Poor soil as a fertile breeding ground: the role of historical agricultural specialization for the persistence of regional differences in crafts

Historically the term “crafts” was used to signify most commercial activities that were unrelated to agricultural production or trade. In Germany, craftsmen were organized in guilds which started to decline after 1500 and were replaced by a more competitive market structure (Ogilvie 2014). Before 1930, the terms “small business” and “crafts” were used interchangeably, in order to distinguish a business from large-scale manufacturing (see discussion in Knöpp 1930: 887). However, the term “crafts” started to acquire a distinct legal meaning in 1897, when mandatory membership in trade and crafts chambers was introduced (Handwerkergesetz [crafts law] 1897). If a business belonged to any one of the listed crafts occupations, the owner had to pay membership dues. In return, the crafts chambers took over a number of administrative and educational tasks in addition to constituting an interest group, lobbying the central government on behalf of their members.

While the term “crafts” is no longer used to describe the small business sector in English speaking countries (where it now pertains to “arts and crafts” or “handicrafts”) it still plays a vital role in Germany, where about 13% of the labor force are working in the crafts sector in 2016 (Federal Statistical Office 2019), and it is currently subject to policy debates with regard to regulation and the free movement of labor within the European Union (see Runst 2018; Runst et al. 2019, Rust and Thoma 2020; Rostam-Afschar 2014). Because the sector was historically defined in opposition to large-scale industries, it represented a motley collection of commercial activities (Knöpp 1930). Nevertheless, the prescribed list of commercial activities labeled as crafts displays a strong continuity over time and most can be described as lower-tech SMEs. For example, butchers, bakers, hairdressers, brewers, opticians, and most construction activity¹ belong to the crafts sector around the year 1900 (Imperial Statistical Office of Germany 1908) as well as today.

We can use data provided by the Imperial Statistical Office as of 1895 in order to evaluate the continuity of the sector over time in more detail. It lists all crafts trades as well as the number of businesses that can be found in a broad sample of German regions. We reproduce this list in the appendix (see Table A.3) and provide an English translation for each crafts trade. To each historical trade title, we also assign the equivalent title that can be found in the current version of the German Trade and Crafts Code. Alternatively, we print “obsolete” if the trade no longer exists. In the final column, we evaluate whether the crafts trade displays continuity in the type of product or service that is being rendered. For example, the trades of hair dressers, bricklayers, and bakers display a strong continuity, as the service is essentially the same as it was 100 years ago although the use of technology has increased. The trade of “Bandagist,” which is now called “orthopedic technician” for example, has become infused with advanced technologies (3D-scanning and printing) while still providing the same type of product/service. The number of businesses in the year 2018 can be obtained from the official crafts statistics of the Federal Statistical Office.

Most trades that existed in 1895 are being recognized by the current Trade and Crafts Code. Apart from the few trades that have become obsolete, and which represent only a negligible share of all business, almost 100% of all businesses in a sample from 1895 would be deemed to be crafts companies under contemporary German law. Reversely, about 76.4% of today’s crafts companies can be found on the historical list of trades. The trades that exist today but did not exist 100 years ago are automotive repair, cleaning, electrotechnical trades, photography, optician, and dental technicians.

Despite one hundred years of technological progress, which changes how products and services are created, the degree of similarity in what kinds of products and services are being produced is remarkable. The similarity is a result of the legal character of the crafts definition which displays a strong continuity throughout the last 100 years. In this paper, we exploit this continuity in the definition of crafts occupations as it allows us to use statistics about the geographic distribution of craftspeople in the past and examine its impact one the present geographic distribution.

Firms benefit from spatial concentration and co-location in various ways (Audretsch and Feldman 1996; Porter 2000; Torre 2008; McKelvey et al 2003; Murata et al. 2014; Devereux et al. 2007; Felzensztein et al. 2010). Much previous research has focused on knowledge spillovers and the innovation benefits of co-location (see Audretsch and Feldman 1996; Porter 2000; Murata et al. 2014), even though there are also production benefits (Torre 2008), marketing benefits (Felzensztein et al. 2010), and benefits from specialized thick labor markets (Addario 2011).

Authors in the knowledge-transfer-literature argue that co-location facilitates interaction by generating information networks and shared informal institutions. Tacit knowledge, as opposed to codified knowledge, cannot be communicated over longer distances. It can be transferred by observation, imitation, and learning by doing, all of which benefits from geographic proximity. Empirical research has mostly focused on larger firms and for example found robust links between the number of firm patents and the geographic distance to related innovations, as measured by cited patents (Jaffe et al. 1993; Thompson and Fox-Kean 2005; Murata et al. 2014). Torre (2008) argues that the benefits of co-location for knowledge transfer should especially accrue to small- and medium-sized firms, as they do not have the resources to overcome distance the same way larger firms can. Co-location benefits also accrue to firms through the existence of thick specialized regional labor markets (e.g., Addario 2011). As the number of skilled individuals increases, the likelihood of matching employer and worker with a suitable skill set is higher. Thus firms have an incentive to locate close to one another as they are competing for employees.

As firms make their location decisions under these circumstances, specific regional patterns of specialization emerge. Specialization patterns can only emerge in the tradable sector, as non-tradeable must be roughly distributed proportional to population and wealth (e.g., Moretti 2012). Once in place, the resulting synergies of co-location will persist over time. However, it is an empirical question for how long these specialization patterns will persist. This is particularly so given that we examine crafts persistence in Germany between 1904 and 2013, a period of two major wars and several political regime changes.

As we will see below, several high-density crafts regions in 2013 are specialized in certain crafts for decades (see Lehmann and Müller 2010). For example, the southern region of the state of Saxony displays a higher crafts density in 2013. It is dominated by woodworking and textile-related crafts as well as musical instrument making, all three of which existed around the year 1900 already. Similarly, the southern regions of Thuringia represent a higher-density crafts region. Again, we already find a high historical concentration of gun makers, which also exists presently even though it significantly decreased in size. Finally, the region around Tuttlingen in Baden-Württemberg was known for its surgical instrument makers, a crafts sector still in existence today.

There are several channels that explain persistence in the crafts sector both of which are indicating path dependency namely that past output matters for current and future economic performance (Martin and Sunley 2006a, b; Setterfield 1995). Martin and Sunley (2006a, b) discuss three types of path dependence that, which are also relevant in the context of the crafts sector. First, path dependence can be the result of technological lock-in. Second, there are formal and informal institutions, or “rules of the game” (North 1990), underlying economic activity. These institutions imply a self-perpetuation of certain types of economic activity and exchange over time. Third, path dependence can be prevalent because of dynamic increasing returns.

In general, path dependence is the result of positive feedback mechanisms and externalities, which has for example been applied to explain the persistence of entrepreneurial activity over time (Andersson and Koster 2011). In their conceptual model, persistence in entrepreneurial activity is driven by sticky regional characteristics and by localized externalities. Sticky regional characteristics comprise hardly changing structural determinants of entrepreneurial activity reflecting path dependence in the sense of technological lock. Localized externalities allude to the emergence of an entrepreneurial climate including learning about entrepreneurship via demonstration and peer effects. This is reflecting path dependence in the sense of dynamic increasing returns. Finally, entrepreneurial activity triggers social acceptance of entrepreneurship and leaves an imprint on informal and formal institutions that become more entrepreneurship-facilitating. This is reflecting the institutional aspect of path dependence (Martin and Sunley 2006a, b).

The primary channel of persistence in the crafts sector can likely be found in its reliance on tacit knowledge which is inducing path dependence. Knowledge is locally embedded as craftsman develop mastery over techniques that are largely uncodifiable and learning by doing (and imitation) represents the main form of instruction (see Sennett 2008). Tacitness is known to work as an important innovation protection mechanism, especially in smaller firms such as in the crafts sector (Nooteboom 1994; Thomä 2013). This also implies a certain regional specialization in the crafts sector like the abovementioned examples of Saxony and Thuringia demonstrate. This regional specialization indicates a technological lock-in while positive effects of learning (e.g., demonstration and peer effects) and transferring tacit knowledge imply dynamic increasing returns facilitating path dependence. Another channel for persistence can likely be found in the aforementioned thick regional labor markets as well as the intergenerational transmission of a craftsman’s identity (see Binder und Blankenberg 2020). This identity is reflecting a local informal institution in favor of crafts also inducing path dependence.

Nevertheless, perpetual turbulence is a feature of history in general, and market economies in particular, and thus the persistence of historical patterns will fade over time. However, as has been demonstrated by Fotopoulos (2014), and Fotopoulos and Storey (2017), turbulence and the corresponding infusion of novelty will most likely not be uniform across space. Some regions will have undergone a more radical economic and social transformation than others, thereby lowering the mechanisms of path dependency described above. For example, one may hypothesize that the considerable economic policy differences between East and West Germany between 1949 and 1990 differentially affect the persistence of economic patterns. The command and control system in the East may be expected to more actively steer economic development, thereby weakening path dependency and persistence. In addition, the growth of large industrial companies has the potential to transform the ecology of firms within its region. The rise of large firms, in addition to their suppliers and other linked companies will crowd out many of the incumbent smaller firms, such as the crafts. Finally, space represents a crucial resource for firm location, the competition for which has changed over the last 100 years. In particular, urbanization has affected the size and density of certain areas vis-à-vis others. In comparison to the year 1880, the largest ten German cities have increased their population by a factor of 2 (Leipzig, Dresden) to 7 (Cologne, Hamburg). The increasing competition for space will drive SMEs out of urban areas into the surrounding sub-urban belt, where they simultaneous benefit from agglomeration effects and lower land prices. Similarly, Acs and Miller (2017) argue that radical innovations are created in so-called campus ecosystems, which are largely urban phenomena, in which student entrepreneurs develop radical innovations. Thus, urban entrepreneurship will be more likely to change the nature of the entrepreneurial ecosystem, causing a stronger deviation from past economic structures. In contrast, rural economic structures develop gradually and incrementally. Similar to Fotopoulos and Storey (2017), who show that the London area displays higher rates of self-employment than can be expected given past values, we argue that urban areas represent centers of social and economic turbulence, thereby lowering path dependency in the crafts sector.

In addition to examining the existence of long-run persistence of regional economic patterns, we investigate how agricultural specialization shaped the emergence of economic activity. It is well established among historians that crafts activity emerged as a secondary occupation alongside agriculture (see Achilles 2008; Kellenbenz 1975, Kintzinger 2000: 207–245). One can easily conceive of a situation, which is historically well established, in which farming activities would initially concentrate in areas of higher agricultural productivity (such as mineral-rich river beds). Further population growth would lead to a progressive settlement of less fertile regions. Under such circumstances, rural farmers would take up additional commercial activities in order to supplement their incomes. In a pre-industrialized world, these activities would most likely be comprised of weaving and woodworking (and other non-perishables), and during the nineteenth century, as transportation networks and food conservation improved, baking, brewing, etc. In other words, small- and medium-sized businesses would be founded in areas of low agricultural productivity.

The historical data on regional crafts distribution is taken from various sources. As the administrative structure, i.e., the crafts chambers, came into existence after 1900, the first data was systematically recorded after that date. Historical statistics (1904, 1926, and 1931) do not contain the number of craftsmen on a regional basis. Instead, data on association memberships is available for those earlier years from the imperial statistical office (Kaiserliches Statistisches Amt 1908), which we use as a proxy variable for the distribution of craftspeople. The data of crafts association memberships is not available per county but on the basis of crafts chamber regions. Each crafts chamber was put in charge of a certain administrative district, each containing a number of counties. For the year 1904, we mapped counties onto administrative districts which largely coincide with crafts chamber districts. We crosschecked our mapping procedure by individually consulting historical anniversary publications (Festschrift) of all crafts chambers, in which the regional areas of the crafts district are being described. These anniversary publications aided us in tracking changes in craft chamber districts over time as well.

We calculated the relative share of each county within the crafts region in terms of population levels. This share is multiplied with the “crafts-data” and yields a “crafts-number” for every historical county. This is the basis for the regional adjustment. Although the definition of counties in the early twentieth century was different from what is defined as a county today, it is possible to assign the historical counties to current ones. If a historical county falls within two or more current counties, we assign data to the respective current counties based on each region’s share of the geographic area (for further details, see Fritsch et al. 2019: chapter 3).²

When we examined the distribution of crafts association membership per 1,000 inhabitants we recognized a strong north–south discrepancy, which caused us to reexamine the underlying data collection process for the regions of Bavaria and Baden-Württemberg. The history of the crafts associations took divergent paths in Prussian and non-Prussian territories in the nineteenth century. While the successor-organizations to the Guilds (Innungen) became more active in the North, the southern German regions followed in the tradition of the Napoleonic liberal tradition (Gewerbefreiheit) (see Wever 1914; Ullmann 1988, in particular pp. 95–97 and 156). Because of their more liberal tradition, the dominant associations in the North (Innungen) were less successful in the South, where craftsmen entered another type of associations (Gewerbevereine). “In northern Germany, there were 2980 Innungen with about 201,312 members, whereas in southern Germany [..] there were only 184 [Innungen] with 17,156 members” (Wever 1914: 786). This discrepancy still existed in the late 1920s, when it started to decline (Ullmann 1988). Given these differences in definition and due to the fact that information for many southern regions is missing, we did not consider areas located in Bavaria and Baden and Wuerttemberg.

The Federal Statistical Office of Germany and the German Confederation of the Skilled Crafts (Zentralverband des Deutschen Handwerks, ZDH) provide county-based data on the number of craftsmen for the year 2013. The distribution of our crafts density variables is plotted in Fig. 1.

We examine the persistence of crafts density in 1904 in the years 1926 and 2013. Crafts density is captured by the number of crafts association memberships per 1,000 inhabitants for the years 1904 and 1926. As we have explained above, we drop the territories of Bavaria and Baden-Württemberg because the density measure does not reflect the true number of crafts people in these regions. For the year 2013, the crafts density measure is equal to the number of craftsmen per 1,000 inhabitants.

The manufacturing employment share in 1907 and the distance to the closest coalfield (1907) are included to control for local economic structures. We also consider historical and current state dummy variables which are included as control variables.

The coefficient captures the average persistence across all regions. As described in the theory section however, we expect to find significant deviations from this average effect. To analyze the spatial pattern of persistence vis-à-vis transformation, we map the residuals of our baseline OLS regression. In addition, we regress the residuals against the dummy variable “East,” the z-transformed population density of 2013, and the z-transformed share of large companies (> 250 employees) in order to capture the effects of political regime change, urbanization, and crowding out, respectively. We also control for educational and age characteristics, as well as the share of non-German citizen. Education and migration in particular are expected to be associated with the density of crafts businesses as the crafts constitute a lower-tech SME sector and the share of crafts individuals with migration background is particularly high compared to other sectors (Runst 2018).

Furthermore, we follow Fotopoulos and Storey (2017) who develop and employ two additional dependent variables to examine the tension between persistence and transformation of economic structures. First, they use the change in self-employment over time. Analogously, we compute the change in the crafts density. Second, the authors generate a Rank Mobility Index (RMI):where RANK_t denotes the rank in the crafts density in 2013 and RANK_t − τ denotes the rank of the crafts density in 1904. The rank change is normalized by n − 1 in order to set a lower and upper bound to the index (− 1 and 1, respectively).

We also employ an instrumental variable (IV) regression approach by exploiting the fact that during the pre-industrial period individuals were more likely to enter a crafts profession if there were few other commercial opportunities, i.e., when agriculture was relatively unproductive (for details, see Sect. 2.3). Our research design does not suffer from endogeneity in the form of reverse causality because of the temporal structure of the data. However, the two-stage regressions allow us to model the effect of historical low productivity agricultural regions on the present crafts density through the channel of past crafts density, and it safeguards against other potential sources of endogeneity. The first stage of our IV regression takes the following form:where “oats_rye” is a dummy variable that is equal to one if the county was dominated by the cultivation of oats and rye in the year 1892, and zero otherwise. The cartographical oat and rye information is available in the Brockhaus-encyclopedia (1897). We used the geographic software QGIS for geo-referencing and digitizing the map. In Fig. 2, we can see that the geographic distribution of rye and oats cultivation is quite similar to the crafts density in the year 1904 and 2013.

Both oats and rye cultivations do not demand high-quality soil in order to grow (Behre 1992; Tiwari 2010; Knutsen 2010). Similarly, mountainous areas, as for the example border region in Saxony (Ore Mountains/Erzgebirge) and Bavaria (Bavarian Forest & the Alps), as well as Baden-Württemberg (Black Forest) and Rhineland-Palatinate (Eifel & Hunsrück) can be clearly identified as oat and rye agricultural regions.

We attempted to relate oat and rye to soil quality and topographical features of regions. However, the number of potential factors is large and requires an extensive study of its own. Because of this complexity, internationally comparable (uniform) methods for soil classification are largely missing (Müller et al. 2007). There is only one data source, that generates a soil quality rating for Germany which seeks to combine the multiple aspects of soil and topography into one soil quality rating (SQR, see Richter et al 2009; Müller et al. 2007), which is available from the Federal Institute for Geosciences and Natural Resources (FIGNR). However, as Figure A2 (appendix) shows, the historical oat and rye cultivating regions are conspicuously similar to the unrated areas (grey regions) in the German soil quality database. These areas are currently not used for agricultural production—which is why no ratings exist—and they are largely characterized by forests and grasslands. On request, experts of the FIGNR and the Thuenen Institute (sub-department of forest research) stated that despite the lack of SQR data, one can assume that unrated areas, i.e., areas currently not being used for agricultural production, are mostly low yield regions, and that low agricultural productivity is the primary reasons for its current status as non-agricultural land.

Even though no SQR rating exists for many oat and rye regions, there are additional indicators for the low agricultural productivity of these regions. Sandy and rocky soils, as well as clay soils lower the agricultural productivity of an area because they restrict rooting and negatively affect water run-off and storage (Müller et al. 2007), in addition to its slope. The cartographic depiction of German soil types shows that the oat and rye agricultural regions are largely characterized by rocky and clayey soils in addition to being situated in upland areas with steeper slopes. The northwestern oat and rye region overlaps with swampy soil, also an indicator of lower agricultural use.

To sum up, while comprehensive soil quality data does not exist, the evidence clearly supports our choice of oat and rye regions as a proxy for low agricultural yield. An area which suffered from lower agricultural productivity in the pre-modern era, as illustrated by its cultivation of oats and rye and its current status as non-agricultural land tended to be relatively poorer and individuals have taken up other commercial ventures. Before the advent of modern industrial production, such commercial activities are composed of small firm crafts production. As stated above, rural craftsmanship is known to be a secondary occupation alongside agriculture (Achilles 2008; Kellenbenz 1975).

In order to validate the oat/rye variable as an instrument, we need to assess the likelihood that there is a direct effect from being an oat and rye agricultural region to the present crafts density, i.e., the exclusion restriction. As the share of the labor force engaged in agricultural activity in 2013 is small (2.4%), the possibility of a direct link between soil quality and current crafts density can be dismissed.

Finally, we employ spatial econometric techniques. As we have argued that co-location benefits exist, the density of crafts in a neighboring county i + 1 may positively affect the crafts density in county i. We therefore include a spatially lagged dependent variable as a predictor in this specification. In addition, there is a possibility of spatially structured error term. For example, if there are unobserved factors (omitted variables) for which spatial autocorrelation exists, we can address this problem by also including a spatially lagged error term.

In this model, the crafts density in region r in 2013 is not only dependent on the crafts density in county r in 1904 but also on the crafts density of neighboring counties in 2013 and 1904. In addition, we allow for a spatial structure of the error term. For both spatial models, we use an inverse distance weighting matrix.

Descriptive statistics for all variables are presented in Table 1. The density of craftsmen in 1904, 1926, and 1931 is measured as the number of crafts association memberships per 1,000 inhabitants. The mean density in all German counties increases between 1904 and 1931. The crafts density in 2013 is measured as the number of craftsmen per 1,000 inhabitants. The variable emp_share_1907 represents the share of the labor force occupied in the industrial sector within a region. The distance to the nearest coalfield is measured in kilometers.

Table 2 reports regression results for the dependent variable “ln_density_1926.” The coefficient for the 1904 density variable is statistically significant and positive in all specifications (1–7). In specification (3), an increase in the crafts density in 1904 by 1 percent raises the crafts density in 1926 by 0.16 percent. If we restrict the sample to the state of Prussia, effect sizes increase to 0.19 percent in specification (6). The manufacturing employment share in 1904 has a moderately negative impact on crafts density in 1926, most likely because of regional specialization in either industrial manufacturing or crafts. Coalfield distance in 1904 affects the crafts density in 1926 positively. Again, the comparative advantage most likely favors industrial specialization closer to coalfields, whereas less developed areas would focus on crafts production.

Table 3 reports OLS regression results for the period from 1904 to 2013. Again, the coefficient for the density variable in 1904 is positive and significant, albeit of somewhat smaller effect size as a result of the longer time span. If the variable “number of craftsmen per 1,000 inhabitants” in 1904 increases by one percent, the dependent variable increases by 0.14 percent in specification (3). Unreported results for the Prussia sample, show similar effect sizes. Likewise, restricting the sample to West German states only (specifications 5 to 8) yields similar results.

The coefficient for the control variable manufacturing employment share is, again, negative and significant. The effect size is large; an increase in the manufacturing share by 1 percent, increases crafts density in 2013 by 0.33 to 0.42 percent, indicating persistent regional specialization. The coefficient for coalfield distance is no longer consistently significant across specifications.

The results of our instrumental variable regression can be found in Table 4 (first stage) and Table 5 (second stage). Oats and rye specialization in 1892 has a positive and statistically significant impact on the crafts density in 1904. If the agricultural production is mostly oat or rye the crafts density increases by 0.28 percent in specification (3). The first stage regression of our full model displays a large F-statistic of 46.16, which is above the often-cited value of 10. In addition, it exceeds the critical value of the Stock and Yogo test for the lowest expected bias of 5 percent (see Stock and Yogo 2005; Stock et al. 2002). Finally, we perform the Olea & Pflueger test (see Olea and Pflueger 2013; Pflueger and Wang 2015). Again, the F-statistic exceeds the critical value for a worst-case bias of 10 percent (as used in Olea and Pflueger 2013) but also the critical value for a worst-case bias of 5 percent (37.42). We can therefore safely reject the hypothesis of weak instruments.

The second-stage regression results can be found in Table 5. The density of craftspeople in 1904, explained by regional agricultural specialization in oats and rye, has a positive impact on the density of craftspeople in 2013. Effect sizes are larger than what can be found in the OLS results.

We run additional regressions, in which we examine the impact of the crafts density in 1904 on the crafts density in 1931 (instead of 1926), as well the impact of the density in 1931 on the density in 2013 (see Table A1). The results confirm previous findings, with similar effect sizes.

Table A2 presents the results for spatial models. When we control for spatially lagged independent variables and include a spatial error term, the coefficients of past density measures remain statistically significant and meaningful in terms of effect sizes. The spatially lagged dependent variable exerts only a minor and inconsistent influence. An increase in the crafts density of a neighboring county slightly lowers the crafts density of the county under consideration in specification 1, 2, and 6. In specification 4 however, in which we add the spatial lag of the 1904 density measure, only the spatially lagged error term exerts an influence. The results remain unchanged in a spatial IV setting (specifications 5 and 6). Overall, our general finding of structural persistence remains robust when considering potential spatial autocorrelation.

Figure 3.A plots the residuals of our baseline OLS regression for all sample regions, in which we regress the crafts density of 2013 against crafts density of 1904 and historical state dummies. There is a visible non-random spatial pattern. The urban regions of Frankfurt, Ruhr valley (Cologne, Düsseldorf, Duisburg, Essen, Dortmund), Kassel, Hannover, Bremen, Hamburg, Kiel, Berlin and Leipzig/ Halle, and Erfurt display considerably lower crafts density values than suggested by the model prediction. In contrast, the rural regions of Eifel, northwestern and northeastern Germany, the Thuringian/ Hessian forest regions and the German–Danish border region display considerably higher crafts density values than the model’s prediction. The results of Fig. 3 suggest that urbanization and increases in population density are related to lower persistence, whereas rural regions display higher persistence levels. The crafts share in the coastal region is also lower than what is predicted by the past crafts value.

Similarly, Fig. 3.B plots the RMI. Rank number 1 signifies the region with the highest crafts share. Urban regions today display a much lower crafts share than in the past, and, reversely, rural areas rank higher in terms of their crafts share. The effect is different in size across urban regions however, the strongest crafts decline can be found in the Hannover, Braunschweig, Wolfsburg regions, as well as in Hamburg, Berlin, Halle, and Leipzig. A smaller decline in the crafts can be found in the Ruhr valley, Bremen, and Dresden. Again, we see the crafts decline in the Baltic coastal areas.

Figure 4 shows a scatterplot, with crafts density in 1904 on the x-axis and the crafts density in 2013 on the y-axis. In addition, the figure shows a linear fit line, which represents the expected level of crafts persistence. Most importantly, the graph suggests that most urban regions (denoted by a triangle) display a decline in the crafts density over time, as they are below the regression line. This supports the hypothesis that the more dynamic economic development in cities leads to a crowding out of traditional economic structures. In addition, there are some non-urban regions which also display a decline in crafts density (diamond shape). Interestingly, most of these regions contain an independent medium-sized city (kreisfreie Stadt) or they are part of a metropolitan area and are therefore in close proximity to larger cities. Thus, these regions are most likely affected by the innovative dynamism of the respective urban centers eventually crowding out traditional crafts companies.

Table 6 displays the results of an OLS regression in which the persistence residual (specification 1 and 2), the change in the RMI (specification 3), or crafts density change (specification 4) is regressed against population density, the share of large companies and the east dummy variable as well as additional controls. The variable coefficient of population density is significantly different from zero across all specifications.

According to specification (1), The residual in the East lies 0.069 below that of West Germany. However, the effect vanishes once we include all controls and does not occur in specification 3 and 4. Thus, there is no evidence that the communist regime change in East Germany, and its active re-organization of regional economic structures, has lowered the degree of crafts persistence over time. As the cartographic depiction of the residual (Fig. 3) shows, the rural eastern regions of Brandenburg, Mecklenburg-Vorpommern, or southern Thuringia display considerable crafts persistence.

Moreover, as the share of large companies increases by one standard deviation, the residual increases by 0.064 (specification 1) and 0.038 (specification 2). The presence of large firms does not seem to negatively affect the persistence of the crafts sector over time. In contrast, the presence of large firms slightly increases crafts persistence. One cautious interpretation compatible with this finding relates to the development of large firms out of the existing regional ecology of firms, where large firms do not emerge as an exogenous shock to the system, but grow out of the existing ecosystem in an evolutionary fashion, thereby allowing for a continuity of sectorial composition.³ However, the share of large firms is not significant in specification 3 or 4. Thus, we conclude that there is little evidence for a crowding out effect of smaller crafts firms by large firms.

Finally, and most importantly, the population density negatively and significantly affects the persistence residual. As the density increase by one standard deviation the residual falls by 0.141 (in specification 2) and 0.098 (in specification 2). It increases the RMI by 0.017, about a third of a standard deviation, and decreases the crafts share change by 4, again, about a third of the standard deviation.

Given these effect sizes, the typical threefold increase in population density of urban centers effectively reduces the persistence of crafts to zero. A portion of this urbanization effect can be explained by the different educational compositions of urban and rural regions, as the coefficient of the tertiary education variable is negative and significant. In other words, the crafts, as a lower-tech SME sector, is unlikely to flourish where the share of highly educated individuals is large. However, the population density is responsible for the largest portion of the overall urbanization effect.

We conclude that the transformative changes that go along with the urbanization process during the twentieth century represent the primary reasons for change and the non-persistence of crafts structures. This interpretation is consistent with applied policy studies about the location decisions of crafts firms, which cluster in rural or semi-urban regions as land prices are lower. In contrast, the share of crafts firms is lower in urban areas (SMWA 2020).