of monasteries as the crucial geographical points and connections. First, we are interested in the implementation of public mechanical clocks in cities and their related growth path. Second, most medieval cities that we study had at least one monastery inside their town walls and all of them had one in their immediate neighborhood. Therefore, the use of cities as geographical loci of potential human capital accumulation and adopters of the clock can be justified. 21 To conclude this descriptive two-stage analysis we give some narrative historical case studies of various towns that were covered before the introduction of clocks by several solar eclipses (as defined later on), had links to astronomy and were among the first adopters of the public mechanical clock: In Mechelen, a Flemish city, the astronomer and philosopher Henry Bate of Mechelen both elaborated tables for predicting eclipses (the so-called Tabulae Mechlinenses) and claimed to have built an astrolabe containing a time component at the end of the 13th century (White (1978) and Zanetti (2017)), and one of the first public mechanical clocks can be found, which also had an astronomical component. Another example is the city of Erfurt, where historical sources document the existence of astronomical and astrological experts (which must have been equipped with astrolabes) during the 13th century (Mentgen (2005)). Indeed Erfurt was also one of the first towns in Germany with a public mechanical clock. Another interesting case is Perpignan (the seat of the King of Aragon) where a strong astronomic interest and a collection of astrolabes by King Pedro IV preceded the building of the first public clock on the Iberian Peninsula. It is also documented that the King had several astronomers and specialized metal workers at the court who developed and maintained his scientific instruments (Beeson (1982) and Perez-Alvarez (2013)). Finally, the city of Padua has one of the earliest and still existing astronomical clocks. Padua became one of the most important centers for astrology in Europe starting during the late 13th and early 14th century, for instance, with Pietro D’Abano, who was a professor at Padua University and an expert in astrology and astronomy with in-depth knowledge of the astrolabe (e.g, Canova (2011)). 3 Data This section contains an overview of the city- and country-level variables considered in the empirical analysis.23 We determine the presence of and the year of adoption of public mechanical clocks during the period of our analysis mainly from four different sources: Bilfinger (1892), who analyzes the introduction of city clocks in France, Germany, England and Italy; Ungerer (1931), who provides a list and description of mechanical clocks in Europe; Dohrn-van Rossum (1996), who historically describes the adoption process, and Glennie and Thrift (2009), who concentrate their attention on the use of time in England. This initial dataset is integrated with an additional and nonpublished 23Table A1 in Appendix A contains the descriptive statistics of these variables. 22 list provided to us by Dohrn-van Rossum with other information from other sources (Cipolla (1967) and Landes (1983)). In addition, when possible, we confirm the date of adoption with the original historical sources. Our final list contains 182 clocks adopted between 1283, the date of adoption of the first mechanical clock in Dunstable, England, and 1450, when it is assumed that the period of yearly adoption concluded.24 Table 1 displays the aggregate number of clocks adopted, the number of cities with more than 5,000 inhabitants in 1400, and the percentage of adoption, which was computed as the ratio of the first two columns, at country level.25 We can observe that the adoption rate has an average of 20%, which covers both areas with low diffusion (e.g., Spain with 3%) and areas with more intense adoption (Switzerland with 90%). Table 1: The Diffusion of the Mechanical Clock in Europe before 1450. Country Cities adopting Cities available in Percentage of the clock Bairoch in 1400 adoption Austria 1 8 13 Belgium 14 33 42 Czechia 1 5 20 France 27 74 36 Germany 45 301 15 Italy 39 101 39 Malta 1 1 100 Netherlands 13 35 37 Poland 5 19 26 Spain 8 262 3 Sweden 8 18 44 Switzerland 10 11 90 Ukraine 1 2 50 United Kingdom 9 60 20 Total (all sample) 182 931 20 Source: Authors’ calculation based on the clock’s dataset. Population data drawn from Bairoch et al. (1988). In addition, we collect population data from Bairoch et al. (1988) which allows us to consider the population in all cities with more than 5,000 inhabitants for ten periods (i.e., 1000, 1200, 1300, 1400, 1500, 1600, 1700, 1750, 1800, and 1850). Following DeLong and Shleifer (1993) and Acemoglu et al. (2005a), we assume that population is a good proxy for urban income because data on urban GDP are not available before 1500.26 24In addition, we build a more extended list to 1600 to compute the penetration rate at the country level. 25In Section 7, we compute the penetration rate weighted by the population. 26Unfortunately, we don’t have population data more frequent than 100 years which would allow us to perform some 23 as whether cities are Atlantic ports for controlling for potential effects of the Columbian Exchange (column (2)), calories (3), and other geographical variables36 described in Section 3 (column (4)). In addition, in column (5) we introduce the level of the population during 1200 which allows to control for potential barrier costs in building the mechanical clocks.37 The results reported in the first part of Table 3 are robust over the different specifications: given the coefficients of the different sets of regressions and computing the marginal effects, we can quantify a positive and significant effect of the introduction of the mechanical clock on a time span of more than two centuries. In this set of regressions, the standard errors that clustered on time-invariant countries may be biased due to potential temporal autocorrelations. We also attempt to adjust the standard errors following the methodology introduced by Conley (1999) and obtain very similar values.38 Furthermore, consistently with the results represented in Figure 3, the estimation results of equation (2) display no effects in the 13th century, while there is a positive impact between 1300 and 1700. 36More precisely, potato crop suitability, amount of land suitable in the Old World, the amount of ruggedness, and elevation. 37In addition, this controls for the dynamics related to the mortality shock of the Black Death, which might lead to the increase to higher income per capita and thus a faster transition to the steady-state. 38More precisely, an estimate for cross sectional OLS corrected for spatial dependence of the specification of column (4) of Table 3 (first part) will give an estimated coefficient for CLOCK ·P ost equal to 0.22 and a corrected standard error of 0.09. 30 Table 3: OLS Regressions Dependent variable: ln P OP (1) (2) (3) (4) (5) 1. Difference-in-difference CLOCK 1.72*** 1.71*** 1.46*** 1.58*** 1.16*** (0.55) (0.55) (0.49) (0.41) (0.41) CLOCK ·P ost ·T rend 0.11*** 0.11*** 0.09*** 0.10*** 0.07** (0.04) (0.04) (0.03) (0.03) (0.03) p-value for joint significance CLOCK 0.01 0.01 0.02 0.00 0.02 Adjusted R20.66 0.66 0.66 0.69 0.70 2. Flexible Difference-in-difference CLOCK· Y EAR = 1200 0.29 0.29 0.25 0.13 (0.28) (0.28) (0.26) (0.27) Y EAR = 1300 0.83*** 0.83*** 0.75*** 0.66*** 0.45*** (0.22) (0.22) (0.21) (0.20) (0.12) Y EAR = 1400 0.80*** 0.80*** 0.69** 0.60** 0.44*** (0.28) (0.28) (0.28) (0.26) (0.14) Y EAR = 1500 0.61** 0.60** 0.56** 0.45** 0.32*** (0.23) (0.23) (0.21) (0.20) (0.11) Y EAR = 1600 0.60** 0.60** 0.56** 0.48** 0.37*** (0.23) (0.23) (0.22) (0.22) (0.13) Y EAR = 1700 0.72** 0.73** 0.74** 0.60** 0.46*** (0.29) (0.29) (0.27) (0.25) (0.14) Y EAR = 1750 0.49* 0.49* 0.48* 0.36 0.26** (0.26) (0.26) (0.24) (0.23) (0.11) Y EAR = 1800 0.36 0.37 0.36 0.24 0.15 (0.27) (0.27) (0.26) (0.25) (0.11) Y EAR = 1850 0.26 0.26 0.26 0.13 0.08 (0.29) (0.29) (0.28) (0.26) (0.12) p-value for joint significance CLOCK 0.00 0.00 0.00 0.00 0.00 Adjusted R20.66 0.66 0.67 0.68 0.69 Geographical coordinates Y Y Y Y Y Atlantic harbors N Y Y Y Y Calories N N Y Y Y Geographical variables N N N Y Y Population at 1200 N N N N Y Number of observations 9,319 9,208 Number of clusters 41 City and year fixed effects included in all regressions. Standard errors are robust and clustered by country. All control variables are interacted by year dummies. Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. 31 As described in the previous section, the OLS results displayed in Table 3 might be affected by several biases. First, reverse causality can be problematic. Endogeneity issues of the variable CLOCK on the dependent variable can arise because a city’s growth may drive the early adoption of the mechanical clock. In addition, equations (1) and (2) can be misspecified: institutional and other city characteristics (e.g., cities’ policies and institutional quality), which are not observable, might also play a role. Finally, the historical variables we use in the specification might be affected by measurement errors. For these reasons, we consider a two-stage-least-squares model based on two different equations, where the first stage considers the relationship between the adoption of the mechanical clock and the eclipses: CLOCKi·It=ρc+γt+X t∈T νtEclipsei·It+X t∈T ατXi·It+ζi(3) In this equation ρcand γtrepresents the city and time fixed effects, respectively, while the second stage is identical to equations (1) and (2) except for having the prediction of (3), i.e., \ CLOCKi·It, instead of the dummy generated by the interaction CLOCKi·It.Eclipse is the number of total solar eclipses that occur in the city in a time-span of a century. Looking at the first stage, we verify that the instrument is not weak considering different sets of logit estimates.39 Table 4 shows a significant and positive relationship between the adoption of the public mechanical clock and the number of eclipses independently on the specification we consider. Furthermore, Table A2 in the Appendix shows that the hypothesis of weakness is once again rejected after having analyzed the multiple partial F-test statistics for each endogenous regressors.40 Finally, we assess whether eclipses are tested to be exogenous using the same type of strategy and estimation suggested by Rubin (2014): as reported in Table A3 in the Appendix; after controlling for different sets of regressors, we find that eclipses did not explain institutional and geographical variable, with the only exception of that of being a Hanseatic center. Table 5 displays the second stage of the IV regression. The results are similar to those obtained by the OLS estimates: while in the first part of the table, the basic instrumented DiD shows a positive and significant effect of the public mechanical clock on the level of population with marginal 39The use of the logistic distribution is motivated by the best representation of the S-shape curve of adoption, as in Geroski (2000). In separate regressions we also consider the variable Eclipse as a dummy and we also estimate linear probability models. In both cases, we obtain very similar results for the relationship between the total solar eclipses and the adoption of the mechanical clock. 40To avoid estimation problems with overidentification, we consider for our set of time dummies interacted with clock we consider all the time dummies interacted for Eclipses except for the year 1200. 32 Table 4: Logit regressions Dependent variable: CLOCK (1) (2) (3) (4) (5) Eclipse 0.72*** 0.73*** 0.96*** 0.99*** 1.10*** (0.22) (0.22) (0.21) (0.23) (0.26) Longitude 0.01 0.01 -0.05*** -0.05*** -0.05** (0.02) (0.02) (0.02) (0.02) (0.02) Latitude 0.03 0.04 0.13*** 0.13*** 0.15*** (0.03) (0.03) (0.02) (0.04) (0.04) Atlantic -0.47 -0.31 -0.22 -0.34 (0.36) (0.30) (0.32) (0.41) Calories 0.37*** 0.32*** 0.30*** (0.00) (0.00) (0.00) Potato 0.06 0.06 (0.08) (0.09) Land suitability 0.22 0.19 (0.19) (0.20) Ruggedness 0.07 -0.16 (0.51) (0.61) Elevation 0.12 0.29 (0.39) (0.43) Population 1200 0.20*** (0.02) Constant -2.76** -2.78** -11.29*** -14.39*** -15.64*** (1.39) (1.40) (1.09) (3.81) (3.81) Pseudo R20.03 0.02 0.11 0.12 0.20 Number of observations 684 City and year fixed effects included in all regressions. Standard errors are robust and clustered by country. All control variables are interacted by year dummies. Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. effects comparable to the ones obtained by the OLS, in the second part we still observe increasing positive and significant effects of the mechanical clock on population dynamics between 1500 and 1700, which confirms the delay of the benefits of the technological adoption. These dynamics can be better visualized in Figure 6, which graphically compares the results of column (4) obtained in Tables 3 and 5, respectively. 33 Table 5: IV Regressions Dependent variable: ln P OP (1) (2) (3) (4) (5) 1. Difference-in-difference CLOCK 6.99** 7.03** 6.88*** 5.70** 5.16* (2.90) (2.96) (2.20) (2.68) (2.70) CLOCK ·P ost ·T rend 0.35** 0.36** 0.36*** 0.30** 0.28* (0.16) (0.17) (0.12) (0.15) (0.15) p-value for joint significance CLOCK 0.03 0.04 0.01 0.07 0.15 2. Flexible Difference-in-difference CLOCK· Y EAR = 1200 1.92 1.92 1.95 0.97 (1.66) (1.66) (1.69) (1.17) Y EAR = 1300 1.28 1.29 1.55 1.46* 0.46 (2.00) (2.00) (1.80) (0.86) (0.49) Y EAR = 1400 0.90 0.91 1.42 1.31 0.43 (2.16) (2.16) (1.76) (0.87) (0.58) Y EAR = 1500 2.47 2.49 2.41 2.15*** 1.09** (1.56) (1.56) (1.49) (0.72) (0.56) Y EAR = 1600 2.56* 2.56* 2.33 2.31*** 1.15* (1.54) (1.54) (1.50) (0.83) (0.67) Y EAR = 1700 2.92** 2.90** 2.45* 2.22*** 0.72 (1.27) (1.27) (1.45) (0.73) (0.60) Y EAR = 1750 1.50 1.51 1.50 1.32* 0.36 (1.65) (1.65) (1.57) (0.80) (0.59) Y EAR = 1800 1.30 1.31 1.28 1.13 0.26 (1.59) (1.59) (1.56) (0.77) (0.56) Y EAR = 1850 0.86 0.86 0.90 0.87 -0.05 (1.70) (1.70) (1.67) (0.95) (0.55) p-value for joint significance CLOCK 0.01 0.01 0.00 0.00 0.04 Geographical coordinates Y Y Y Y Y Atlantic harbors N Y Y Y Y Calories N N Y Y Y Geographical variables N N N Y Y Population at 1200 N N N N Y Number of observations 9,319 8,572 Number of clusters 41 Instruments considered: number of total solar eclipses in a time-span of a century interacted by year. City and year fixed effects included in all regressions. Standard errors are robust and clustered by country. All control variables are interacted by year dummies. Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. Exogenous regressors are partialled out. 34 Figure 6: The effects of CLOCK on population. OLS and IV coefficients of the flexible DiD. -.5 0 .5 1 Coefficients 1200 1400 1600 1800 year -1 0 1 2 3 4 Coefficients 1200 1400 1600 1800 year Figure 6 displays the coefficients of specification (4) of the flexible DiD reported in column (4) of Tables 3 (on the left) and 5 (on the right). Confidence intervals at 95% reported. Our instrument is based on the assumption that the eclipses had a direct effect on the adoption of the mechanical clock. We test this hypothesis by considering a set of placebo tests that exploit both the past and the future trajectories of total and annular eclipses. The data provided by Espenak (2015) allow us to track these astronomical events for a period before the contribution of our instrument, i.e., from 2000 B.C. to 800 A.D. In addition, we collect information over the period 1450 to 3000 A.D. We construct the instruments while summing up the number overlapping eclipses for cities in intervals of 100 years.41 We move the sum each year, which allows us to construct 168 different placebos. We rerun the regressions by substituting one of the placebos each time. Similar to the graphical representation of placebos in Madestam et al. (2013), the left hand side of Figure 8 shows the empirical cumulative distribution function of the different coefficients of Eclipse from the 168 different placebos of the logit regression that we use as a first stage.42 On the right-hand side we report the comparison of the respective z−statistics. The figure suggests that not only are the actual estimates (represented by the red line) is an outlier with respect to the entire distribution (only 3.57% of the coefficients are higher) but also that only 5.95% of the coefficients are significantly different (99%) from zero, which provides additional support for the positive causal relationship between the frequencies of the actual eclipses and the adoption of the public mechanical clock. 41To be restrictive in our exercise, we consider all the cities available in the dataset, independently on the appearance in particular years in the Bairoch et al. (1988) dataset. 42More precisely, we consider the specification reported in column (5) of Table 4 compared to the actual estimates. 35 Figure 7: Placebo test based on 168 historical and future total solar eclipses On the left we compare the effective estimates of the logit regression reported in Table 4 (red line) with the empirical cumulative distribution function (ECDF) obtained by 168 placebos. On the right the figure displays the effective z−statistics with the ECDF of the placebos. 5 Additional robustness checks In this section we are going to test the robustness of the results reported in our OLS and IV estimations in several ways. First, we look at the potential effect of another important technology introduced at the end of the Middle Ages, i.e., the printing press. Second, we propose a different instrument. Third we test the relationship between the adoption of the public mechanical clock and the population exploiting a propensity score matching technique. Fourth, we test whether our results are affected by particular subsets of the dataset considered. Finally, we exploit the time of adoption as additional information for our estimation. Considering the first issue, we study the potential relationship between the mechanical clock and a more recent innovation, i.e., the printing press, since recent contributions have underlined the relevance of this technology on both economic growth (Dittmar (2011)) and the adoption of Protestantism (Rubin (2014)).43 To have results comparable to the work of Dittmar (2011), we consider the following econometric specification, which is inspired by the classical equation formula that was derived by 43Boerner et al. (2019) study the effect of both mechanical clocks and the printing press on the choice of religious beliefs and, in particular, Calvinism. 36 Mankiw et al. (1992), where economic growth is explained by a set of variables that are related to the input of production and human capital and the initial economic condition of the period studied (Barro (1991)): ∆ ln POPitx =β0+β1CLOCKi+β2Pressi+β3CLOCKi·Pressi+β4Xit +β5ln P OPi0+vit (4) where for each city iand time t= 1400,1500,1600,1700,1800 and x=t−100, t −200, t −300, t − 400, ∆POPit−x= ln P OPit P OPix , ln Popi0is the initial level of population, and Xis a set of control variables that were described in the previous section. In addition, we include the two dummies of the technological innovations (CLOCK and Press) and their interaction. The OLS estimates of equation (4) are reported in Table 6. To study the long-run effects, we consider six different intervals of time: 1200-1300, 1300-1400, 1400-1500, 1500-1600, 1500-1700, and 1500-1800. While these estimates display very similar results to the work of Dittmar (2011), we again find a positive effect of the mechanical clock on population growth during the period 1500-1700. A further empirical discussion of the relationship between both technologies can be found in Boerner et al. (2019): Table 6: OLS Regressions Dependent variable: ∆ ln P OP (1) (2) (3) (4) (5) Periods: 1300 −1400 1400 −1500 1500 −1600 1500 −1700 1500 −1800 CLOCK 0.10 0.03 0.02 0.17** 0.08 (0.08) (0.10) (0.05) (0.06) (0.08) Press 0.19** 0.22** 0.27** (0.07) (0.08) (0.11) CLOCK ·P ress 0.02 -0.01 0.08 (0.06) (0.10) (0.11) Adjusted R20.02 0.17 0.16 0.22 0.31 N. of observations 277 228 489 511 609 Controls considered in column (4) of Table 3 included in all regressions. Standard errors are robust and clustered by country. Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. As an additional robustness check, we also consider as an alternative instrument the distance from the very first adopters.44 As previously outlined, the adoption of mechanical clocks occurred in a few 44The distance from a knowledge source has been already examined in relation to education and labor economics 37 cities in a first wave in different regions in Europe. Then, diffusion in these and neighboring regions can be observed. The diffusion pattern can be explained by the fact that only a few experts had clockmaking skills. These clock makers traveled from city to city to sell their expertise by building mechanical clocks, and their expertise was shared slowly (Cipolla (1967), Landes (1983) and Dohrnvan Rossum (1996)). Thus, the likelihood of the implementation of a clock in a town depended on the distance from one of the first adopters. Thus, we can follow an established research methodology that has been used in related empirical historical studies (for instance, see Becker and Woessmann (2009) and Dittmar (2011)). However, in our case distance might be endogenous to the growth rate of a city before the implementation of the clock. The results in Table A4 in the appendix consider the distance from the very first innovators. As we illustrated in Figure 2, the cities adopting the mechanical clocks between the years 1283 and 1350, and shows a similar pattern as found in the Section 3. Alternatively, the increasing effect of the period between 1500 and 1700 is also confirmed by a propensity score matching technique (Rosenbaum and Rubin (1983)). More specifically, we match treatment and controls based on a logit specification reported in column (5) of Table 4. Figure 8 reports the yearly average difference of the treatment effects on the treated with respect to the control and the relative standard errors. An additional and legitimate concern can be that the the fact that by chance total and solar eclipses might have overlapped with very active economic area (e.g., the Southern part of England, the Netherlands, and the Northern part of Italy) by chance, thus creating potential problems related to the geographical balance of the dataset. Even if regressions excluding parts of the dataset seem to exclude this,45 we address this potential bias by adopting the estimation technique of the Entropy balancing algorithm proposed by Hainmueller and Xu (2013) and recently used by Angelucci et al. (2017). In practice, the algorithm creates in an artificial way a geographical and random balance of the cities that are not similar to the ones in the treatment group. The results are reported in Table 7, which is divided into two parts: in the upper part, we observe the comparison between treated and control before and after the weighting, respectively, while the lower part reports the results of the weighted regression. Additionally, we find consistent results as in the previous regressions. (e.g., Card (1993)) by other studies on innovation of product and culture and economic growth (Becker and Woessmann (2009) and Dittmar (2011)). 45Table A4 in the Appendix shows the OLS estimates excluding Germany, Italy, the Netherlands, Switzerland, and the Atlantic ports, respectively. 38 Figure 8: Difference of the ATT -.5 0 .5 1 1.5 ATT: difference 1200 1400 1600 1800 year Difference of the ATT derived from propensity score with 95% confidence interval. Source: Authors’ calculations based on the authors’ dataset of clocks. Finally, we exploit the fact that we have precise information on the data of the construction of the mechanical clock. This would allow us to check whether the results displayed before may be affected by two potential issues: first, the econometric models previously estimated do not consider the different dynamics related to the time of adoption of the public mechanical clock, which spans over a period of three different centuries; second, particularly in the OLS regressions, a potential reverse causality between the technology and the level of population. For these reasons, we can consider an event-study regression analysis in the spirit of Autor (2003),46 which is an extended version of equation (1) with the exception of using a set of dummy variables that take the value of one once the the lags of one or two centuries are considered (i.e., Adoptiont−200 and Adoptiont−100, respectively), the century of adoption (Adoptiont), and the four centuries after the innovations (i.e., Adoptiont+100, Adoptiont+200,Adoptiont+300, and Adoptiont+400, respectively). We consider cities with at least three consecutive observations. Table A6 in the Appendix shows the results of the event-study considering the different moment of the adoption of the technology as different experiments over time, while Figure 9 shows the estimated impact of the public mechanical clock over the time relative to the 46See Angrist and Pischke (2008) for an introduction and a comparison with respect the DiD methodology. 39 ible) difference-in-difference, event-study, propensity-score analysis, and entropy-tests, and last but not least, we introduced solar eclipses as a new instrument for clocks to deal with misspecification, measurement errors, and endogeneity. Finally, to extend and further explore the impact of mechanical clocks on contemporary behavior we concluded the paper with an analysis of intertemporal long-run correlation between the early implementation of clocks and contemporary cultural values and routines. These results indicate a positive relationship between an early implementation of the new technology and a tighter organization and coordination of societies, and stronger affinity to time. However these results create first of all an outlook for further investigation to understand the long-run evolution of daily routines related to fundamental technological changes in more detail. 46 References Acemoglu, D., D. Autor, D. Dorn, G. H. Hanson, and B. Price (2014): “Return of the Solow Paradox? IT, Productivity, and Employment in US Manufacturing,” American Economic Review, 104, 394–99. Acemoglu, D., S. Johnson, and J. A. Robinson (2005a): “The Colonial Origins of Comparative Development: An Empirical Investigation,” American Economic Review, 91, 1369–1401. ——— (2005b): “The Rise of Europe: Atlantic Trade, Institutional Change, and Economic Growth,” American Economic Review, 95, 546–579. al Jazari (1973): The Book of Knowledge of Ingenious Mechanical Devices: Kit´ab f´ı ma’rifat al-hiyal al-handasiyya, translated by Donald R. Hill, Berlin: Springer Science+Business Media. Algan, Y. and P. Cahuc (2010): “Inherited Trust and Growth,” American Economic Review, 100, 2060–2092. Andersen, T. B., J. Bentzen, C.-J. Dalgaard, and P. Selaya (2012): “Lightning, IT Diffusion, and Economic Growth Across U.S. States,” The Review of Economics and Statistics, 94, 903–924. Angelucci, C., S. Meraglia, and N. Voigtlaender (2017): “The Medieval Roots of Inclusive Institutions: From the Norman Conquest of England to the Great Reform Act,” NBER Working Papers 23606, National Bureau of Economic Research, Inc. Angrist, J. D. and J.-S. Pischke (2008): Mostly Harmless Econometrics, Princeton, N.J.: Princeton University Press. Atkeson, A. and P. J. Kehoe (2007): “Modeling the Transition to a New Economy: Lessons from Two Technological Revolutions,” American Economic Review, 97, 64–88. Autor, D. H. (2003): “Outsourcing at Will: The Contribution of Unjust Dismissal Doctrine to the Growth of Employment Outsourcing,” Journal of Labor Economics, 21, 1–42. Bairoch, P., J. Batou, and P. Chevre (1988): The Population of European Cities from 800 to 1850, Geneva: Librairie Droz. 47 Barro, R. J. (1991): “Economic Growth in a Cross Section of Countries,” The Quarterly Journal of Economics, 106, 407–43. Becker, S. O. and L. Woessmann (2009): “Was Weber Wrong? A Human Capital Theory of Protestant Economic History,” The Quarterly Journal of Economics, 124, 531–596. Beeson, C. F. C. (1982): The Making of a Tower Clock and Bell for the King’s Castle., London: Antiquarian Horological Society. Bettencourt, L. M. A., J. Lobo, D. Helbing, C. K¨ uhnert, and G. B. West (2007): “Growth, innovation, scaling, and the pace of life in cities,” Proceedings of the National Academy of Sciences, 104, 7301–7306. Bilfinger, G. (1892): Die Mittelalterlichen. Horen und die Modernen Stunden Ein Beitrag zur Kulturgeschichte, Stuttgart: Kohlhammer. Blake, S. P. (2014): Astronomy and Astrology in the Islamic World, Edinburgh: Edinburgh University Press. Boerner, L., J. Rubin, and B. Severgnini (2019): “A Time to Print, a Time to Reform,” Tech. rep. Borst, A. (1989): Astrolab und Klosterreform an der Jahrtausendwende, Heidelberg: Universit¨at Verlag. Bresnahan, T. F., E. Brynjolfsson, and L. M. Hitt (2002): “Information Technology, Workplace Organization, And The Demand For Skilled Labor: Firm-Level Evidence,” The Quarterly Journal of Economics, 117, 339–376. Bresnahan, T. F. and M. Trajtenberg (1995): “General purpose technologies ’Engines of growth’?” Journal of Econometrics, 65, 83–108. Canova, G. M. (2011): “Padua and the Stars: Medieval Painting and Illuminated Manuscripts.” in VI. Proceedings of a conference held October 18-23, 2009 in Venezia, Italy., ed. by E. M. Corsini, ASP Conference Series, San Francisco, vol. 441. 48 Cantoni, D. (2015): “The Economic Effects Of The Protestant Reformation: Testing The Weber Hypothesis In The German Lands,” Journal of the European Economic Association, 13, 561–598. Card, D. (1993): “Using Geographic Variation in College Proximity to Estimate the Returns to Schooling,” in Aspects of Labour Market Behaviour: Essays in Honor of John Vanderkamp, University of Toronto Press. Caselli, F. and W. J. Coleman (2001): “Cross-Country Technology Diffusion: The Case of Computers,” American Economic Review, 91, 328–335. Chabas, J. B. and B. R. Goldstein (2012): A survey of European astronomical tables in the late middle ages, Leiden: Brill. Cipolla, C. M. (1967): Clocks and Culture, 1300-1700, New York: Norton. Comin, D., W. Easterly, and E. Gong (2010): “Was the Wealth of Nations Determined in 1000 BC?” American Economic Journal: Macroeconomics, 2, 65–97. Conley, T. G. (1999): “GMM estimation with cross sectional dependence,” Journal of Econometrics, 92, 1–45. Crafts, N. (2002): “The Solow Productivity Paradox in Historical Perspective,” CEPR Discussion Papers 3142, C.E.P.R. Discussion Papers. Crombie, A. (1961): “Quantification in Medieval Physics,” Isis: A Journal of the History of Science Society, 61, 143–160. Crosby, A. W. (1997): The Measure of Reality. Quantification and Western Society, 1250-1600., Cambridge: Cambridge University Press. Czernich, N., O. Falck, T. Kretschmer, and L. Woessmann (2011): “Broadband Infrastructure and Economic Growth,” Economic Journal, 121, 505–532. Dart, J. (1726): The History and Antiquities of the Cathedral Church of Canterbury, London: J. Cole. David, P. A. (1990): “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox,” American Economic Review, 80, 355–61. 49 ——— (1991): “Computer and Dynamo: The Modern Productivity Paradox in a Not-Too-Distant Mirror,” in Technology and Productivity: the Challenge for Economic Policy, OECD Publishing, Paris. Deimann, W. and D. Juste (2015): Astrologers and Their Clients in Medieval and Early Modern Europe, Vienna: B¨ohlau Verlag. DeLong, J. B. and A. Shleifer (1993): “Princes and Merchants: European City Growth before the Industrial Revolution,” Journal of Law and Economics, 36, 671–702. Devine, W. D. (1983): “From Shafts to Wires: Historical Perspective on Electrification,” Journal of Economic History, 43, 347–372. Dittmar, J. E. (2011): “Information Technology and Economic Change: The Impact of The Printing Press,” The Quarterly Journal of Economics, 126, 1133–1172. Dohrn-van Rossum, G. (1996): History of the Hour: Clocks and Modern Temporal Orders, Oxford: The University of Chicago Press. ——— (2005): “Die ”Erfindung der Erfinder” im Sp¨atmittelalter,” in Tradition, Innovation, Invention: Fortschrittsverweigerung und Fortschrittsbewusstsein im Mittelalter, de Gruyter, 27–49. Einstein, A. and L. Infeld (1938): The Evolution of Physics: The Growth of Ideas from Early Concepts to Relativity and Quanta, Cambridge: Cambridge University Press. Engamarre, M. (2009): On Time, Punctuality, and Discipline in Early Modern Calvinism, Cambridge: Cambridge University Press. Espenak, F. (2015): “Eclipse Predictions,” NASA/GSFC, http: // eclipse. gsfc. nasa. gov/ eclipse. html . Fagniez, G. C. (1877): Etudes sur l’industrie et la classe industrielle a Paris au XIIIe et au XIVe siecle, Paris: Vieweg. Fernandez, D., M. Lechner, and N. Rodriguez-Planas (2016): “Difference-in-difference estimation by FE and OLS when there is panel non-response,” Journal of Applied Statistics, 43, 2044–2052. 50 Fogli, A. and R. Fernandez (2009): “Culture: An Empirical Investigation of Beliefs, Work, and Fertility,” American Economic Journal: Macroeconomics, 1, 146–177. Galor, O. and O. ¨ Ozak (2016): “The Agricultural Origins of Time Preference,” American Economic Review, 106, 3064–3103. Geroski, P. A. (2000): “Models of technology diffusion,” Research Policy, 29, 603–625. Gibbon, J., C. Malapani, C. L. Dale, and C. Gallisteld (1997): “Toward a neurobiology of temporal cognition: advances and challenges,” Current Opinion in Neurobiology, 7, 170–184. Glennie, P. and N. Thrift (1996): “Reworking E.P. Thompson’s ’Time, Work-discipline and Industrial Capitalism’,” Time and Society, 5, 275–299. ——— (2009): Shaping the Day: A History of Timekeeping in England and Wales 1300-1800, Oxford: Oxford University Press. Gordon, R. (2016): The Rise and Fall of American Growth: The U.S. Standard of Living since the Civil War, Princeton: The Princeton Economic History of the Western World. Granger, C. W. J. (1969): “Investigating Causal Relations by Econometric Problems,” Econometrica, 37, 424–38. Griliches, Z. (1957): “Hybrid Corn: An Exploration of the Economics of Technological Change,” Econometrica, 25, 501–522. Hainmueller, J. and Y. Xu (2013): “ebalance: A Stata Package for Entropy Balancing,” Journal of Statistical Software, 54. Hamermesh, D. S., C. K. Myers, and M. L. Pocock (2008): “Cues for Timing and Coordination: Latitude, Letterman, and Longitude,” Journal of Labor Economics, 26, 223–246. H¨ arter, K. (2007): “Zeitordnungen und Zeitverbrechen: Reglementierung, Disziplinierung und Fragmentierung von Zeit in der fr¨uhneuzeitlichen Policeygesetzgebung,” in Die Autoritt der Zeit in der Frhen Neuzeit, ed. by A. B. et al., Lit Verlag, 187–232. Hawkings, G. S. (1988): Stonehenge decoded, New York: Hippocrene Books. 51 Hayek, F. A. (1945): “The Use of Knowledge in Society,” American Economic Review, 35, 519–530. Hofstede, G., G. J. Hofstede, and M. Minkov (2010): Cultures and organizations: software of the mind: intercultural cooperation and its importance for survival, New York: McGraw-Hill, 3 ed. Jorgenson, D. W. (2005): Productivity, Vol. 3 Information Technology and the American Growth Resurgence, Cambridge, MA: MIT press. Jorgenson, D. W. and K.-Y. Vu (2005): “ The Efficiency of Capital Allocation,” Scandinavian Journal of Economics, 107, 631–650. King, D. A. (2011): Astrolabes from Medieval Europe, Aldershot and Burlington: Ashgate Variorum. Landes, D. (1983): Revolution in Time, Cambridge: Harvard University Press. Lazear, E. P. (1996): “Performance Pay and Productivity,” NBER Working Papers 5672, National Bureau of Economic Research, Inc. Le Goff, J. (1971): “The Town as an Agent of Civilisation, 1200-1500,” in The Fontana Economic History of Europe: The Middle Ages, Collins, vol. 1. ——— (1982): Time, Work, and Culture in the Middle Ages, Oxford: The University of Chicago Press. Leontieff, W. (1983): “National Perspective: The Definition of Problems and Opportunities,” in The Long-Term Impact of Technology on Employment and Unemployment, National Academy of Engeneering. Levine, R. V. (1998): A Geography Of Time: The Temporal Misadventures of a Social Psychologist, New York: Basic Books, revised ed. Levine, R. V. and A. Norenzayan (1999): “The Pace of Life in 31 Countries,” Journal of Cross- Cultural Psychology, 30, 593–650. Lindsay, J. B. (1858): The Chrono-Astrolabe, London: Dundee. 52 Lippincott, K., U. Eco, and E. Gombrich (1999): The Story of Time, Greenwhich: National Maritime Museum. Long, P. O. (2011): Artisan/Practitioners and the Rise of the New Sciences, 1400-1600, Corvallis, OR: Oregon State University Press. Macey, S. L. (1979): Clocks and the Cosmos: Time in Western Life and Thought, Lancaster: Gazelle Book Services Ltd. Maddison, A. (2007): Contours of the World Economy, 1-2030 AD. Essays in Macro-Economic History, Oxford: Oxford University Press. Madestam, A., D. Shoag, S. Veuger, and D. Yanagizawa-Drott (2013): “Do Political Protests Matter? Evidence from the Tea Party Movement,” The Quarterly Journal of Economics, 128, 1633–1685. Mankiw, N. G., D. Romer, and D. N. Weil (1992): “A Contribution to the Empirics of Economic Growth,” The Quarterly Journal of Economics, 107, 407–37. Marx, K. (1863): “Letters to Engels,” in Collected Works. McClelland, D. (1961): The Achieving Society, Princeton: Van Nostrand Co. McCluskey, S. C. (2000): Astronomies and Cultures in Early Medieval Europe, Cambridge: Cambridge University Press, 2nd ed. McEvedy, C. and R. Jones (1978): Atlas of World Population History, New York: Facts on File. Mentgen, G. (2005): Astrologie und ¨ Offentlichkeit im Mittelalter, Stuttgart: A. Hiersemann. Mihm, M. and A. Mihm (2007): Mittelalterliche Stadtrechnungen im historischen Prozess, Die Aelteste Duisburger Ueberlieferung (1348-1449), Cologne: Boehlau Verlag. Mokyr, J. (1992): The Lever of Riches: Technological Creativity and Economic Progress, Oxford: Oxford University Press. ——— (2002): The Gifts of Athena: Historical Origins of the Knowledge Economy, Princeton, NJ: Princeton University Press. 53 ——— (2005): “Long-Term Economic Growth and the History of Technology,” in Handbook of Economic Growth, ed. by P. Aghion and S. N. Durlauf, Elsevier, vol. 1, chap. 17, 1113–1180. Mumford, L. (1934): Technics and Civilization, New York: Harcourt, Brace & World. Munro, J. H. A. (2005): “Before and After the Black Death: Money, Prices, and Wages in Fourteenth-Century England,” Working papers, University of Toronto, Department of Economics. Needham, J. (1986): Science and Civilization in China, Cambridge: Cambridge University Press, 6 ed. Neugebauer, O. and R. A. Parker (1988): Egyptian Astronomical Texts: volume 1, London: Brown University Press. Nuessli, C. (2011): “Geographical Maps, http://www.euratlas.com/about.html,” . Nunn, N. and D. Puga (2011): “Ruggedness: The Blessing of Bad Geography in Africa,” Review of Economics and Statistics. Nunn, N. and N. Qian (2011): “The Potato’s Contribution to Population and Urbanization: Evidence From A Historical Experiment,” The Quarterly Journal of Economics, 126, 593–650. Ogle, V. (2015): The Global Transformation of Time. 1870-1950, Cambridge: Harvard University Press. Perez-Alvarez, V. (2013): “Mechanical clocks in the medieval Castilian Royal Cour,” Antiquarian Horology, 34, 498–502. Price, D. J. d. S. (1955): “Clockwork Before the Clocks,” Horological Journal, 81–84. ——— (1956): “The Prehistory of Clocks,” Discovery, 153–157. ——— (1959): “An Ancient Greek Computer,” Scientific American, 200, 60–67. Ricardo, D. (1821): On the Principles of Political Economy and Taxation, London: John Murray, 3rd ed. Rogers, E. M. (2003): Diffusion of Innovations, New York: Free Press, 5th ed. 54 Rosenbaum, P. R. and D. B. Rubin (1983): “The central role of the propensity score in observational studies for causal effects,” Biometrika, 70, 41–55. Rosenberg, N. and L. J. Birdzell (2008): How the West Grew Rich: The Economic Transformation of the Industrial World, New York: Basic Books. Rubin, J. (2014): “Printing and Protestants: An Empirical Test of the Role of Printing in the Reformation,” The Review of Economics and Statistics, 96, 270–286. Sadowski, R. M., M. Ziolkowski, and K. Piasecki (1993): “Stone rings of northern Poland,” in Archaeoastronomy in the Old World, edited by Douglas C. Heggie, Cambridge University Press. Samuelson, P. A. (1988): “Mathematical Vindication of Ricardo on Machinery,” Journal of Political Economy, 96, 274–82. Sloley, R. (1931): “Primitive Methods of Measuring Time: With Special Reference to Egypt,” The Journal of Egyptian Archaeology, 17, 166–178. Sombart, W. (1921): Luxus und Kapitalismus, Munich: Duncker & Humblot. Squicciarini, M. P. and N. Voigtl¨ ander (2015): “Human Capital and Industrialization: Evidence from the Age of Enlightenment,” The Quarterly Journal of Economics, 130, 1825–1883. Steele, J. (2000): Observations and Predictions of Eclipse Times by Early Astronomers, Dordrecht, Boston and London: Kluwer Academic Publishers. Stephenson, F. R. (1997): Historical Eclipses and Earth’s Rotation., Cambridge: Cambridge University Press. Thompson, E. (1967): “Time, Work-Discipline, and Industrial Capitalism,” Past and Present, 38, 56–97. Thorndyke, L. (1941): “Invention of the Mechanical Clock about 1271 A.D.” Speculum, 16, 242–3. Truitt, E. R. (2015): Medieval Robots: Mechanism, Magic, Nature, and Art, Philadelphia, PA: University of Pennsylvania Press. 55 Table A3: Exogeneity test (1) (2) (3) (4) (5) (6) (7) (8) Dependent Press F ree imperial Bishop Water Hanseatic Log Distance Log Sixteenth−Log F ifteenth− variable: city to W ittenberg Century Growth Century Growth Eclipse 0.01 -0.01 -0.08 0.11 0.18*** -0.04 -0.08 0.03 (0.05) (0.04) (0.06) (0.08) (0.06) (0.08) (0.08) (0.11) Adjusted R20.35 0.70 0.37 0.11 0.44 0.84 0.02 -0.01 N. of observations 248 248 248 248 248 248 174 122 OLS regression based on the dataset and specification provided by Rubin (2014) and introducing the variable Eclipse. Significance at the 90%, 95%, and 99%, respectively. 62 Table A4: IV Regressions Dependent variable: ln P OP (1) (2) (3) (4) (5) CLOCK· Y EAR = 1200 0.17 0.17 0.01 -0.20 (0.69) (0.70) (0.74) (0.79) Y EAR = 1300 1.17** 1.18** 1.13** 1.01** 0.98* (0.48) (0.48) (0.45) (0.49) (0.51) Y EAR = 1400 0.82* 0.83* 0.63 0.41 0.42 (0.48) (0.49) (0.41) (0.47) (0.56) Y EAR = 1500 0.78* 0.78* 0.74* 0.58 0.88* (0.40) (0.41) (0.39) (0.46) (0.53) Y EAR = 1600 0.74** 0.74** 0.71** 0.66 1.31* (0.34) (0.34) (0.32) (0.45) (0.68) Y EAR = 1700 1.09** 1.08** 1.15** 1.11* 1.32** (0.52) (0.52) (0.51) (0.57) (0.62) Y EAR = 1750 0.79 0.79 0.86* 0.58 1.24* (0.51) (0.51) (0.51) (0.52) (0.69) Y EAR = 1800 0.79 0.79 0.88* 0.54 1.28* (0.52) (0.52) (0.52) (0.50) (0.66) Y EAR = 1850 0.65 0.66 0.76 0.24 0.80 (0.53) (0.54) (0.54) (0.49) (0.72) Geographical coordinates N N N N N Atlantic harbors N Y Y Y Y Calories N N Y Y Y Geographical variables N N N Y Y Population at 1200 N N N N Y p-value for joint significance CLOCK 0.00 0.00 0.00 0.00 0.01 Number of observations 9,319 8,572 Number of clusters 41 Instrument considered: distance from the first adopters interacted by year. City and year fixed effects included in all regressions. Longitude and latitude not included. Standard errors are robust and clustered by country. All control variables are interacted by year dummies. Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. Exogenous regressors are partialled out. 63 Table A5: OLS Regressions Dependent variable: ln P OP (1) (2) (3) (4) (5) Sample Germany the Italy Switzerland Atlantic without: Netherlands ports CLOCK∗ Y EAR = 1200 0.29 0.13 0.17 0.13 0.12 (0.22) (0.25) (0.30) (0.24) (0.25) Y EAR = 1300 0.81*** 0.68*** 0.75*** 0.69*** 0.65*** (0.18) (0.18) (0.23) (0.17) (0.18) Y EAR = 1400 0.78*** 0.58** 0.60* 0.58** 0.55** (0.17) (0.24) (0.30) (0.22) (0.23) Y EAR = 1500 0.56*** 0.43*** 0.45** 0.42*** 0.41** (0.14) (0.16) (0.20) (0.14) (0.16) Y EAR = 1600 0.59*** 0.46*** 0.50** 0.46*** 0.44** (0.16) (0.16) (0.20) (0.15) (0.16) Y EAR = 1700 0.76*** 0.56*** 0.64*** 0.58*** 0.55*** (0.19) (0.19) (0.22) (0.18) (0.19) Y EAR = 1750 0.50** 0.33* 0.42* 0.34* 0.33 (0.19) (0.19) (0.23) (0.18) (0.20) Y EAR = 1800 0.39* 0.22 0.28 0.23 0.20 (0.21) (0.22) (0.28) (0.22) (0.23) Y EAR = 1850 0.27 0.12 0.17 0.10 0.11 (0.24) (0.24) (0.30) (0.24) (0.26) p-value for joint significance CLOCK 0.00 0.00 0.00 0.00 Adjusted R20.68 0.68 0.66 0.68 0.68 Number of observations 8,100 9,012 7,533 9,211 9,156 Controls considered in column (4) of Table 3 included in all regressions. City and year fixed effects included in all regressions. Standard errors are robust and clustered by country. All control variables are interacted by year dummies. Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. 64 Table A6: Event Study Dependent variable: ln P OP (1) (2) (3) Adoptiont+400 0.16** 0.13* 0.15* (0.08) (0.08) (0.08) Adoptiont+300 0.26** 0.20** 0.24** (0.10) (0.10) (0.11) Adoptiont+200 0.25** 0.17* 0.21* (0.12) (0.10) (0.11) Adoptiont+100 0.29* 0.19* 0.22** (0.14) (0.10) (0.11) Adoptiont0.35** 0.26** 0.30** (0.15) (0.11) (0.13) Adoptiont−100 0.111 0.0350 0.0343 (0.19) (0.16) (0.15) Adoptiont−200 -0.48* -0.41 -0.54* (0.19) (0.19) (0.17) Adjusted R20.81 0.82 0.83 Interaction Y ear ∗Country YYY Interaction Y ear2∗Country NYY Other variables N N Y Number of observations 7,477 Other variables: controls considered in column (4) of Table 3 included in all regressions. Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. 65 B. Mechanical clocks and GDP per capita This section allows us to test whether the adoption of public mechanical clocks had a direct impact not only on the growth of towns but also on productivity in wider geographical areas. Such an aggregate study can be motivated by our consideration of GDP per capita as a measure of the aggregate performances, which allows us to make better comparisons with the related research on the recent economic impact of the information technology discussed in the introduction.50 To test this, we take inspiration from the empirical framework introduced by Czernich et al. (2011), who analyze the effect of broadband Internet access on economic growth in European countries. The estimation is based on two stages: in the first stage, we study and predict the rate of penetration of the mechanical clock at the country level; in the second stage, we estimate the factors that are important for economic growth and detect the importance of the new technology using the findings of the first stage. For our analysis, we consider a slightly modified version of the traditional growth regression model introduced by Mankiw et al. (1992), which is considered in the second stage, and study the different effects of physical factors and technology: ∆ ln yct = ln Ac+γ2∆ ln POPct +γ3ln yc0+ξct (6) where for country cat time t∆ ln yis the gross domestic product (GDP, henceforth) per capita growth rate, ∆ ln P OP is the population growth, ln yc0is the initial level of GDP per capita and ξis the error term. To study the effect of clocks, we decompose the country-level of technology Acinto a general technological effect, represented by the parameter γ0and the contribution of the clock Bct, for which the penetration rate serves as proxy: Act =exp(γ0+γ1Bct) (7) Substituting (7) into (6), we obtain the modified growth equation ∆ ln yct =γ0+γ1Bct +γ2∆ ln POPct +γ3ln P OPc0+ξct (8) 50Dale Jorgenson and associates studied in depth the relationship between IT and productivity: for example, using growth accounting techniques Jorgenson (2005) and Jorgenson and Vu (2005) analyze ITs contribution to total factor productivity for the American and the world economies, respectively. 66 Similar to the discussion in the previous section, here, an OLS estimate of (8) may be affected by two different endogeneity issues: first, a problem of reverse causality can arise because country economic growth can positively drive the adoption of the mechanical clock; second, there could be a problem of misspecification because particular government policies and different institutional quality, which is difficult to measure, could play a role in the adoption of the mechanical clock. For these reasons, we imitate the strategy that was introduced by Czernich et al. (2011) by considering an instrumental variable approach, which is also useful for studying the pattern of diffusion, and by considering an instrumental variable approach, which is also useful for studying the pattern of diffusion. Following the previous contributions to the adoption rate (Griliches (1957), Geroski (2000), and Comin et al. (2010)) that are also motivated by the S-shaped diffusion of technology, we model the impact of the clock Bfollowing a logistic distribution Bct =φc 1 + exp h−˜ β(t−τ)i+ect (9) where φcis the saturation level, i.e., the maximum amount of adoption, ˜ βis a parameter displaying the double amount of maximum growth rate, τprovides information on the inflection point of the curve and eis the error term. To provide a value for φ, we assume that the saturation can be positively related to the percentage of the population living in an area covered by the combinations of eclipses studied in the previous section, eclipse sharec φc=φ0+λ1eclipse sharec(10) The availability of yearly population data described in Section 3 allow us to study the diffusion using more than 7,000 observations. More precisely, we consider population data for eight countries (Austria, Belgium, Denmark, Finland, France, Germany, Italy, Sweden, Switzerland, and the United Kingdom) for the period of 1250-1750. Table B1 shows the results of the regression of the equation obtained by substituting (10) into (9), which provide several pieces of information. We can observe that the parameter of eclipses (λ1) enters in a positive and significant way, with a a penetration rate of approximately 20%. In addition, while the parameter τsuggests that the inflection rate is situated at the year 1430, a period which is similar to those analyzed in the empirical analysis in Section 2, ˜ β 67 indicates a maximum growth rate of approximately 1.5% In addition, Figure B.1 compares the actual rate and the the fitted adoption rate generated by the first stage estimates and suggests a prevalence of logistic distributions in the countries analyzed. Table B1: NLS estimation for the diffusion curve λ1φ1˜ β τ 0.21*** 0.00*** 0.02*** 1430.68*** (0.00) (0.00) (0.00) (5.30) R20.30 N. of observations 8,421 Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. Figure B.1.: Mechanical adoption rate: Actual (blue line) and predicted (red line) 0 .2 .4 .6 .8 1 1000 1200 1400 1600 1800 year Adoption rate Fitted Belgium 0 .2 .4 .6 .8 1 1000 1200 1400 1600 1800 year Adoption rate Fitted France 0 .2 .4 .6 .8 1 1000 1200 1400 1600 1800 year Adoption rate Fitted Germany 0 .2 .4 .6 .8 1 1000 1200 1400 1600 1800 year Adoption rate Fitted Italy 0 .2 .4 .6 .8 1 1000 1200 1400 1600 1800 year Adoption rate Fitted Switzerland 0 .2 .4 .6 .8 1 1000 1200 1400 1600 1800 year Adoption rate Fitted United Kingdom Table B1 contains the results of the second stage. The first four columns show the impact of the contemporaneous adoption of the mechanical clock ( b Bct) on economic growth, while columns (5)-(8) 68 consider a lag of a century ( b Bct−100). In all the cases, we consider both cluster and bootstrapped standard errors errors based on 50 replications. Our estimates are based on the GDP per capita each 100 years collected by Maddison (2007), and confirm the findings of the regressions based on city-level data, i.e., the penetration of the GPT has a positive and significant impact on GDP per capita growth. We find that an increase of 10 percentage points in the diffusion of mechanical clocks can raise the GDP per capita growth approximately 30 % in a century. 69 Table B1: The effects of mechanical clocks on GDP per capita: second stage. (1) (2) (3) (4) (5) (6) (7) (8) Bootstrapped Bootstrapped Bootstrapped Bootstrapped Std. Errors Std. Errors Std. Errors Std. Errors b Bct 0.32*** 0.32*** 0.32*** 0.32*** b Bct−100 0.33*** 0.33*** 0.33*** 0.33*** (0.78) (1.13) (0.81) (1.69) (0.84) (1.23) (0.86) (1.82) ln y00.00 0.00 0.00 0.00 ln y00.00 0.00 0.00 0.00 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) ∆ ln POP 0.32 0.32 0.32 0.32 ∆ ln P OP 0.32 0.32 0.32 0.32 (0.24) (0.24) (0.25) (0.25) (0.24) (0.24) (0.25) (0.25) Constant 1.23* 1.23* 1.23** 1.23** 1.23* 1.23* 1.23** 1.23** (0.56) (0.56) (0.51) (0.51) (0.56) (0.56) (0.51) (0.51) R20.68 0.68 0.68 0.68 0.68 0.68 0.68 0.68 N. of observations 36 Significance at the 90%, 95%, and 99% confidence levels are indicated by *, **, and ***, respectively. 70