The substitutability between brick-and-mortar stores and e-Commerce: the case of books

Götz, Georg; Klotz, Phil-Adrian; Schäfer, Jan Thomas; Herold, Daniel Article — Published Version The substitutability between brick-and-mortar stores and e-Commerce: the case of books Journal of Cultural Economics Suggested Citation: Götz, Georg; Klotz, Phil-Adrian; Schäfer, Jan Thomas; Herold, Daniel (2025) : The substitutability between brick-and-mortar stores and e-Commerce: the case of books, Journal of Cultural Economics, ISSN 1573-6997, Springer US, New York, Vol. 49, Iss. 4, pp. 811-854, https://doi.org/10.1007/s10824-025-09544-2 This Version is available at: https://hdl.handle.net/10419/333382 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. http://creativecommons.org/licenses/by/4.0/ Vol.:(0123456789) Journal of Cultural Economics (2025) 49:811–854 https://doi.org/10.1007/s10824-025-09544-2 ORIGINAL ARTICLE The substitutability betweenbrick‑and‑mortar stores ande‑Commerce: thecase ofbooks GeorgGötz1· DanielHerold1 · Phil‑AdrianKlotz2· JanThomasSchäfer1 Received: 21 April 2023 / Accepted: 2 May 2025 / Published online: 25 June 2025 © The Author(s) 2025, corrected publication 2025 Abstract We analyze competition between the online and the offline retail channel by using data on the German book market, which is characterized by fixed book prices. The analysis sheds light on the extent to which consumers perceive e-Commerce and traditional brick-and-mortar stores as substitutes. We find that, on average, when a bookstore closes, sales of print books decrease by around 744 units per month. This explains about 37% of the total loss in sales of print books in our sample. These findings indicate imperfect substitutability between the online and the offline retail channel. Substitutability between the channels remains imperfect when we incorporate information on e-book sales. The magnitude of the effect is genre-dependent. For instance, sales of fiction titles decrease more strongly than sales of school books. The data used in our study were provided by media control GmbH, GfK GmbH and Acxiom Deutschland GmbH. We received funding from the German Publishers and Booksellers Association (“Börsenverein des Deutschen Buchhandels e. V.”) to buy the data. Moreover, the German Publishers and Booksellers Association funded a research project at the Chair of Georg Götz from 2018 to 2020. Our study is a product of this project. The project itself was scientific in nature (i. e., no commercial research project). To conduct this project, the positions of the coauthors Jan Thomas Schäfer and Daniel Herold were funded by the German Publishers and Booksellers Association during the aforementioned period. Keywords Product differentiation· Book market· Retailing· e-Commerce We thank the German Booksellers’ Association, Joel Waldfogel, Samuel de Haas, David Finck, Maximilian Maurice Gailand Daniel Lüke for continuous support and fruitful discussions. * Daniel Herold [email protected] 1 Chair forIndustrial Organization, Regulation andAntitrust, Department ofEconomics, Justus Liebig University Giessen, Licher Strasse 62, 35394Giessen, Germany 2 Düsseldorf Institute forCompetition Economics (DICE), Heinrich Heine University Düsseldorf, Universitätsstraße 1, 40225Düsseldorf, Germany 812 Journal of Cultural Economics (2025) 49:811–854 JEL Classification L13· L81· D12· L42 1 Introduction With the rise of e-Commerce, conventional brick-and-mortar retail sectors have experienced a substantial increase in competitive pressure across various industries (see, e.g., Burt and Sparks 2003; Srinivasan etal. 2002). It remains an open question to what extent and in which industries consumers perceive e-Commerce and traditional, physical stores as substitutes (see, e.g., Wang and Goldfarb 2017; Sinai and Waldfogel 2004; Brynjolfsson etal. 2009). In this article, we investigate the substitutability between e-Commerce and physical retailers in the German book market. Physical bookstores and e-Commerce apparently have different “service” features. e-Commerce typically offers convenient product search options among potentially huge inventories (see, e.g., Waldfogel 2017). However, ordering online entails waiting costs and e-Commerce does not offer the possibility of physical inspection (Loginova, 2009; Guo and Lai, 2017). Books are considered experience goods, so that prepurchase uncertainty potentially affects demand (Reinstein and Snyder, 2005; Hilger etal., 2011). Against this background, consumers have the possibility to receive advice at physical bookstores or via online reviews or ratings (see also Reimers and Waldfogel 2021; Lizzeri 1999; Clement etal. 2007; Marvel and McCafferty 1990). Cowen (2008) and Gilbert (2015) address that while reducing the importance of physical stores, e-Commerce might expand overall demand for books by attracting new customers. However, given the differences in services offered by the two channels, it is unclear to what extent consumers view the offline and online channels as substitutes. If they are considered relatively far substitutes by sufficiently many consumers, it could be that a decrease in the number of physical bookstores (potentially triggered by advancing digitization) decreases sales when (potentially captive) consumers of the offline channel reduce their demand for books. Analyzing the substitution patterns between the two retail channels can thus improve the understanding of how digitization affects the retail landscape. The German book market is particularly well-suited for such an analysis because it is characterized by fixed book prices such that the price-dimension plays no role in the comparison between the two channels. To analyze to what extent consumers perceive the online and offline channels as substitutes, we investigate how closures of physical bookstores affect book sales. A novel panel data set is used which consists of monthly sales data of physical retailers and e-Commerce in Germany covering the period 2011–2017. The data set also contains information on the number of physical retailers. The relationship between the number of physical bookstores and book sales may be bidirectional, i.e., demand may increase in the number of outlets or the number of outlets may increase in demand. We employ an instrumental variable (IV) approach using a proxy for population as an instrument to obtain consistent estimates. Our analysis suggests that, on average, the closure of one physical bookstore results in a monthly decline of 744 book sales per federal state. Between 2011 and 2017, bookstore closures collectively contributed to a decrease in monthly sales of 813 Journal of Cultural Economics (2025) 49:811–854 around 1 million units, accounting for approximately 37% of the overall decline in monthly book sales. The drop in sales appears to be genre-specific. For example, sales of fiction titles decrease by around 208 units on average per month and federal state when a book store closes, whereas we find no effect for schoolbooks. Remarkably, for fiction titles and children books we find that a decrease exit of physical bookstores implies also a weak decrease in sales in e-Commerce. This finding indicates complementary between the two channels in the sense that e-Commerce benefits from the presence of physical bookstores. Overall, our findings provide evidence that e-Commerce is no perfect substitute to physical bookstores. We contribute to the literature on the impact of digitization on “traditional” physical retailers. There are numerous articles analyzing competition between the two retail channels, such as Brown and Goolsbee (2002), Jin and Kato (2007), Ofek etal. (2011), Gauri etal. (2021) or Couture etal. (2021). The impact of digitization on the retail landscape has been studied from different angles, including the topics crowd ratings (Reimers and Waldfogel, 2021), the role of niche-products (Brynjolfsson etal., 2003, 2006; Reimers and Waldfogel, 2017) and innovation, diffusion and copyright protection (Waldfogel and Reimers, 2015; Waldfogel, 2017; Reimers, 2016). Particularly closely related to this article are the papers of Goolsbee (2001), Prince (2007) and Goldmanis etal. (2010). Goolsbee (2001) and Prince (2007) analyze the market for computer equipment and find evidence for intensifying competition between the two channels throughout the late 1990s. Goldmanis etal. (2010) analyzes competition between the online and offline channel in three markets in the US: travel agencies, bookstores and new car dealers. With respect to bookstores, they document a shift of market shares from small bookstores to larger chains in markets that are exposed to a larger degree of online competition, leading to a decline in the number of bookstores and employment. In the context of the impact of digitization on the book market, our paper is closely related to Brynjolfsson and Smith (2000), Clay etal. (2002) and Chevalier and Goolsbee (2003). These studies, however, analyze markets where retailers compete in prices. With fixed book prices (i.e., in the absence of price competition between retailers), we find that the decrease in sales triggered by the closure of a bookstore are not fully compensated by increased sales of other bookstores or in e-Commerce. The article is also related to the literature and policy discussion on vertical restraints, particularly with respect to the evaluation of potential efficiency effects arising from resale price maintenance (RPM). This evaluation has become an important topic of competition policy at least since the Leegin-Case, which led to RPM being shifted from per se illegality to a rule of reason based approach in the US in 2007. In the EU, RPM is still practically considered per se illegal (see Akman and Sokol 2017) for a more detailed overview). However, in various European countries such as France, Austria and Germany (and other countries such as Japan),1 fixed book price systems are in place, which occur in the form of RPM. As the goal of 1 See, for instance, Global Fixed Book Price Report of the International Publishers Association, May 23, 2014, https:// bit. ly/ 2Wg1t pz. According to Poort and van Eijk (2017), in 15 OECD-countries (ten EUmembers), there is a fixed book price system in place. 814 Journal of Cultural Economics (2025) 49:811–854 fixed book prices is usually to protect books as a cultural or merit good, they constitute an exception when it comes to the application of competition law. Whether such an exemption from some fundamental pillars of competition law is justified, crucially depends on fixed book prices being an appropriate tool to achieve the policy goal to promote the demand and supply of books. Against the background of our findings, one way through which fixed book prices could affect book sales would be securing margins and thereby potentially promoting market entry or preventing exit of physical retailers (Bouckaert, 2000; Elzinga and Mills, 2008; Marvel and McCafferty, 1985). In particular, Williams (2024) finds that, on average, book sales are higher in countries with RPM, while prices are similar. He explains this findings by a positive impact of fixed book prices on the network of bookstores. The article is structured as follows. Section2 contains a brief conceptualization of our empirical analysis based on economic theory. Our data set is described in Sect.3. Section4 presents our empirical analysis. Section5 provides an overview of the robustness checks that are presented in the Appendix. Section6 concludes. 2 Theory The following section provides a brief theoretical background for the empirical analyses presented in the remainder of this article. The starting point is the notion that the online and offline retail channels have different “service” features, as was already explained in the introduction. Consumers vary regarding their preferences toward these features. This is an example of (horizontal) product differentiation, which, in the literature, is usually formalized using Hotelling and Salop models where consumers’ preferences regarding the two sales channels are reflected in different transport costs (Balasubramanian, 1998; Bouckaert, 2000; Chu etal., 2012; Guo and Lai, 2017; Legros and Stahl, 2019). In our empirical setting, we study the effect of a physical bookstore’s market exit. In a Hotelling or Salop model, ceteris paribus such an exit has no effect on consumers buying online or patronizing a bookstore that is not closing. Net utility of those consumers who patronized a bookstore that is closing decreases due to an increase in inconvenience costs from either patronizing another bookstore or purchasing online. This can lead to a decrease in overall demand if that increase is so high that, for given prices, consumers net utility from purchasing books at another bookstore or online becomes negative.2 It is possible in such a setting that exit of physical bookstores ceteris paribus leads to a decrease in market coverage and, thus, demand for books. We interpret this pattern as imperfect substitution between the online and the offline channel. That is, some consumers who patronized a physical bookstore and who have high 2 Such a formalization requires a model that allows for partial market coverage, i.e., a model in which not all consumers always purchase the product by assumption. To construct such a model, one would need to depart from the assumption of symmetric inconvenience costs from buying online, as it is usually used in the literature (see above). Instead, one could assume that these costs differ between consumers. An earlier version of this paper contained such a model. It is available from the authors upon request. 815 Journal of Cultural Economics (2025) 49:811–854 inconvenience costs from buying online may stop buying altogether because neither e-Commerce nor the remaining bookstores are perceived as “close enough” substitutes. This notion may serve as a theoretical interpretation of the following empirical analyses. 3 Data anddescriptive statistics For our empirical analysis, we use monthly data on the number of book retailers as well as sales volumes and revenue data on a federal state level. The data comprise sales of books in brick-and-mortar stores as well as e-commerce sales, for example, books shipped by Amazon. We distinguish between nine different product groups to control for genre-specific effects. The database covers the period from January 2011 to December 2017. We utilize a combination of data from multiple sources. In what follows, we present some descriptive statistics and we describe the data sources in more detail. 3.1 Brick‑and‑mortar bookstores The German Publishers and Booksellers Association3 maintains several databases on the number of brick-and-mortar stores, such as its membership database and the Address Book for the German-Language Book Trade4, which we have combined to obtain an accurate picture of the market and its development over time. We checked the consistency of our dataset using data on the order volume of bookstores provided by three of the largest German wholesalers.5 Appendix A.1 contains an indepth description of the data generation process. Figure1 shows the development of the number of physical bookstores in Germany over time, based on data from the German Publishers and Booksellers Association. The data include information on outlets of chain stores as well as independent bookstores. Between 2011 and 2017, the number of physical bookstores decreased from over 6,300 to less than 4,900. Figure2 provides summary statistics of bookstores on the federal state level. On average, the number of brick-and-mortar bookstores decreased by 20 percent over the observation period. In absolute numbers, around 1,435 bookstores exited the market. Figure3 shows the number of brick-and-mortar stores per 100,000 residents in beginning of 2011 and end of 2017.6 3 Börsenverein des Deutschen Buchhandels, see https:// www. boers enver ein. de. 4 Adressbuch für den deutschsprachigen Buchhandel, see https:// adbonline. de; last accessed December 19, 2023. 5 Libri GmbH (https:// www. libri. de), G. Umbreit GmbH & Co. KG (https:// www. umbre it. de), and Koch, Neff & Volckmar GmbH (KNV, http:// www. knv. de). 6 Note that we use labor force as a proxy for population. This is because population data are only published on a yearly frequency. See Sect.4 for further discussion. 816 Journal of Cultural Economics (2025) 49:811–854 Fig. 1 Development of physical bookstores over time. Source: German Publishers and Booksellers Association Fig. 2 Development of physical bookstores on the federal state level. Source: German Publishers and Booksellers Association 817 Journal of Cultural Economics (2025) 49:811–854 3.2 Sales data In order to study the development of sales over the observation period, we use data from two sources. First, we have access to scanner data provided by media control GmbH.7 Second, we use consumer panel data provided by GfK GmbH.8 Both databases differ with respect to coverage, frequency and accuracy. The scanner data are based on quasi real time data and comprises information on sales of print books in independent bookstores, chain stores as well as online retailers (see Appendix A.2 for more characteristics of the data set). The data provider claims to cover more than 80–90% of the German print book market. However, the scanner data do not contain information on the sale of e-Books. To the best of our knowledge, the most reliable information on e-Book sales in Germany available to researchers are consumer panel data, which, for the observation period 2014–2017, is provided by GfK. The sales data projections of GfK are based on roughly 20,000 consumers being surveyed on a regular basis and contain information on the purchases of print books, e-Books and audio books. In contrast to the panel data, the survey data are only available on a quarterly level for the period from Q1.2014 to Q4.2017 (see Appendix A.3 for more characteristics of the data set). For the years 2011 to 2013, survey data Fig. 3 Number of physical bookstores per 100,000 residents in beginning of 2011 and end of 2017. Source: German Publishers and Booksellers Association 7 see https:// www. mediacontr ol. de/ buch1. html; last accessed December 20, 2023. 8 see https:// www. gfk. com/ de; last accessed December 20, 2023. 818 Journal of Cultural Economics (2025) 49:811–854 are only available as yearly aggregates and for whole Germany, rather than on the federal state level and are included for illustrative purposes only.9 Figure 4 shows sales figures for physical books from the scanner data set. It becomes obvious that the sales of print books in Germany exhibit a decreasing trend. In 2011, 297.2 million print books were sold in brick-and-mortar and online stores with a corresponding revenue of 3.66 billion Euros. This implies an average book price of 12.31 Euro. In 2017, 264 million books were sold with a revenue of 3.54 billion Euros (i.e., 13.41 Euro per book). The decline in sales was thus at least partly offset by a price increase so that revenues of brick-and-mortar and online stores remained fairly constant. As can be seen in Fig.5, the number of print books sold per resident decreased between 2011 and 2017. At the same time, e-Book sales increased from 4.30 million in 2011 to 29.15 million in 2017. As can be seen from Fig.6, the growth rate of e-book sales stalled after a sharp increase and corresponding revenues even declined between 2016 and 2017. It is important to note that e-Book sales data cannot be directly compared to the data on print book sales that have been presented in Fig.4. From the comparison of GfK’s print book sales data (see Appendix A.3) with the more accurate scanner data, we know that the consumer panel projections might be overestimating "true" sales by up to 25 percent. However, even without correcting for this potential bias, it becomes clear that e-Book sales did not completely compensate the decline in sales of print books in all years. For example, print book sales dropped by 6.1 Million copies between 2013 and 2014, while e-Book sales only increased by 3.31 Million. Fig. 4 Annual revenues (in million Euros) and sales (in million books sold) of print books in brick-andmortar stores and e-Commerce. Cumulative losses since 2011 in parentheses. Thuringia and Saarland are excluded due to inconsistencies in the data. Source: media control GmbH 9 For e-Book sales see https:// de. stati sta. com/ stati stik/ daten/ studie/ 232191/ umfra ge/ absatzvon-ebooksindeuts chland; last accessed December 20, 2023. For total sales (including e-Books and audio books) see https:// de. stati sta. com/ stati stik/ daten/ studie/ 416380/ umfra ge/ absatzvonbuech ernindeuts chland; last accessed December 20, 2023. 825 Journal of Cultural Economics (2025) 49:811–854 4.2 Regression analyses This section presents the results of the analysis for the period 2011–2017 and focuses on sales of physical books. Before we present the regression results, we first explain our 2SLS estimation approach. The structural equation of our basic model takes the following form: where the dependent variable are the sales of print books per capita in federal state i and month t. Our treatment variable  # stores i,t refers to the fitted values from the firststage for the absolute number of bookstores in federal state i in month t. The year-month fixed effects are given by 𝜉t . As mentioned above, book sales are decreasing over time. This trend, which could, for instance, be associated with more consumers substituting reading with movie streaming or online gaming, is captured by those dummies. Moreover, the year-month fixed effects also control for the seasonality of book sales, e.g., the large Christmas effect. The variable gtrendsi,t captures the Google Trends Index for the topic “book”, ereaderi,t gives the Google Trends Index for the search item “E-book reader” and 𝜉i depicts time-invariant fixed effects for each federal state i.19 As explained in Sect.4.1, we use popi,t and pop 2 i,t as instruments for the potentially endogenous variable #storesi,t . #storesi,t is correlated with popi,t and pop 2 i,t , which is a prerequisite for the relevance condition. Note that, for the relevance condition to be satisfied, pop 2 i,t needs to be implemented, as otherwise the correlation is not sufficiently strong. As explained in Sect.4.1, the orthogonality assumption should be satisfied as well because it is reasonable to assume that population does not affect how many books a person reads. Thus, the projection in the first-stage regression of our base model can be formalized as follows: The corresponding first-stage regression results are presented in Table11 of Appendix B.2. 4.2.1 Baseline results Table2 presents the regression results for Eq. (1).20 We run the regression for total sales, i. e. the sum of sales in physical bookstores and e-Commerce (column (1)), (1) sales _pc i,t =𝛽 1 #stores i,t +𝛽 2 gtrends i,t +𝛽 3 ereader i,t +𝜉 i +𝜉 t +u i,t, (2) # stores i,t = 𝛿1 pop i,t + 𝛿2 pop 2 i,t + 𝛿3 gtrends i,t + 𝛿4 ereader i,t + 𝜉i + 𝜉t + 𝜀i,t. 19 Note that we only investigate 14 out of 16 federal states in Germany. We excluded two federal states (Saarland and Thuringia) from our analyses due to potential errors in the data. 20 We present the results of a naive OLS estimation for Eq. (1) in Table10 of Appendix B.1. In this regression, the number of bookstores has a positive effect on the print book sales, which is significant on the 5%-level. However, as discussed in the previous sections (esp. Sect.4.1), this naive OLS regression is prone to endogeneity. Neglecting this endogeneity issue could lead to a biased estimation, which is the reason why we apply an IV estimation approach. 826 Journal of Cultural Economics (2025) 49:811–854 and for the two different sales channels physical bookstores and e-Commerce separately (columns (2) and (3)). The coefficient of #stores measures how a change in the total number of bookshops affects book sales per capita in an average month for an average federal state. Thus, based on column (1) in Table2, we can conclude that if one physical bookstore closes, ceteris paribus, sales per capita decrease by 0.000248 on average (average month, average federal state). The effect can be disentangled into the offline and online channel. The effect on sales in physical bookstores is 0.000183 and in e-Commerce 0.0000559. Before turning to the interpretation of the findings, some technical aspects have to be discussed. The potentially endogenous variable in Equation (1), #stores , is identified by the instrument as indicated by the Kleibergen–Paap rk Wald F-statistic which clearly exceeds the IV critical value from Stock and Yogo (2005). Furthermore, the Null of the Anderson–Rubin Wald F-statistic can be rejected for the regression with aggregated (column (1) in Table2) and offline sales (column (2)), indicating that the potentially endogenous regressor is relevant in the structural equation. Additionally, a test for exogeneity of the variable #stores using the difference of two SarganHansen statistics (also GMM distance or C-statistic) can be rejected.21 A positive coefficient for #stores (column (1) in Table2) means that when a physical bookstore exits the market, aggregate sales decrease. The same applies to sales in the offline and the online channel in isolation (columns (2) and (3) in Table2). This implies that, on average, market exit of physical bookstores is not fully compensated for by sales in other physical bookstores. (Full compensation would occur if the effects were not statistically different from zero. In that case, depending on which effect is considered, the closure of a bookstore would not affect total sales, sales in the offline or sales in the online channel.) Our results thus indicate that, on average, market exit of physical bookstores leads to a significant decrease in sales. The effect is different for the offline and the online channel. One can see that offline sales decrease more strongly upon market exit of physical bookstores than online sales (columns (2) and (3) in Table2). The positive coefficient of e-Commerce might seem confusing at first glance. It indicates complementarity between the channels, which is consistent with the wellknown problem of free-riding on service provision as in, for example, Telser (1960): The features or “services” of one sales channel affect sales in the other channel. Even though the exact mechanism cannot be clearly identified based on our data, there are potential explanations for this pattern. One potential explanation could be “showrooming” (Zhang etal., 2018). Although RPM eliminates price differences between the offline and online channel, non-price factors like product fit and wait times, as highlighted by Gensler etal. (2017), can still drive such an effect. Another potential explanation would be word-of-mouth (Beck, 2007), where initial exposure to a title in the offline channel, which is characterized by a much higher sales volume in Germany (see Sect.A.3 in the Appendix), can lead to increased online 21 This endogeneity test statistic is numerically equivalent to a Durbin-Wu Hausman test under conditional homoskedasticity (see Hayashi (2000)). 827 Journal of Cultural Economics (2025) 49:811–854 purchases. In Sect. 4.2.2, the effect will be disentangled for different genres and types of bookstores. To get a better understanding of the magnitude of our findings, it is illuminating to express the decrease in sales in absolute terms. Population, which we approximate using labor force (see above), is, on average, 3 million per federal state (42 million total). Thus, when one bookstore closes, average book sales decrease by 0.000248 ⋅3 million =744 units per month in an average federal state. Based on these figures, it is possible to answer the following question: How did the closures of bookstores affect print book sales in the period 2011–2017? To answer this question, first note that monthly sales went down by 2,765,000 from 24,767,500 in 2011 to 22,002,500 in 2017 (see Fig.4). In the 14 federal states examined in this section, 1,382 bookstores exited the market.22 With an average decrease in sales upon the closure of a bookstore of 744 units per month on average, store closures have accounted for a drop of 1,028,208 in monthly sales. This suggests that bookstore closures accounted for about 37% of the sales decrease.23 When interpreting these findings, one has to keep in mind that they occur in the German market, which is characterized by a relatively high density of bookstores. According to our data, in Germany there is approximately one physical bookstore per 16,400 inhabitants in 2018. For instance, in the USA the number of physical bookstores is much lower. The number of stores that sold books peaked at 12,000 in 1992 (see Wu etal. 2018), so that there was approximately one physical bookstore per 21,400 inhabitants. It remains an open question whether the effect of a closure on sales becomes more pronounced the lower the number of physical stores because with each closure it becomes increasingly more difficult for consumers to find an Table 2 Main estimations with sales volumes on print books per capita as dependent variable Standard errors in parentheses Significance levels:*P < 0.05, **P < 0.01, ***P < 0.001 (1) (2) (3) Aggregated Offline e-Commerce # Stores 0.000248*** 0.000183*** 0.0000559* (0.0000611) (0.0000486) (0.0000260) Google Trends 0.00376*** 0.00317*** 0.000739* (0.000811) (0.000637) (0.000354) Google Trends e-Reader 0.000470 0.000169 0.000225 (0.00113) (0.000815) (0.000622) Federal State FE Yes Yes Yes Year-Month FE Yes Yes Yes Anderson–Rubin Wald F-statistic 8.009 7.281 2.095 Kleibergen–Paap rk Wald F-statistic 2238.1 2238.1 2238.1 # of observations 1176 1176 1176 22 Including Thurungia and Saarland, 1,435 were closed between 2011 and 2017, see Sect.3. 23 We thank one of the anonymous reviewers for suggesting this interpretation. 828 Journal of Cultural Economics (2025) 49:811–854 appropriate substitute offline (e.g., increasing physical traveling distances). It is left open for future research to investigate whether the effect of the closure of physical bookstores is different in markets with fewer or more physical bookstores than Germany. Note that the coefficient of Google Trends for the term “book” is statistically significant. This means that, ceteris paribus, with increasing search volume, sales of books tend to increase. In contrast, the coefficient of Google Trends e-Reader is insignificant in Table2 implying that the search behavior of the consumers for e-Readers does not significantly affect the physical book sales per capita. In other words, it does not appear to be the case that consumers substitute e-Books for print books in a systematic fashion, as would be indicated by a negative coefficient for that index. This can also be seen as an indication that consumers do not substitute print books by e-Books when bookstores close, or at least not one-for-one. E-Book sales will be discussed in Sect.4.3 in more detail. Lastly, as a first robustness check24, we compare the results of our IV estimation presented above with four alternative specifications: an IV estimation using yearly data (a), a naive OLS estimation (b) and two OLS specifications with lags (c) and (d). The IV estimation using yearly data (a) has the benefit that seasonal patterns in sales or the number of bookstores (e.g., the Christmas effect or memberships of the German Booksellers Association running out) play no role in the estimations. However, the number of observations drops from 1,176 to 98, which strongly reduces the power of the analysis. The naive OLS estimation (b) ignores potential endogeneity that arises when book sales are regressed on the number of bookstores (see the beginning of this chapter). The lagged OLS specifications (c) and (d) can be seen as more casual attempts to control for this endogeneity than an IV approach, since the book sales in period t should not directly affect the number of bookstores in t−1 . Specifications (c) and (d) were run in levels and in log levels, respectively. The results of the respective estimators for #stores are presented in Table3. The second to fifth rows in Table3 depict the estimation results for the aforementioned, alternative specifications, whereas the first row depicts the results derived from the main IV. In comparison to the latter, the IV with yearly variation (a) shows that the effects of a closure on aggregated sales or sales in e-Commerce are not significantly different from zero at the 5 %-level. This is not surprising given the decrease in the number observations. Interestingly, the coefficient of #stores is still significantly different from zero at the 10 %-level. Moreover, the effects of a closure on offline sales remains significant at the 0.1 %-level. With a naive OLS (b), the coefficient of #stores is not statistically significant at the 5 %-level. The coefficient is significant at the 10 %-level for all channels (Aggregated, Offline and Online). However, given that the naive OLS does not account for potential endogeneity, the effect may be biased, anyway. The lagged OLS estimates (c) and (d) are statistically significant at the 5 %-level for the offline channel. Specification (d) also shows a significant effect at the 5 %-level on aggregate. The magnitude of some of the effects in the OLS specifications is generally higher than in the IV estimation. The coefficient for aggregated sales (column (1)) 24 Further robustness checks are presented in Sect.5. 829 Journal of Cultural Economics (2025) 49:811–854 implies an average decrease in sales per bookstore that closes of around 1,980 units (avg. annual sales per federal state (19,8 million books, see Table1) times the coefficient (0.0001)). The effects in e-Commerce are far from significant and should not be interpreted. The estimation equations for (c) and (d) are presented in Appendix C.4. Overall, it remains to conclude that, at that point, the results of the main IV estimation appear to be robust. As expected, the significance levels strongly decrease when using yearly data, however, they are in line with what we observe when using monthly data. Also the effects of naive or lagged OLS estimations point in a similar direction as the IV results. It should be noted, though, that the OLS estimations should be interpreted carefully as they are expected to be affected by endogeneity. 4.2.2 Genre‑specific effects In this section, we will check whether there are genre-specific differences in the effects of exit of physical bookstores on sales. In doing so, we run the IV regressions explained above for the nine different book genres (see in Sect.3) separately. Table4 represents the results for our treatment variable #stores in the respective second stages of the 2SLS regressions with aggregated sales, offline sales and sales in e-Commerce as dependent variables. The results presented in the first column of Table4 are largely in line with the results presented above. For instance, if one bookstore closes, 0.0000692 print books of the fiction genre are sold less in both retail channels on average (month and federal state). This means that 0.0000692 ⋅3 million =207.6 fiction books are Table 3 Comparison of IV and OLS estimations Standard errors in parentheses Significance levels:*P < 0.05, **P < 0.01, ***P < 0.001 (1) (2) (3) Aggregated Offline e-Commerce IV 0.000248*** 0.000183*** 0.0000559* (0.0000611) (0.0000486) (0.0000260) Yearly 0.00298 0.00589*** 0.000565 (0.00167) (0.0000521) (0.000971) OLS 0.000515 0.000470 0.000106 (0.000249) (0.000224) (0.0000493) Lagged OLS 0.000557 0.000562* −0.00000515 (0.000272) (0.000222) (0.0000983) Lagged OLS (log level) 0.0001* 0.00128** −0.00084 (0.0004682) (0.0005409) (0.0017939) Federal State FE Yes Yes Yes Time FE Yes Yes Yes 830 Journal of Cultural Economics (2025) 49:811–854 sold less per month, on average. Similar interpretations can be applied to the genres nonfiction, humanities, children books, natural sciences, guidebooks and travel. In particular, note that the effect for children books is even slightly stronger than that for fiction ( 0.0000701 ⋅3 million =210.3 ). However, the change in the number of physical bookstores has no significant impact on offline sales of school books and books belonging to the category social sciences. In other words, if physical bookstores close, total sales as well as sales in the offline and online category remain unaffected. This finding is in line with expectations as the demand for school books can be expected to be highly inelastic and independent of the number of bookstores. In that genre, service aspects such as expert opinion, ad hoc sales, etc., should be almost irrelevant. Table 4 IV regressions by genres Standard errors in parentheses Significance levels:*P < 0.05, **P < 0.01, ***P < 0.001 Aggregated Sales Offline Sales e-Commerce Fiction # Stores 0.0000692*** 0.0000557*** 0.0000151* (0.0000191) (0.0000166) (0.00000678) Nonfiction # Stores 0.0000150** 0.0000112** 0.00000379 (0.00000459) (0.00000359) (0.00000215) Humanities # Stores 0.0000106*** 0.00000695*** 0.00000209 (0.00000200) (0.00000138) (0.00000125) Children books # Stores 0.0000701*** 0.0000573*** 0.0000113* (0.0000109) (0.00000914) (0.00000455) Natural sciences # Stores 0.00000655*** 0.00000358*** 0.00000319* (0.00000143) (0.000000474) (0.00000131) Guidebooks # Stores 0.0000400*** 0.0000323*** 0.00000975 (0.00000766) (0.00000413) (0.00000588) Travel # Stores 0.0000128*** 0.0000106*** 0.00000233* (0.00000251) (0.00000211) (0.00000118) School books # Stores 0.00000456 0.00000533 0.00000281 (0.0000369) (0.0000332) (0.00000520) Social sciences # Stores 0.00000276 0.000000745 0.00000275 (0.00000184) (0.000000853) (0.00000148) 831 Journal of Cultural Economics (2025) 49:811–854 The coefficients in the third column in Table4 (Sales e-Commerce) require some discussion. As already explained, these coefficients capture a complementarity between the online and the offline channel in a sense that online sales are positively affected by the number of bookstores. This is the case for fiction, children books, natural sciences and travel. However, note that this effect is only significant on the 5%-level. Moreover, for fiction (45.3 units) and children books (33.9 units) the effect is far stronger than for natural sciences (9.57) and travel (6.99). 4.3 Estimation withE‑books It is reasonable to assume that at least some consumers substitute e-Books for print books when physical bookstores close, especially since there is evidence that e-Books expanded book sales (see Gilbert, 2015,p. 167–170, for an overview). It is thus appropriate to check whether the effects of exit of physical stores on book sales presented above are overestimated due to missing information on e-Book sales. Thus, we will now deploy the consumer panel data set that includes information on e-Book sales (see Appendix A.3 for more details). The data are used to repeat the analyses presented in Sect.4.2. Estimation results are reported in Table5 as follows. We run the regression for total book sales (column (1)) and for the two book formats print and e-Books separately (columns (2) and (3)). As shown in Table5, the effect of the number of physical bookstores on book sales remains statistically significant in all three regression approaches. The reported coefficient in column (1) implies a decrease in total book sales per capita, including e-Books, of 0.00257 in a given quarter per federal state (significant on the 1%-level), when a physical bookstore closes. (For a more detailed discussion of the order of magnitude of the effect, see below.) The effect in the number of bookstores on book sales is also significantly positive when we run the regression for print and e-Books separately (columns (2) and (3)). Thereby, the number of physical bookstores has a larger effect on print book sales per capita (0.00239) than on e-Book sales per capita (0.000189). Again, Table 5 Estimation with total book sales (print books and e-Books) per capita as dependent variable using consumer survey data from GfK Standard errors in parentheses Significance levels:*p < 0.05, **p < 0.01, ***p < 0.001 (1) (2) (3) All books Print e-Books # Stores 0.00257*** 0.00239*** 0.000189*** (0.0000841) (0.0000849) (0.0000228) Federal State FE Yes Yes Yes Year-Quarter FE Yes Yes Yes Anderson–Rubin Wald F-statistic 479.6 443.6 29.27 Kleibergen–Paap rk Wald F-statistic 8475.0 8475.0 8475.0 # of observations 320 320 320 832 Journal of Cultural Economics (2025) 49:811–854 the significant result for e-Books indicates complementarity between the online and the offline channel. These findings also show that the main results reported in Sect.4.2 are qualitatively robust when information on e-Book sales are included into the analyses, as e-Book sales apparently do not compensate for the decrease in sales triggered by the exit of physical bookstores, as this would require a negative coefficient of #stores . Thus, a significantly positive (or insignificant) coefficient does not challenge the previous findings. Note that the effects appear to be stronger when we use consumer panel data. The coefficient for “All books” reported in column 1 implies a decrease in average quarterly sales of 0.00257 ⋅ 3 million = 7, 710 , i.e., 2,570 units per month. This effect is around 3.5-times stronger than the one reported in the Sect.4.2.1. A possible explanation for this difference is that information in the consumer panel data set are extrapolated by GfK based on survey data, whereas the scanner data are based on actual sales (see Sect.3). Thus, the two data sets differ in terms of absolute sales figures. As explained in Sect.3, we consider the scanner data to be a more accurate description of actual book sales than survey data. 5 Robustness checks Robustness checks were performed for the analyses presented in the previous sections. All estimation results are presented in Appendix C. Our results remain robust with respect to the following modifications. • To rule out that book prices affect some of the results, the change in the monthly average book prices is integrated as a “bad” control variable into the second stage of the IV regression. Average book prices are computed based on scanner data for the period 2011–2017. Including prices as a control can rule out that decreasing book sales are driven by increasing book prices and not by the decreasing number of physical bookstores. The control variable is “bad” because the relationship between sales and prices is bidirectional (see Appendix C.1). • The number of students is included as a covariate. This is done to control for the education levels in the 14 German federal states of our data set. The fixed effects included into the regressions capture nationwide time trends and time-invariant effects that affect individual federal states. However, sociodemographic effects such as education can be time-variant and specific to a federal state. Those effects potentially confound our analyses (see Appendix C.2). • The analyses presented in the Sect.4.2 can be performed using a combination of survey and scanner data to include e-Book sales. Therefore, we use the ratio of e-Book sales from the survey data to calculate absolute e-Book sales based on the scanner data. This e-Book sales data then can be used to repeat our estimations for the period 2014–2017 (quarterly level) with data on print and digital book sales (see Appendix C.3). 833 Journal of Cultural Economics (2025) 49:811–854 • We also regress the book sales per capita on the lagged number of physical bookstores in an OLS estimation approach. Lagging our treatment variable #stores does at least partially solve the reverse causality issue between the number of bookstores and the book sales per capita (see Appendix C.4). • To account for potential differences in the development of book sales across German federal states, we extend our IV estimation by incorporating state-specific time trends. The results confirm our main findings while yielding slightly larger point estimates (see Appendix C.5). 6 Conclusion We find that, overall, e-Commerce does not pose a perfect substitute to physical retailers from the viewpoint of the consumers. Our results predict that when a physical bookstore closes, on average, monthly book sales decrease by 744 units. Taking into that between 2011–2017, 1,382 bookstores were closed across the 14 federal states comprising our data set, our results indicate that around 37% of the drop in monthly print book sales can be traced by to the closures of bookstores. Apparently, a large enough number of consumers prefers to buy books at physical bookstores for closures of those stores to have a statistically significant, negative impact on the total sales of books. Consumers’ preferences toward offline “services” might potentially be affected by expert opinion, a more careful selection and presentation of titles, ad hoc purchases or simply the atmosphere of physical bookstores. The magnitude of the effect differs between genres. The effect is particularly strong for fiction and children books titles, where we not only find a significant drop in offline sales but also in online sales following the market exit of physical bookstores. This finding indicates complementarity between the channels for some genres. On the other hand, we find no such relationship between the two channels when it comes to school books. This observation appears intuitive as the demand for schoolbooks should be determined by reasons other than features or services offered by online and offline bookstores. It is important to note that the magnitude of the effects of market exit of physical bookstores was determined for the German market, which has a relatively high number of physical bookstores. It remains an open question whether the effect of a closure on sales becomes more pronounced the lower the number of physical stores because with each closure it becomes increasingly more difficult for consumers to find an appropriate substitute offline (e.g., increasing physical traveling distances). It is left open for future research to investigate whether the effect of the closure of physical bookstores is different in markets with fewer or more physical bookstores than Germany. Our finding that consumers perceive online and offline retailers as imperfect substitutes provides an additional facet to the policy evaluation of fixed book prices. The goal of these vertical restraints is usually to protect books as cultural or merit goods. It is described in the literature that fixed book prices can promote market entry (Bouckaert (2000), Guo and Lai (2017), Elzinga and Mills (2008,p. 1848)) and prevent exit by, for example, securing margins, in particular in the presence of 834 Journal of Cultural Economics (2025) 49:811–854 online competition (Marvel and McCafferty (1985, 376), Bouckaert (2000), Guo and Lai (2017), Elzinga and Mills (2008,p. 1848), Legros and Stahl (2019)). There is evidence from the UK that after the abrogation of the Net Book Agreement, UK’s and Ireland’s fixed book price system abandoned in the 1990s, the book market consolidated (Davies etal., 2004; Fishwick etal., 1997; Dearnley and Feather, 2002). In combination with our finding that a larger number of physical bookstores promotes book sales, fixed book price systems may thus support the policy goal of securing a broad supply of books. Even though systematic analysis is warranted, this efficiency effect of fixed book prices would have to be taken into account when evaluating the welfare effects of suppressing price competition among retailers. Appendix: A: Data A.1: Brick‑and‑mortar bookstores The numbers presented in Fig.1 are derived from the membership database of the German Publishers and Booksellers Association. Roughly 85–90% of all book retailers in Germany are members of the German Publishers and Booksellers Association. The database contains information on the name, the address, the founding year of a bookstore, the date of the beginning as well as the end of a membership in the Association. Since 1963 a unique identifier (“Verkehrsnummer”) is assigned to each business entity that is member of the German Publishers and Booksellers Association. This identifier is used for communication between business partners, i.e., for communication between bookstores, publishers, and wholesalers. Using this identifier we are able to match bookstores to other databases, like the so-called Address Book for the German-Language Book Trade.25 This address book was established in 1839, and is now managed by a subsidiary of the German Publishers and Booksellers Association.26 The database contains over 30,000 addresses of publishers, bookstores, music stores, and publishing representatives in the German-speaking region. The address book is updated regularly and changes (entry, exit, change of location or legal name) are published online. Members of the German Publishers and Booksellers Association can be listed in the address book for free, nonmembers have to pay a small annual fee. We apply several steps in order to transform the merged address book/membership database into a monthly panel that contains the number of active brick-andmortar stores for each Federal State. First, we drop all publishers, head offices and warehouses as well as online bookstores from the database. Second, we obtain information on the Federal State in which each brick-and-mortar store is located, using the Google Maps API. Third, we group stores by location. In doing so, we avoid situations where a bookstore would receive a new identifier due to, for example, a 25 Adressbuch für den deutschsprachigen Buchhandel, see https:// adbonline. de; last accessed December 19, 2023. 26 MVB Marketingund Verlagsservice des Buchhandels GmbH, see https:// mvbonline. de; last accessed December 20, 2023. 841 Journal of Cultural Economics (2025) 49:811–854 Table 8 Summary statistics by Federal State Federal State N Mean SD Min Median Max Baden-Wuerttemberg Quantity all channels 84 3,054,151 817,311 2,197,958 2,826,255 6,033,089 Number B&M stores 84 797.6 63.7 698 793 905 Population (labor force) 84 5,832,617 175,385 5,538,385 5,825,910 6,179,156 Google Trends Book 84 67.39 6.267 57 66 88 Google Trends Amazon Kindle 84 29.06 13.22 6 26 82 Bavaria Quantity all channels 84 3,523,258 967,749 2,578,736 3,254,845 6,786,929 Number B&M stores 84 855.6 65.84 769 833 967 Population (labor force) 84 6,991,562 187,336 6,670,375 7,014,883 7,368,733 Google Trends Book 84 66.75 6.856 58 65 87 Google Trends Amazon Kindle 84 30.68 14.63 6 28.5 87 Berlin Quantity all channels 84 1,095,268 300,601 779,227 1,012,833 2,094,380 Number B&M stores 84 248.4 11.45 230 249 274 Population (labor force) 84 1,807,867 53,918 1,710,468 1,823,527 1,900,722 Google Trends Book 84 69.99 9.021 57 68 100 Google Trends Amazon Kindle 84 27.36 11.35 5 25.5 70 Brandenburg Quantity all channels 84 629,316 213,298 414,295 533,336 1,213,088 Number B&M stores 84 91.93 6.259 80 94 101 Population (labor force) 84 1,331,518 8977 1,311,513 1,332,724 1,345,946 Google Trends Book 84 62.45 7.922 47 61 83 Google Trends Amazon Kindle 84 24.83 12.71 0 23.5 75 Bremen Quantity all channels 84 279,569 97,812 180,552 246,384 671,877 Number B&M stores 84 53.01 3.012 49 53 58 Population (labor force) 84 337,925 9175 321,313 340,915 350,670 Google Trends Book 84 64.96 8.074 53 63 91 Google Trends Amazon Kindle 84 27.33 13.31 3 25 75 842 Journal of Cultural Economics (2025) 49:811–854 Table 8 (continued) Federal State N Mean SD Min Median Max Hamburg Quantity all channels 84 764,794 217,571 569,237 705,560 1,587,100 Number B&M stores 84 122.4 11.04 105 121.5 139 Population (labor force) 84 972,280 29,745 919,795 974,548 1,025,176 Google Trends Book 84 65.24 7.824 51 63 96 Google Trends Amazon Kindle 84 27.62 13.25 5 26 85 Hesse Quantity all channels 84 1,800,033 493,730 1,276,866 1,674,170 3,667,458 Number B&M stores 84 505.5 45.83 439 504 576 Population (labor force) 84 3,219,236 78,415 3,073,820 3,216,483 3,368,375 Google Trends Book 84 65.48 6.647 55 64 85 Google Trends Amazon Kindle 84 29.05 13.39 6 26 80 Lower Saxony Quantity all channels 84 2,639,531 865,852 1,783,210 2,266,358 4,885,420 Number B&M stores 84 504 43.91 439 502.5 579 Population (labor force) 84 4,128,667 89,228 3,974,260 4,136,114 4,289,250 Google Trends Book 84 61.05 6.784 49 60 80 Google Trends Amazon Kindle 84 25.29 11.45 3 24 70 Mecklenburg-Western Quantity all channels 84 409,381 111,777 281,548 370,741 697,173 Pomerania Number B&M stores 84 74.02 6.499 66 71 86 Population (labor force) 84 838,488 11,969 822,790 832,014 862,432 Google Trends Book 84 59.42 7.409 45 59 84 Google Trends Amazon Kindle 84 24.64 12.44 2 22 63 843 Journal of Cultural Economics (2025) 49:811–854 Table 8 (continued) Federal State N Mean SD Min Median Max North Rhine-Westphalia Quantity all channels 84 5,218,629 1,560,673 3,288,458 4,670,050 10,464,162 Number B&M stores 84 1194 101.4 1038 1186 1362 Population (labor force) 84 9,255,150 169,422 8,962,720 9,257,316 9,556,575 Google Trends Book 84 62.6 6.559 53 61 82 Google Trends Amazon Kindle 84 27.06 12.93 5 25 73 Rhineland-Palatinate Quantity all channels 84 1,178,232 358,714 770,241 1,045,224 2,169,307 Number B&M stores 84 329.8 30.79 284 333.5 383 Population (labor force) 84 2,149,753 38,968 2,080,827 2,155,255 2,213,204 Google Trends Book 84 65.88 6.185 55 64 84 Google Trends Amazon Kindle 84 30.79 14.74 7 27.5 84 Saxony Quantity all channels 84 1,194,030 367,516 852,609 1,097,054 2,535,721 Number B&M stores 84 271.2 21.27 238 269.5 307 Population (labor force) 84 2,120,833 9956 2,091,661 2,120,822 2,141,032 Google Trends Book 84 61.9 7.279 51 60 82 Google Trends Amazon Kindle 84 22.48 10.31 5 21 63 Saxony-Anhalt Quantity all channels 84 573,037 201,924 284,508 510,070 1,149,577 Number B&M stores 84 115.5 11.75 97 116 136 Population (labor force) 84 1,173,376 17,382 1,142,313 1,176,068 1,205,128 Google Trends Book 84 58.56 6.354 47 58 75 Google Trends Amazon Kindle 84 24.57 13.5 2 21.5 71 844 Journal of Cultural Economics (2025) 49:811–854 Table 8 (continued) Federal State N Mean SD Min Median Max Schleswig-Holstein Quantity all channels 84 780,612 237,037 546,278 720,590 1,579,033 Number B&M stores 84 197.3 12.2 176 200 219 Population (labor force) 84 1,484,578 35,646 1,422,040 1,489,927 1,544,259 Google Trends Book 84 64.39 7.709 53 63 85 Google Trends Amazon Kindle 84 27.45 13.27 3 26 76 845 Journal of Cultural Economics (2025) 49:811–854 Table 9 Summary statistics by-product group Product group N Mean SD Min Median Max Children Books Quantity all channels 1176 329,661 303,005 28,008 211,995 2,139,147 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 Fiction Quantity all channels 1176 596,405 584,793 77,323 355,013 4,753,941 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 Guidebooks Quantity all channels 1176 224,833 210,871 19,005 135,797 1,561,292 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 Humanities Quantity all channels 1176 53,430 50,631 2944 34,048 362,306 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 Natural Sciences Quantity all channels 1176 30,242 27,125 2045 18,891 171,151 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 Nonfiction Quantity all channels 1176 133,848 130,071 13,981 85,486 1,036,771 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 846 Journal of Cultural Economics (2025) 49:811–854 B: IV Approach B.1: Naive OLS estimation ofEquation (1) See Table10. Table 9 (continued) Product group N Mean SD Min Median Max School Books Quantity all channels 1176 170,586 272,208 5384 76,379 2,135,074 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 Social Sciences Quantity all channels 1176 30,410 28,521 1610 18,847 189,124 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 Travel Quantity all channels 1176 83,379 79,644 8028 53,826 434,636 Number B&M stores 1176 382.8 338.5 49 251 1362 Population (labor force) 1176 2,974,561 2,560,595 321,313 1,990,775 9,556,575 Google Trends Book 1176 64 7.834 45 63 100 Google Trends Amazon Kindle 1176 27.01 13.08 0 25 87 Table 10 Naive OLS estimation Standard errors in parentheses Significance levels:*p < 0.05, **p < 0.01, ***p < 0.001 (1) Aggregated # Stores 0.000576* (0.000259) Google Trends 0.00341* (0.00144) Google Trends e-Reader 0.000000498 (0.00117) Federal State FE Yes Year-Month FE Yes # of observations 1176 847 Journal of Cultural Economics (2025) 49:811–854 B.2: First‑stage IV regression result ofEquation (2) See Table11. C: Robustness checks C.1: Book prices as“bad” control The change in the monthly average book prices is integrated as a “bad” control variable into the second stage of the IV regression. We calculate the change of the monthly average book price for the period 2011–2017 based on our scanner data and use this parameter as a “bad” control variable in our IV estimation approach from Equation (1). Of course, the price is a “bad” control variable in this approach because we now regress a quantity parameter on a price parameter in the second stage of our 2SLS estimation. The results of our IV estimation when controlling for price changes is presented in Table12. Table12 shows that the newly included variable Book price has no significant effect on the dependent variable total print book sales. The coefficient of this variable is expected to be biased due to reverse causality because quantities are basically regressed on prices. Nevertheless, from Table12 one can see that the number of bookstores is still significant and has a positive sign. This implies that a decreasing number of physical bookstores also leads to lower print book sales per capita when we control for book price changes, which is in line with the findings presented in the main text. Table 11 First-stage regression results of baseline IV estimation Standard errors in parentheses Significance levels:*P < 0.05, **P < 0.01, ***P < 0.001 (1) # Stores Pop 0.000410*** (0.00000652) Pop. squared −4.71e−11*** (9.16e−13) Google Trends 0.592** (0.184) Google Trends e-Reader 0.691*** (0.159) Federal State FE Yes Year-Month FE Yes # of observations 1176 848 Journal of Cultural Economics (2025) 49:811–854 C.2: Number ofuniversity students ascontrol State and time specific sociodemographic effects might confound our baseline analysis, in particular changes in the education level. To control for those effects, we use data on the number of university students by federal states30 as a proxy for education. Given that these data are only available annually, we aggregate sales taken from scanner data to that level and run a IV regression on a yearly basis. In particular, we regress print book sales per capita on our treatment variable, the number of physical bookstores, and the lagged number of students (as well as other covariates mentioned in Equation (1)) in the second stage of our 2SLS estimation. The results are presented in Table13. One can see that the lagged number of university students has no significant effect on print book sales per capita in Germany. Note that there is little variation in the number of students such that the federal state fixed effect and the number of students are closely related. However, the lagged number of physical bookstores still has a positive and significant effect on print book sales, which again is in line with the findings presented in the main text. It appears surprising that the result remains statistically significant even though the number of observation is only one twelfth of that in the main text (yearly data instead of monthly data). Table 12 IV estimation with monthly average book price change as bad control variable Standard errors in parentheses Significance levels:*P < 0.05, **P < 0.01, ***P < 0.001 (1) Aggregated # Stores 0.000319*** (0.0000683) Google Trends 0.00419*** (0.00107) Google Trends e-Reader 0.000595 (0.00112) Book Price −0.00887 (0.00638) Federal State FE Yes Year-Month FE Yes Anderson–Rubin Wald F-statistic 10.24 Kleibergen–Paap Wald F-statistic 1194.4 # of observations 1176 30 These data are provided by the German Federal Statistical Office, see https:// www. stati stisc hebib lioth ek. de/ mir/ recei ve/ DESer ie_ mods_ 00000 113. 849 Journal of Cultural Economics (2025) 49:811–854 C.3: Estimations using acombination ofscanner andsurvey data In order to incorporate e-Book sales into the scanner data set, we follow these steps: 1. Aggregate the scanner data based on federal state quarters for the period 2014Q1 − 2017Q4 . 2. Utilize survey data to calculate the ratio of e-Book to print book sales for each federal state quarter. 3. Apply the calculated ratio to estimate e-Book sales using the scanner data. Table 13 IV estimation on a yearly basis when controlling for the number of university students Standard errors in parentheses Significance levels:*P < 0.05, **P < 0.01, ***P < 0.001 (1) Aggregated # Stores 0.00642* (0.00316) Google Trends −0.0351 (0.0550) Google Trends e-Reader 0.0828 (0.0470) Studentst−1 −0.000000970 (0.00000912) Federal State FE Yes Year FE Yes Anderson–Rubin Wald F-statistic 2.853 Kleibergen–Paap Wald F-statistic 22.99 # of observations 84 Table 14 IV regression results including e-Book sales Standard errors in parentheses Significance levels:*P < 0.05, **P < 0.01,***P < 0.001 (1) Sales e-Commerce (incl. e-Books) # Stores 0.000547*** (0.0000502) Federal State FE Yes Year-Quarter FE Yes Anderson–Rubin Wald F-statistic 60.75 Kleibergen–Paap Wald F-statistic 13927.8 # of observations 224 850 Journal of Cultural Economics (2025) 49:811–854 Subsequently, the obtained e-Book sales data can be used to replicate our estimations in Sect.4.2 with comprehensive data on both print and digital book sales. Estimation results using print and calculated e-Book sales for the sales channel e-Commerce as dependent variable are presented in Table14. One can see that the number of physical bookstores still has a significant positive effect on the book sales per capita in the e-Commerce when also including e-Book sales into the regression for the book scanner data (cf. column (3) in Table2 of Sect.4.2.1). This finding also shows that the findings reported in Sect.4.2 are robust when information on e-Book sales are included into the analyses, as e-Book sales apparently do not compensate for the decrease in sales triggered by the exit of physical bookstores, as this would require a negative coefficient of #stores . Thus, a significantly positive (or insignificant) coefficient does not challenge the previous findings. C.4: OLS Estimation withlagged number ofbookstores Additional variants of Equation (2) were run based on OLS using the yearly or monthly lagged number of bookstores as explanatory variables. In general, this estimation strategy should at least partially resolve potential reverse causality between the number of bookstores and book sales per capita since book sales in t should not affect the number of bookstores in t−1 . The estimation equation can be formulated as follows: The results based on yearly data are presented in Table3. Here, salesi,t is either measured in levels (column 4) or in logs (column 5). The results the lagged OLS estimation using monthly data are depicted in Table15 and imply that the number of bookstores have a positive effect on the sales per capita in the offline sales channel (column (2)). This means that a closing bookstore in t=−1 significantly lowers the book sales per capita in t=0 (significant on salesi,t = 𝛼 + 𝛽bookstoresi,t − 1 + 𝜇i + 𝜇t + 𝜖i,t. Table 15 OLS Estimation with lagged number of bookstores as treatment variable Standard errors in parentheses Significance levels:*P < 0.05, **P < 0.01, ***P < 0.001 (1) (2) (3) Aggregated Offline e-Commerce #Storest−1 0.000557 0.000562* −0.00000515 (0.000272) (0.000222) (0.0000983) Google Trends 0.00276 0.00238* 0.000378 (0.00168) (0.000920) (0.00111) Google Trends e-Reader −0.000279 −0.000431 0.000152 (0.00141) (0.00105) (0.00109) Federal State FE Yes Yes Yes Year-Month FE Yes Yes Yes # of observations 1008 1008 1008