Abstract
This paper contributes to the understanding of the long-run consequences of Roman rule on economic development. In ancient times, the area of contemporary Germany was divided into a Roman and a non-Roman part. The study uses this division to test whether the formerly Roman part of Germany are more developed than the non-Roman part. This is done using the Limes Germanicus wall as geographical discontinuity in a regression discontinuity design framework. The results indicate that economic development—as measured by luminosity—is indeed significantly and robustly larger in the formerly Roman part of Germany. The study identifies the persistence of the Roman road network until the present as an important factor causing this developmental advantage of the formerly Roman part of Germany both by fostering city growth and by allowing for a denser road network.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
By adopting a BDD to identify the causal effect of Roman transport infrastructure, the paper adds to a growing literature that exploits geographical or political discontinuities in space to identify causal effects of certain variables on economic outcomes like e.g., luminosity, institutional quality or political attitudes (e.g., Becker et al. 2015; Dell 2010; Dell et al. 2015; DehdariandGehring 2016; Fontana et al. 2016; Michalopoulos and Papaioannou 2014; Ochsner and Roesel 2016; Pinkovskiy 2017; Schumann 2014; Wysokinska 2015).
As discussed in detail in Sect. 3, the fact that the border is identical to the Danube or the Rhine in many countries gives rise to the problems of compound treatment, e.g. due to the fact that the Rhine was also the border of e.g., the states conquered by the French during the Napoleonic period. A border that is located in a mountainous area might also not be suitable for such an analysis as no large-scale economic activity can be found in mountain regions (independent of Roman treatment). Other countries divided by the historical Roman border are located in Northern Africa. However, most of the Roman parts of North Africa are located in the Sahara, making large settlements anywhere other than the coastal area almost impossible. Moreover, the findings of Bosker et al. (2013) suggest that Roman roads played no or only a very limited role for the development of cities in North Africa and the Middle East.
The segment that is a straight line is also clearly visible in Figure B.1. in the Appendix.
He also built the so called “Odenwald-Limes” (a more or less straight line connecting the Main area with the Neckar).
Appendix B.1.2 gives additional information on the evolution of the Roman border in Germany from the beginning of the conquest of the area in around 30 BC until the end of the Roman Empire in the 5th century. I also conducted regressions for those segments of the border that were held by the Romans until the end of the Empire. This area (roughly corresponding to the part of Germany south of the Danube and west of the Rhine) experienced Roman rule for a longer period than the area that was given up after the Limes wall was invaded. Results for these two segments are almost identical to those for the whole Limes area, but, with respect to the area south of the Danube the results indeed show a larger effect suggesting that this area might have profited from longer Roman rule (more on this in Appendix B.1.2).
The sources for the Celtic and Roman settlements and their locations are described in Appendix B.2.
Archaeologists and pre-historians speak of the “Helvetier-” and “Boier-Einöde” (“Helvetii-” and “Boier desert”) to indicate that there are almost no findings from this period in this area (Rieckhoff and Biel 2001; Rieckhoff 2008) which was later one of the core areas of Roman Germany. And Buringh et al. (2012) note that this area of Germany was “largely devoid of settlements before the Romans arrived” (Buringh et al. 2012, p. 12).
On the equator this is equivalent to a grid cell area of 0.86 km\(^2\).
The kernel density estimate of the probability distribution of luminosity is reported in Figure A.1 in the Appendix. There are also around 7% of zero values. However, these zeros very likely do not reflect areas with absolutely no economic activity but are due to some “bottom-coding” because satellites cannot distinguish very low levels of luminosity (roughly below 3) from luminosity. More on this in Appendix A.1.1.
All sources on which the shapefile is based are available here http://www.euratlas.net/shop/bibliography/gis_200.html. A generally positive review of the collection of digital maps on which the shapefile is based is Gibbs (2011).
A map showing the course of the Limes is available from the “Deutsche Limeskommission ” (German Limes committee) here http://www.deutsche-limeskommission.de/index.php?id=6&L=1; accessed on September 26th, 2015.
Finally, I check whether the results change substantially when I exclude the area 5 km around the border to ensure that the results are not driven by inaccuracies of the exact course of the Limes.This check is somehow related to the idea of a “Donut RDD” (Barreca et al. 2011). Results turned out to be robust to this exclusion. Results of estimating Table 3 without the border area are reported in Table B.10 in the Appendix. To further test whether the results are driven by particular areas within the studied region that, e.g., have a very high or very low level of luminosity, I separate the studied region (i.e., the buffer area 100 km to the south and north of the Limes) into 20 equally sized 5 km distance bands. That is, the first distance band represents the area 0–5 km away from both sides of the border, the second one the area 5 to 10 km away and so on, and then re-run the BDD from Table 3, column (6) leaving out one of these distance bands in each regression. The results are shown in Appendix B.4.5, Table B.11. Overall, the coefficients remain quite stable and always significant. Hence, the results are not driven by particular areas of the sampling region.
The course, building and characteristics of Roman roads have been extensively studied by historians and archaeologists (e.g., Laurence 1999). From such works the Roman road network can be reconstructed with some certainty. Nevertheless, for some of the roads there is a scholarly disagreement about their exact course. These roads are classified in the Barrington Atlas as being uncertain and are excluded from the subsequent analysis.
An additional natural factor that was of considerable importance for ancient and pre-historic societies was climate. However, I do not include a measure of climate as control variable due to several reasons. First, I am not aware of any large-scale dataset on pre-medieval climate in Europe. Existing data sets start in the medieval period and have such a high spatial resolution (e.g., 5 degrees in latitude and longitude in the case of the Guiot and Corona 2010 data set) that the whole studied area would fall into one grid cell. Second, I also control for segment fixed effects, higher order polynomials of latitude and longitude and study an area with a maximum of 200 km north-south extent, I presume that climate should not vary much within this comparatively small area.
This is probably not true for a small area in today’s Hesse (approximately at the point where the border has a turning point to north–south instead of east–west orientation. This area is the “Wetterau”, which is an area with particularly high soil quality that the Romans wanted to secure for their own purposes. However, as I control for agricultural suitability this should be no concern. Furthermore, I show that the positive discontinuity in economic development also holds if I focus only on the segments of the border without the Wetterau area (see Table B.12 in the Appendix).
As throughout the paper, the order of the distance polynomial used in the respective columns is chosen according to the Akaike information criterion (AIC). The AIC is most commonly used to select the order of polynomials in RDD analyses (Lee and Lemieux 2010). It enables to choose among statistical models by balancing the fit of the model (measured by the variance of the residuals) and the complexity of the model (measured by the number of parameters). Its formula is \(AIC=Nln(\hat{\sigma }^2)+ 2p\). Where \(\hat{\sigma }^2\) is the mean squared error of the regression and p is the number of parameters in the regression model (Lee and Lemieux 2010). The model with the lowest AIC value is selected.
In Table C.3 I check for the continuity of soil suitability only taking into account crops that were available before the Columbian Exchange (before 2000). I do so by using the average caloric suitability index developed by Galor and Özak (2015, 2016). This measure has the additional advantage of being historically more accurate as it also assumes low input agriculture and abstracts from irrigation methods. However, it comes with the cost of a much higher spatial resolution of around 85 km\(^{2}\). It turns out that caloric suitability for pre-Columbian Exchange crops is continuous at the Limes border for 15, 10 and 5 km bandwidths.
For more on this matter the reader is referred to Appendix B.3. There, Figure B.11 shows the Holy Roman Empire and its sovereign states in the year 1378, after the death of Charles IV. as well as the Roman border in 200 CE. Figure B.12 shows the borders of the states of the German Empire, in particular that of Prussia in 1871 and the Roman border in 200 CE. Finally, Figure B.18 shows the spatial distribution of Protestants and Catholics in the HRE around 1560. It also suggests that the historical Roman border did not reflect the schisms and religious upheavals of early-modern Europe.
Buringh et al. (2012) show that settlement densities along the lower Germanic Limes, which was completely identical to the Rhine, were larger than in the regions along the landlocked parts of the German Limes. They estimated that, even when accounting for military demand, the presence of a navigable river lead to an additional 28% higher urbanization. Thus, an endogeneity problem is very likely.
Descriptive statistics of the actual estimation sample for the BDD regressions can be found in the Appendix Table A.1.
Results for re-estimating Table 1 for those buffer areas but without the river segments of the border are shown in Table C.1 in the Appendix.
Actually, to keep the zero values in the data, here and throughout the whole analysis I use ln(\(1+Luminosity_{s,i}\)) as the dependent variable. Results would be virtually identical if one would exclude the zero values from the luminosity data set (see Appendix A.1.1 for a discussion of this issue). The results would also not be affected by using the unlogarithmized version of the luminosity variable.
As the segments that are identical to Rhine and Danube are excluded I consider only the border segments 2–4 in the BDD estimation sample. I also estimated the semi-parametric BDD using ten instead of five equally long border segments. The results of those regressions are reported in Appendix B.4.6, Table B.13. They are virtually identical to those with five border segments, if anything they are even strong with respect to the estimated size of the effect.
I will use 2nd or 3rd order coordinates polynomial of the following form: \(f(x,y)=x+y+xy+x^2+y^2+x^2*y+y^2*x+(x^3+y^3)\).
In all figures the bins are chosen according to the IMSE-optimal evenly-spaced method using polynomial regression.
Is this a large effect when compared to other historical treatments that happened long ago? To get an idea of this it is useful to compare this effect to those found by other studies on long-run impacts of treatments that happened a very long time ago. Ashraf and Galor (2013) assess benefits and costs of within country genetic diversity that was determined thousands of years ago by the “Out of Africa” migration process and find very large effects on contemporary development levels, especially for countries away from the optimum. They also find that around 16% of today’s variation in income per capita is explained by genetic diversity. Putterman (2008) finds that between one third and half of cross-country income p.c. differences in 1997 is explained by different transition years to agriculture. When I estimate a bivariate regression of the distance to Roman border variable on the natural logarithm of luminosity I find that around 6% of the variation in nighttime luminosity is explained by distance to the Roman border. In light of the results of the other studies, this seems not to be out of range (regressions available upon request). Finally, Bosker et al. (2013) and Bosker and Buringh (2017) explain the location and long-run development of cities in Europe and the MENA region. Their estimates for the effect of being located on a Roman road and a hub of Roman roads is around 0.10 and hence in a similar range to the effect estimated in this study.
As the year 2010 might still be too close to the financial crisis I also re-estimate Table 3 using luminosity values from the year 2000. Results are available in Table C.2 in the Appendix and show a qualitatively similar results even suggesting a slightly larger development advantage of Roman Germany. Grid cell population originates from version 4 of the Gridded Population of the World shapefile for 2010. It can be downloaded after free registration from here: http://beta.sedac.ciesin.columbia.edu/data/set/gpw-v4-population-count/data-download (accessed last on October 20th, 2016). As in the case of luminosity I include ln(\(1+Population_{s,i}\)) into the regression to keep grid cells with zero population. Only 0.006% of all grid cells have a population value of zero. Hence, the results do not change when leaving them out (regressions available upon request).
The radiance calibrated luminosity data is provided by the same source as the standard luminosity. It can be downloaded from this url: http://ngdc.noaa.gov/eog/dmsp/download_radcal.html (accessed on September 10th, 2016). However, one of the shortcomings of the radiance-calibrated luminosity series is, that it is measured with a slightly lower quality than the top-coded luminosity measures. The bivariate correlation between standard luminosity and radiance calibrated luminosity is very high (0.847).
There are 3,995 municipalities within our 100 km buffer area around the Limes. Average municipality level radiance calibrated luminosity is 20.481 with a standard deviation of 17.61. The 40 cities with the highest luminosity levels (1% of all municipalities) are those with values larger than 96.5357 (the maximum is 186.986). Among them, 35 are located in the Roman area and five in the non-Roman area. Municipal borders are taken from a shapefile provided by official sources (the “Bundesamt für Kartographie und Geodäsie”). It can be downloaded from here: http://www.geodatenzentrum.de/geodaten/gdz_rahmen.gdz_div?gdz_spr=deu&gdz_akt_zeile=5&gdz_anz_zeile=1&gdz_unt_zeile=13&gdz_user_id=0; accessed on December 19th, 2016.
Appendix B.4.8.2 discusses the sources on which this assignment is based an also shows maps of the spatial distribution of Protestant and Catholic territories and grid cells (Figures B.18 and B.19).
Appendix B.4.8.4 gives more information e.g., provides a descriptive discussion of recent regional migration patterns in Germany.
Another relevant concern refers to differences resulting from different allied policies in the years between 1945 and 1949, when Germany was occupied and split in four different occupation zones. As shown by recent research (Ochsner and Roesel 2016; Schumann 2014) the allied countries pursued different policies in their occupation zones in post-war Germany and Austria (e.g., regarding openness for refugees) that had significant effects on the subsequent development of these areas. To rule out that my results reflect differences in the policies of the allied countries, I assign each grid cells in the data set to either the French, British or US occupation zone. In the next step, I include dummy variables for grid cells in the French and British zone to the regression specification reported in Table 3, column (6) and look at how the estimated coefficients change. Results are shown in Table B.15, column (3) in the Appendix. The border of the occupation zones are based on the shapefiles of Schumann (2014) and on the borders of contemporary German states. For more information the reader is referred to Appendix B.4.8.3. The estimated coefficient of the Roman area dummy is 0.115 (\(p<0.01\)). Hence, the results in Table 3 are not driven by differences between the former allied occupation zones.
This is important, as the 50 km shift, ensures that only actually untreated or only treated units are included in the estimation sample, regardless of the chosen bandwidth.
An alternative—non-parametric way—of conducting such a placebo border test is to run a Zivot–Andrews structural breakpoint test. This is done in Appendix B.4.9 and also suggests that there is a structural break in luminosity at the historically Roman border.
Among historians, one can find different opinions about the long-run importance of Roman roads. Bairoch (1988) or Lopez (1956) for example, are skeptical about the importance of Roman roads for medieval trade. They doubt that many of the important Roman roads were maintained or represented the most cost-saving path to the trade centers.
Furthermore, the dense road network connecting Frankfurt am Main and the Rhine-Neckar area with Saarbrücken (on the mid-west of the map) probably also follows historical Roman roads as Saarbrücken and Frankfurt began as Roman settlements. However, McCormick et al. (2013b) classified these roads (or their course) as uncertain and thus I do not consider them in the analysis, leading me to underestimate the possible persistence of the road network.
Another channel through which the persistent importance of the Roman cities could be explained is because of the surviving ecclesiastical administration in the Roman bishoprics (Cologne, Mainz, and Trier have remained important bishoprics until today).
It actually also suggests that the OLS results are downwards biased as the coefficient from an OLS regression of distance to highway on luminosity with the same controls would yield a significant but notably lower coefficient of −0.263. Regression not shown but available upon request.
This would also work with a dummy variable indicating grids that intersect a highway. Regression not shown but available upon request.
This result also holds if one were to control for luminosity to account for the fact that the higher road density could also be the result of higher economic development that in turn could have been the result of higher levels of urbanization and agglomeration caused by Roman legacies. The inclusion of luminosity would reduce the coefficient to −0.0544 which would still be significant at the 1% level.
The data can be downloaded here: http://www.cgeh.nl/sites/default/files/def%20europe.xls; accessed on July, 10th, 2015.
Information about Roman settlements is taken from the shapefile “Europe in 200 AD” provided by Euratlas Nüssli and Nüssli (2008).
From the findings of the long-run city development literature one can argue that other factors, like e.g., the ecclesiastical or commercial importance of a city should also be included here. However, it is likely that these variables, e.g. a dummy for residences of archbishops or medieval trade centers, are endogenous to the Roman treatment and hence would be bad controls.
Figure C.3 in the Appendix shows the locations of the cities (cities on the Roman side of the border in red and cities on the non-Roman side of the border in blue), their size in 1800 (Figure C.3a) and 2000 (Figure C.3b) indicated by the size of the dots, as well as the Roman road network. The visual impression suggests that cities on the Roman side of the border seem to be larger on average than their counterparts in the non-Roman area.
In Appendix B.4.10 (Table B.17) I also include a dummy for cities becoming Protestant in the 16th century as additional control to rule out that north–south differences in religion drive the results. It turns out that Protestantism is always insignificant and that the coefficients are almost identical.
I code a city as being located on a Roman road if it is located within a 5 km buffer around the road.
Again, one can ask whether these large effects are reasonable. While it is right that 60% is a very large effect, it is not out of the range of effects found by previous studies on long-run city development. Bosker et al. (2013) for example, estimate that being a capital city increases population on average by around 90%. They also estimate that residence cities of archbishops or of cities located on the coast are on average around 40% larger, roughly corresponding to the size advantage I estimated for cities in the Roman part of Germany.
More on these data and the construction of the famous people migration index can be found in Appendix B.4.8.4.
I do not include the Celtic settlement dummy, the distance to Heuneburg and the distance to the closest Roman market and mine variables to this regression, as they were always insignificant in Table 8—and would be insignificant in this regression.
Another way to look at the temporal evolution of the Roman effect is to run separate cross-sections for each century. This has the advantage that one can directly interpret the estimated coefficients as the effect of Roman legacies in the respective century. However, it comes with the cost of a considerably reduced sample size making the estimates more imprecise. Nevertheless, I estimate separate cross-sections of the regression in Table 8 column (1) for the whole sample of cities (including the river areas) and for the 15th century, in order to have enough observations. Results are shown in Appendix B.4.12, Table B.20. I find significant and positive effects beginning in the 16th century with a notable decrease in the coefficient size in the 18th century. Hence, the results of the separate cross-sections do in principle confirm the pattern observed when interacting the Roman dummy with the century fixed effects.
References
Acemoglu, D., Cantoni, D., Johnson, S., & Robinson, J. A. (2011). The consequences of radical reform: The french revolution. American Economic Review, 101, 3286–3307.
Acemoglu, D., Johnson, S., & Robinson, J. A. (2001). The colonial origins of comparative development: An empirical investigation. American Economic Review, 91, 1369–1401.
Ashraf, Q., & Galor, O. (2013). The ’Out of Africa’ hypothesis, human genetic diversity, and comparative economic development. American Economic Review, 103, 1–46.
Baatz, D. (2000). Der römische Limes. Archäologische Ausflüge zwischen Rhein und Donau. Berlin: Gebr. Mann Verlag.
Bairoch, P. (1988). Cities and economic development. From the dawn of history to the present. Chicago, IL: University of Chicago Press.
Bairoch, P., Batou, J., & Chevre, P. (1984). The population of European cities from 800 to 1850. Geneva: Librairie Droz.
Barreca, A. I., Guldi, M., Lindo, J. M., & Waddell, G. R. (2011). Saving babies? Revisiting the effect of very low birth weight classification. Quarterly Journal of Economics, 126, 2117–2123.
Becker, S. O., Boeckh, K., Hainz, C., & Woessmann, L. (2015). The empire is dead, long live the empire! Long-run persistence of trust and corruption in the bureaucracy. Economic Journal, 126, 40–74.
Becker, S. O., Pfaff, S., & Rubin, J. (2016). Causes and consequences of the protestant reformation. Explorations in Economic History, 62, 1–25.
Berger, T., & Enflo, K. (2017). Locomotives of local growth: The short- and long-term impact of railroads in Sweden. Journal of Urban Economics, 98, 124–138.
Bleakley, H., & Lin, J. (2012). Portage and path dependence. Quarterly Journal of Economics, 127, 587–644.
Bosker, M., & Buringh, E. (2017). City seeds: Geography and the origins of the European city system. Journal of Urban Economics, 98, 139–157.
Bosker, M., Buringh, E., & Van Zanden, J. L. (2013). From Baghdad to London. Unraveling urban development in Europe, the Middle East, and North Africa, 800–1800. Review of Economics and Statistics, 95, 1418–1437.
Buggle, J. C. (2016). Law and social capital: Evidence from the code napoleon. European Economic Review, 87, 148–175.
Buringh, E., van Zanden, J. L., & Bosker, M. (2012). Soldiers and booze: The rise and decline of a Roman market economy in north-western Europe. CGEH working paper no. 32.
Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014a). Robust nonparametric confidence intervals for regression-discontinuity designs. Econometrica, 82, 2295–2326.
Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014b). Robust data-driven Inference in the regression-discontinuity design. The Stata Journal, 13, 1–36.
Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2011). Robust inference with multiway clustering. Journal of Business & Economic Statistics, 29, 238–249.
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.
Cogneau, D., & Moradi, A. (2014). Borders that divide: Education and religion in Ghana and Togo since colonial times. Journal of Economic History, 74, 694–729.
Dalgaard, C.-J., Kaarsen, N., Olsson, O., & Selaya, P. (2017). Roman roads to prosperity: The long- run impact of infrastructure on economic activity. Presentation at London School of Economics. url: https://dl.dropboxusercontent.com/u/17139768/roman_slides.pdf.
de la Croix, D., & Licandro, O. (2015). The longevity of famous people from Hammurabi to Einstein. Journal of Economic Growth, 20, 263–303.
Dehdari, S., & Gehring, K. (2016). Building national identiy? The causal effect of homogenization policies in Alsac-Lorrain. CIS working paper no. 89.
Dell, M. (2010). The persistent effects of Peru’s mining mita. Econometrica, 78, 1863–1903.
Dell, M., Lane, N., & Querubin, P. (2015). State capacity, local governance and economic development in Vietnam. Mimeo.
Dittmar, J. E. (2011). Information technology and economic change. The impact of the printing press. The Quarterly Journal of Economics, 126, 1133–1172.
Fontana, N., Nannicini, T., & Tabellini, G. (2016). Historical roots of political extremism. The effects of Nazi occupation of Italy. CEPR discussion paper, no. 11758.
Galor, O., & Özak, Ö (2015). Land productivity and economic development: Caloric suitability vs. agricultural suitability. Brown University working paper.
Galor, O., & Özak, Ö. (2016). The agricultural origins of time preference. American Economic Review, 106, 3064–3103.
Gibbon, E. (Ed.). (1983). Histoire du d’eclin et de la chute de l’Empire Romain. Paris: Robert Laffont.
Gibbs, P. (2011). Marc-Antoine Nüssli & Christos Nüssli. Euratlas Periodis Expert: Periodical Historical Atlas of Europe 1–2000 (digital cartography program). Antiquity, 85, 1101–1102.
Giuliano, P., & Nunn, N. (2017). Understanding cultural persistence and change. Mimeo.
Glick, T. F. (1979). Islamic and christian Spain in the early Middle Ages. Princeton, NJ: Princeton University Press.
Guiot, J., & Corona, C. (2010). Growing season temperatures in Europe and climate forcings over the past 1400 years. PLoS ONE, 5(4), 1–15.
Henrich, P., (Ed.) (2012). Der Limes vom Niederrhein bis an die Donau. Vol. 6. Beitra ge zum Welterbe Limes. Stuttgart: Konrad Theiss Verlag.
Henrich, P., (Ed.) (2014). Der Limes in Raetien, Ober- und Niedergermanien vom 1. bis 4. Jahrhundert. Vol. 8. Beiträge zum Welterbe Limes. Darmstadt: Konrad Theiss Verlag.
Hohenberg, Paul M., & Lees, L. H. (1995). The making of urban Europe, 1000–1994. Cambridge, MA: Harvard University Press.
Holl, A. (2004). Manufacturing location and impacts of road transport infrastructure: Empirical evidence from Spain. Regional Science and Urban Economics, 34, 341–363.
Imbens, G. W., & Lemieux, T. (2008). Regression discontinuity designs: A guide to practice. Journal of Econometrics, 142, 615–635.
Imbens, G., & Kalyanaraman, K. (2012). Optimal bandwidth choice for the regression discontinuity estimator. The Review of Economic Studies, 79, 933–959.
Jacob, M. (2010). Long-term persistence: The free and imperial city experience in Germany. Mimeo.
Jedwab, R., Kerby, E., & Moradi, A. (2017). History, path dependence and development: Evidence from colonial railroads, settlers and cities in Kenya. Economic Journal. doi:10.1111/ecoj.12347.
Keele, L., & Titiunik, R. (2014). Geographic boundaries as regression discontinuities. Political Analysis, 22, 814–863.
Kristiansen, K. (2000). Europe before history. Cambridge: Cambridge University Press.
Kuckenburg, M. (2000). Vom Steinzeitlager zur Keltenstadt. Siedlungen der Vorgeschichte in Deutschland. Stuttgart: Konrad Theiss Verlag.
Laurence, R. (1999). The roads of Roman Italy: Mobility and cultural change. London: Routledge.
Lee, D. S., & Lemieux, T. (2010). Regression discontinuity designs in economics. Journal of Economic Literature, 48, 281–355.
Lopez, R. S. (1956). The evolution of land transport in the Middle Ages. Past and Present, 9, 17–29.
Margary, I. D. (1973). Roman roads in Britain. London: John Baker Publishing.
McCormick, M. (2001). Origins of the European economy. Communication and commerce AD 300–900. Cambridge: Cambridge University Press.
McCormick, M., Huang, G., Zambotti, G., & Lavash, J., (Eds.). (2013a). Digital atlas of Roman and medieval civilizations. Cambridge, MA: DARMC, Center for Geographic Analysis, Harvard University.
McCormick, M., Huang, G., Zambotti, G., & Lavash, J. (2013b). Roman road network (version 2008). DARMC Scholarly Data Series. Data Contribution Series No. 2013-5.
Menghin, W. (1995). Kelten, Römer und Germanan. Archäologie und Geschichte in Deutschland. Augsburg: Weltbild Verlag.
Michaels, G., & Rauch, F. (2017). Resetting the urban network: 117–2012. Economic Journal. doi:10.1111/ecoj.12424.
Michalopoulos, S., & Papaioannou, E. (2014). National institutions and subnational development in Africa. Quarterly Journal of Economics, 129, 151–213.
Nunn, N. (2008). The long-term effects of Africa’s slave trades. Quarterly Journal of Economics, 123, 139–176.
Nüssli, C., & Nüssli, M.-A., (Eds.) (2008). Georeferenced historical vector data 200. Euratlas, www.euratlas.com.
Ochsner, C., & Roesel, F. (2016). Migrating extremists. CESifo working paper no. 5799.
Okoye, D., Pongou, R., & Yokossi, T. (2016). On the dispensability of new transportation technologies. Evidence from colonial railroads in Nigeria. MPRA paper no. 75262.
Pinkovskiy, M. L. (2017). Growth discontinuities at borders. Journal of Economic Growth, 22, 145–192.
Pirenne, H. (1944). Medieval cities: Their origins and the revival of trade. Princeton, NJ: Princeton University Press.
Planck, D., & Beck, W. (1987). Der Limes in Südwestdeutschland. Stuttgart: Konrad Theiss Verlag.
Planitz, H. (1966). Die deutsche Stadt im Mittelalter. Von der Römerzeit bis zu den Zunftkämpfen. Wiesbaden: VMA-Verlag.
Putterman, L. (2008). Agriculture, diffusion and development: Ripple effects of the neolithic revolution. Economica, 75, 729–748.
Rieckhoff, S. (2008). Geschichte der Chronologie der Späten Eisenzeit in Mitteleuropa und das Paradigma der Kontinuität. Leipziger Online-Beiträge zur Ur- und Frühgeschichtlichen Archäologie, No. 30.
Rieckhoff, S., & Biel, J. (Eds.). (2001). Die Kelten in Deutschland. Stuttgart: Konrad Theiss Verlag.
Riemer, U. (2006). Die Römische Germanienpolitik. Von Caesar bis Commodus. Darmstadt: Wissenschaftliche Buchgesellschaft.
Scarre, C. (Ed.). (2000). Atlas de rome antique. Paris: Autrement.
Schallmayer, E. (2011). Der Limes. Geschichte einer Grenze. Munich: C.H. Beck.
Schich, M., Chaoming, S., Yong-Yeol, A., Mirsky, A., Mauro, M., Barabasi, A.-L., et al. (2014). A network framework for cultural history. Science, 345, 558–562.
Schumann, A. (2014). Persistence of population shocks: Evidence from the occupation of West Germany after World War II. American Economic Journal. Applied Economics, 6, 189–205.
Seidel, T., & Von Ehrlich, M. (2015). The persistent effects of regional policy. Evidence from the West-German Zonenrandgebiet. CESifo working paper series no. 5373.
Talbert, R. J. A. (Ed.). (2000). Barrington atlas of the Greek and Roman world. Princeton, NJ: Princeton University Press.
Van Zanden, J. L. (2008). The Road to the Industrial Revolution: Hypotheses and Conjectures about the Medieval Origins of the European Miracle. Journal of Global History, 3, 337–359.
Von Schnurbein, S. (1995). Vom Einfluss Roms auf die Germanen (Vol. 331). Vorträge: Nordrheinwestfälische Akademie der Wissenschaften.
Wolters, R. (2011). Die Römer in Germanien. Munich: C.H. Beck.
Wysokinska, A. (2015). Invisible wall. Role of institutions and culture in long term development. Natural experiment. In Proceedings of the EEA annual congress 2015.
Zabel, F., Putzenlechner, B., & Mauser, W. (2014). Global agricultural land resources. A high-resolution suitability evaluation and its perspectives until 2100 under climate change conditions. PLoS ONE, 9, 1–12.
Acknowledgements
The author would like to thank the editor (Oded Galor) and three anonymous referees for their helpful and constructive comments which substantially improved the paper. Furthermore, he gratefully acknowledges the comments of Davide Cantoni, Sibylle Lehmann-Hasemeyer and Aderonke Osikominu as well as seminar participants in Hildesheim, Hohenheim, Mannheim and Warwick, especially Jan Ditzen, Andreas Peichl and Agnieszka Wysokińska for valuable comments.
Author information
Authors and Affiliations
Corresponding author
Electronic supplementary material
Below is the link to the electronic supplementary material.
Rights and permissions
About this article
Cite this article
Wahl, F. Does European development have Roman roots? Evidence from the German Limes. J Econ Growth 22, 313–349 (2017). https://doi.org/10.1007/s10887-017-9144-0
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10887-017-9144-0