The Impact of Ebook Distribution on Print Sales: Analysis of a Natural Experiment

Digital distribution introduces many new strategic questions for the creative industries, notably how the use of new digital channels will impact sales in established channels. We analyze this question in the context of ebook and hardcover sales by exploiting a natural experiment that exogenously delayed the release of a publisher’s new Kindle ebooks in April and May 2010. Using new books released simultaneously in ebook and print formats in March and June 2010 as the control group, we find that delaying ebook availability results in a 43.8% decrease in ebook sales but no increase in print book sales on Amazon.com or among other online or offline retailers. We also find that the decrease in ebook sales is greater for books with less pre-release buzz. Together we find no evidence of strong cannibalization between print books and ebooks in the short term, and no support for the sequential distribution of books in print versions followed by ebook versions.


Introduction
Digital distribution channels introduce a variety of new opportunities and challenges for the creative industries (e.g., music, film, and publishing). The book market represents one prominent (and understudied) setting where publishers face the challenge of selling books in both print and digital formats. Since the launch of the Kindle ebook reader platform in late 2007, the publishing industry has witnessed significant growth in ebook adoption and an associated growth in ebook sales. The Association of American Publishers reported that, ebook sales revenue was $345 million from January to October 2010, a 171 percent growth from the same period in 2009, and that ebook sales made up nearly 9 percent of total trade book sales during this period in 2010 (Association of American Publishers 2010). Similarly, Amazon.com, the largest online seller of print and digital books, reported that Kindle sales surpassed Amazon's total hardcover sales in July 2010, and surpassed total print sales as of April 1, 2011 (Amazon.com 2011). Growth in the popularity of the ebook platform has continued and from 2013 to 2015 ebooks comprised around 20% of all book sales (Association of American Publishers 2016).
As the popularity of ebooks grows, the book industry is engaged in an active debate about how the use of ebook channels will impact print sales. On one side of the debate ebook platform providers, such as Amazon, claim that ebooks do not cannibalize print sales, but rather represent mostly incremental sales.
For example, Jeff Bezos, CEO of Amazon, observed in an earnings call with analysts (SeekingAlpha 2009), "When people buy a Kindle, they actually continue to buy the same number of physical books going forward as they did before they owned a Kindle. And then incrementally, they buy about 1.6 to 1.7 electronic books, Kindle books, for every physical book that they buy." On the other side of the debate many book publishers worry that because print books and ebooks offer essentially the same content in different formats, and that the consumption of ebooks will cannibalize print book sales, particularly those that are priced significantly higher than their ebook counterparts.
Carolyn Reidy, CEO of Simon & Schuster, put forward this view in a New York Times article (Rich 2009), "What a (book) consumer is buying is the content, not necessarily the format." "The publishing plan is focused on maximizing velocity of the hardcover before Christmas." Brian Murray, CEO HarperCollins November 2009: Viacom/Schribner delays the ebook release of Stephen King's new novel by 6 weeks after the hardcover release date.
"We think that this publishing sequence gives us the opportunity to maximize hardcover sales." Adam Rothberg, Spokesperson Scribner Early 2010: Hachette Book Group delays the ebook release of nearly all of its titles by 3-4 months after the hardcover release date.

"I can't sit back and watch years of building authors sold off at bargain-basement prices."
David Young, CEO Hachette Early 2010: Simon & Schuster delays the ebook release for 35 major titles by 4 months after the hardcover release date.
"The right place for the e-book is after the hardcover but before the paperback." Carolyn Reidy, CEO Simon & Schuster These sorts of cannibalization concerns have led many publishers to delay ebook releases in the hopes of not cannibalizing hardcover sales (see Table 1). However, discussions we have had with publishers suggest that these decisions have been made based on gut feel and instinct, rather than careful empirical analysis.
In this study we attempt to address the research question of whether and how ebooks cannibalize print book sales by providing empirical evidence on how delaying the release of ebooks impacts both hardcover sales and ebook sales. To do this we use a novel natural experiment where, over the course of a 2-month period, a particular publisher stopped providing Kindle ebooks for its new releases to Amazon due to a dispute between the publisher and Amazon. During the dispute period, however, the publisher continued to provide its new releases in hardcover format to Amazon and other print book retailers.
Because this publisher had previously released ebooks at the same time as the print release, this event, which we describe in more detail below, creates a scenario where ebook release dates are exogenously delayed by between 1 to 8 weeks (compared with hardcover release dates) for a set of titles, providing a unique opportunity to study the cross-channel effect between ebooks and print books.
Our analyses show that, in total, delaying ebook release dates relative to their hardcover counterparts results in no increase in hardcover sales either on Amazon or among other physical and online retailers.
However, delaying the ebook release date does result in a large decrease in ebook sales and total profit to the publisher. 1 The decrease in ebook sales is evident not only in the lost sales during the period of release delay but also through lower demand in the post-release period. Additional analyses reveal that books with a smaller number of pre-release reviews suffer a greater decrease in ebook sales. In light of these findings, we suspect that some ebook buyers may stay loyal to the digital channel or format and wait to purchase ebooks in the event of a delayed release, especially for books with more pre-release buzz.
Further, we find that delaying Kindle release does not lead to an increase of ebook sales in a competing digital channel, Apple iBooks.
We believe this study makes several key contributions. First, to the best of our knowledge, this study is the first one to investigate the sales performance of the sequential distribution of digital and physical products in the publishing industry. Our empirical results offer important managerial implications for book publishers and potentially firms in other creative industries as well. The key implication is that the strategy of delaying the release of ebooks could do more harm than good by not increasing physical sales but significantly reducing digital sales. Second, our study adds to a growing academic literature analyzing the impact of digital distribution channels on existing sales channels. We extend this literature by using a novel natural experiment and sales data collected from the understudied electronic publishing industry, an industry that is growing in importance. Third, these results suggest that consumers can form strong channel (or format) preferences between physical and digital products and even strong platform preferences among different digital providers. Finally, our framework of how cross-channel effects are moderated by different demand-related characteristics can assist academics and mangers in designing optimal digital release strategies for digital distribution channels.

Literature Review and Theoretical Background
Our research is related to the information systems and marketing literature analyzing cross-channel cannibalization (e.g., Deleersnyder et al. 2002, Smith and Telang 2009, Gong et al. 2015, among many 1 We note that these lower ebooks sales will also lead to lower total profits in our setting given that publishers' margins on ebooks are very similar to their margins on higher priced print books. See for example Rich (2010) who notes that publisher profit (before overhead) on a $26.00 print book is approximately $4.05 while publisher profit on the same book sold in ebook format is between $3.51-$4.26 for a $9.99 ebook and $4.56-$5.54 for the same ebook sold at $12.99. others) and optimal timing of sequential distribution (e.g., Moorthy and Png 1992, Lehman and Weinberg 2000, and August et al. 2015. One prominent research stream on cross-channel cannibalization investigates whether and how the addition of the Internet channel cannibalizes sales in established offline channels. Deleersnyder et al. (2002) find that the introduction of online newspapers resulted in a relatively small cannibalization of physical newspaper sales; Biyalogorsky and Naik (2003) find that the introduction of online storefronts for music did not significantly cannibalize physical record sales; Waldfogel (2009) shows that YouTube viewing has only a small negative impact on television viewing. In recent years, there are also studies that investigate how the opening of an offline retail store affects online sales in the growing omnichannel retailing literature (e.g., Brynjolfsson et al. 2013). For instance, Avery et al. (2012) find that the introduction of a retail store selling high-end apparel, accessories, and home furnishings cannibalizes sales in the catalog channel but not the Internet channel in the short run and increases sales in both direct channels in the long run. In addition, Brynjolfsson et al. (2009) show that the cross-channel competition between online and offline retailers is more significant for mainstream products compared with niche products. (notably for our study) impact physical sales. They show that the removal of legal video content on iTunes is associated with a significant increase in the demand for pirated content but has no statistical impact on physical DVD sales. Similarly, Danaher et al. (2015) analyze how the addition of ABC television series to the Hulu platform impacted piracy and DVD sales, finding that the use of digital distribution significantly reduced the demand for pirated content but had no impact on DVD sales.
Finally, Gong et al. (2015) study the cross-channel cannibalization between digital rentals and digital downloads of movies by analyzing the impact of price discounts on own-and cross-channel sales, finding that price promotions for digital downloads lead to increased digital rentals.
Although cross-channel effects have been studied in the context of movies and television shows, they have received limited attention in the context of the publishing industry. One exception is the study by Kannan et al. (2009) who investigate the optimal pricing strategies of the National Academies Press (NAP) that sells both print and digital (PDF) books through its official website. Another related study by Ghose et al. (2006) examines the online sales of both used and new print books. They find that Internet channels for used books result in a relatively small cannibalization of new book sales. Our study extends this prior work by providing empirical evidence of cross-channel effects between ebooks and print books sold in both online and offline stores.
Our study also contributes to the literature on the optimal timing of sequential distribution. In this literature, Moorthy and Png (1992) analyze how a seller can sequentially introduce a high-end and a lowend version of a durable product, finding that sequential introduction is preferred over simultaneous introduction when cannibalization is a concern. In the movie industry, release windows for different channels are an important business decision for practitioners and have been extensively investigated in the literature (e.g., Lehmann and Weinberg 2000, Prasad et al. 2004, August et al. 2015. Movies are traditionally released first in theaters and later in home video channels (purchase and rentals). However, several analytical and empirical studies suggest that simultaneous distributions may outperform sequential distributions. For instance, Hennig-Thurau et al. (2007) conduct a conjoint analysis and show that a simultaneous release of movies in theaters and on rental home video can be optimal for movie studios but will hurt other players such as theater chains. In their theoretical work, Calzada and Valletti (2012) also find that the simultaneous release of theatrical and video versions of movies can be optimal when the movie studio is integrated with the exhibition and distribution channels. In addition, Mukherjee and Kadiyali (2011) empirically test cross-channel substitution between two post-theatrical channels, home video purchases and rentals, and show that windowing (or sequential distribution) reduces the sum of revenues across two channels.
In the traditional publishing industry, books are often published sequentially first in hardcover and then in paperback formats, which follows the logic that the most expensive format should be released first (Lulu 2015). To avoid cannibalization between ebooks and print books, the main approach identified by publishers is to release them sequentially and more specifically delay the release of ebooks. This study thus adds to the literature on optimal timing of sequential distribution by comparing sequential and simultaneous distributions of print books and ebooks. Our research differs from prior movie studies in that we use a novel natural experiment that creates an exogenous and unintended delay in the availability of some ebook titles. This quasi-experimental setting allows us to establish causality and avoid many confounding factors common in related cross-channel settings.
From a theoretical perspective, when the digital release of a book is delayed, it may or may not increase the sales of its print version. Previous research on product stockouts has shown that consumers who face stockouts of a certain brand may switch to other sizes and varieties of the same brand (e.g., Emmelhainz et al. 1991). Similarly, ebook consumers may switch from the focal title's ebook to its print book when its ebook is delayed, simply because ebooks and print books offer the same content and could be strong substitutes for each other. If this happens, then the print sales of the book would increase when ebooks were delayed. On the other hand, there are reasons to believe that ebook and print markets could be distinct due to ebook consumers' preferences for digital distribution channels. Marketing theory shows that markets can be segmented based on a consumer's channel preferences (Kotler 2002). Thus, it is possible that some ebook consumers may not consider print books as substitutes for ebooks. In this case, the sales of the print book might not change, and digital sales could fall if consumers were unwilling to wait for the eventual digital release of the book.
Our finding that the effects of delaying digital releases can be moderated by pre-release buzz is related to several papers documenting similar types of effects. Simester at al. (2009) show that the effects of current advertising on future sales are moderated by brand awareness among consumers. Earlier studies such as Elberse and Eliashberg (2003) also report that marketing buzz can directly affect product sales. With the rise of Internet channels, online word-of-mouth has become an important indicator of consumers' awareness and an important driver of product sales (e.g., Luan and Sudir 2010). Publishers often actively court, and in some cases pay, top reviewers to create online word-of-mouth for their new books (Coster 2006). Furthermore, our result that delaying digital releases leads to no increase in print sales is consistent with the literature suggesting that consumers potentially have a strong preference for their chosen channel. Vernik et al. (2011) show that music consumers may have a strong preference for the digital format, and this preference can drive product sales. Frambach et al. (2007) demonstrate that consumers can develop higher preferences for the online purchasing channel when they have a more favorable Internet experience. Zentner et al. (2013) find that online channels could change consumers' consumption patterns and product choices.

Data and Description of Natural Experiment
An ideal starting point for understanding how delayed release dates for ebooks affect sales would be a pure experiment where a publisher randomly assigns books to the treatment group with different amounts of delay. Due to the business consequences involved in this setup, most publishers are reluctant to implement this type of randomized experiments. Lacking this data, our research employs a natural experiment that approximates this ideal setup. This event results in the release schedule of Kindle and print titles shown in Table 2. Notably for our purposes, this event creates a "natural experiment" where a sample of Kindle titles are delayed by between one to eight weeks relative to their print counterparts.
To analyze the results of this "experiment", we obtained data from the publisher in question, covering and digital sales but also between overall print and digital sales. 3 We then divide this sample into two groups as follows: 1. Control Group: new books that are released from March 1 to March 31 and from June 1 to June 30. These books are released simultaneously in both print and Kindle formats.
2. Experiment Group: new books that are released from April 1 to May 31. These books are released first in print and then one to eight weeks later (on June 1) in Kindle format.

Comparing the Control and Experiment Groups
Based on discussions with the publisher in question there is no reason to believe that the control group will be systematically different from the experiment group. Book release schedules are set at least six months in advance, they did not change as a result of the dispute, and the timing of the dispute was not influenced by the publisher's upcoming releases. The publisher's statements that the control and treatment groups are similar is consistent with available data regarding whether books in the control group have a similar profile to books in the experiment group on observable dimensions -which in our case include the list price, genre category (fiction or nonfiction), pre-release buzz, book weight, height, length, width, and number of pages. Summary statistics on these dimensions are provided in Table 3 separately for books in the control group and books in the treatment group. Fiction is a dummy variable, which equals 1 if the genre of a book belongs to the fiction category and 0 otherwise. 4 Pre-releaseReviews is the number of Amazon reviews prior to print release.
Other variables are either count or continuous variables. The statistics in Table 3 show that the control and experiment titles are very similar along all these observable dimensions. We can also empirically test whether any of the observable variables can be used to predict a particular book belonging to the experiment group. To accomplish this, we estimate the following Probit model: where Experiment i is an indicator of whether book i belongs to the experiment group, and Z i is a vector of independent variables described in Table 3. The results from estimating this model are reported in Table   4.
The results in Table 4 show that none of the coefficients is statistically significant at the 5% level. This is again consistent with comments from our publisher that books in the control group are materially similar to -and thus a reasonable control for -books in the experiment group.

Descriptive Analyses of Book Sales
When studying the sales patterns of these books, we focus on the sales in the first twenty weeks since each book's initial release in their respective channels. We do this because this is the period where the majority of sales occur, allowing us to keep the sales numbers comparable across different books. Figure   1 presents the average weekly sales for books in our sample in the Amazon Print, BookScan Print, and Kindle Digital formats. Amazon Print unit sales are included in the BookScan Print unit sales, so the curve for Amazon Print is always below the curve for BookScan Print. All three curves follow a similar and overall decreasing temporal pattern.

Figure 1: Average Weekly Sales (Kindle Digital on the secondary axis)
To get an initial assessment of whether there are differences in the sales patterns between the control and experiment groups, we present the summary statistics in Table 5

Digital Sales Print Sales
Week Since Release

Amazon Print
BookScan Print Kindle Digital suggest that digital sales are a significant sales channel. Digital sales make up nearly half of total sales on Amazon for this publisher, which is generally consistent with available data regarding sales patterns (e.g., Amazon.com 2011). Second, a quick comparison of the control and experiment groups reveals that ebook sales in the experiment group are significantly lower than those in the control group regardless of whether the ebook sales are measured since the print release or Kindle release. We note that print and Kindle release dates are the same for the control group; the digital sales after print release are smaller than the digital sales after Kindle release for the experiment group because digital sales are zero in the initial few weeks after print release due to the treatment. Although the mean and median of Amazon print sales in the experiment group are slightly higher than in the control group in these summary statistics, they seem to be nowhere near large enough to compensate for the lost ebook sales. Third, the means of the BookScan print sales for these two groups are different, but the medians are quite close (173 vs. 161), suggesting that a few very popular books in the control group are driving up the average sales (and standard deviation as well) recorded by BookScan.

Analyses and Results
Because our dependent variables (weekly print and digital sales) are count data and we find evidence of overdispersion, we estimate a negative binomial panel regression model (Hausman et al. 1984, Cameron andTrivedi 2013). Let Y it represent the unit sales generated by a particular version (print or digital) of book i in week t and follow a Poisson distribution with the conditional mean : ( | ) = ! , = 0,1,2,3, … An advantage of this model is that it allows us to directly control for any factors that may have affected book sales during this timeframe that may differ between the control and experiment groups. We note that if sales of the control books are statistically representative of what sales of the experiment books would have been if the Kindle version had been released along with the print version (as we believe is the case), then having Experiment i only in the explanatory variable is sufficient to test the effects of the delay of digital releases.
As our data is a book-week panel, we report the results from the population averaged negative binomial estimators for panel models (Cameron and Trivedi 2013) to account for any potential within-panel correlation. Alternatively, we could employ random effects estimators for the panel model, which produce the subject specific estimates (i.e., what would happen to a particular book if its digital release was delayed). Since we are more interested in the effect of the digital release delay on the general population, 5 As a robustness check, we also include a squared time trend term, WeeksSinceRelease it 2 , in addition to the linear time trend term, WeeksSinceRelease it , our results remain largely similar as reported in Table 6 and are available upon request. 6 There are many observations that have zero new reviews in a given week. To avoid missing values for the average rating variable, we control for the average rating of all Amazon reviews received by book i until week t, instead of the average rating of new Amazon reviews received by book i in week t.
or what would happen if the digital release of an average book were delayed, we present the results from the population averaged panel model. Note that the population averaged estimates and subject specific estimates are equivalent for linear models but not for nonlinear models such as negative binomial regressions.  (3) examine the weekly print and digital sales in Weeks 1 to 20 after print release. As the digital sales of the initial few weeks (since print release) are zero for the experiment group, we examine the digital sales in Weeks 1 to 20 after Kindle release in Column (4) so that the experiment group also has 20 weeks of positive sales and the sales patterns are aligned between the two groups. The coefficient estimates on the Experiment variable in Columns (1) and (2) are both statistically insignificant, indicating that delaying Kindle releases does not lead to a significant change in overall and Amazon print sales, consistent with our summary statistics above. This implies that when the digital version of a book is unavailable on Amazon, consumers do not seem to switch to buying print books from the same retailer (Amazon), or from other online or brick-and-mortar print book retailers. The coefficient estimates on the Experiment variable in Columns (3) and (4) (3), but the analysis in Column (4) further indicates that the experiment group also faces lower demand after the Kindle ebook is released. In sum, 7 The magnitude of the coefficient estimate on Experiment in Column (4) is larger than that in Column (3) because in Column (4) the digital sales after Kindle release are consistently lower for the experiment group than for the control group, but in Column (3) the digital sales of the experiment group can be larger in some weeks than that of the control group (e.g., for a book whose digital release is delayed by 8 weeks, its digital sales after print release in Week 9 is likely to be larger than the digital sales of a control book in the corresponding week), although the digital sales of the initial few weeks are zero for the experiment group.

The Effects on Print Sales and Digital Sales
these results suggest that delaying the digital release of books results in a decrease in digital sales but no increase in print sales.

Table 6: Effects of the Delay of Digital Release on Print and Digital Sales
Robust standard errors are in parentheses. **Significantly different from zero, p < 0.01. * p < 0.05. Time dummies are included in the estimation.
The results for the control variables are largely consistent with expectations. First, the coefficient estimates on WeeksSinceRelease in all columns are negative and statistically significant at the 1% level.
The coefficient estimate of -0.124 in Column (1) of Table 6 means that print sales decline at roughly 11.7% (i.e., e -0.124 -1=-0.117) per week as shown in Figure 1. Second, consistent with expectations the coefficient estimates on PrintPrice are negative and statistically significant at the 1% level in Columns (1) (1)

BookScan Print Sales
(2) Amazon Print Sales releaseReviews are all positive and statistically significant at the 1% level, implying that the pre-release buzz of a book is predictive of both its print and digital sales after release. Fourth, the coefficient estimates on Fiction are statistically significant at the 1% level in Columns (2) to (4) but not in Column (1). This suggests that the aggregate print sales of fiction and non-fiction books are similar in all channels, but consumers on Amazon seem to purchase fewer fiction print books than non-fiction print books and more fiction ebooks than non-fiction ebooks. Finally, among the other book characteristics, Length and Width are statistically correlated with the book sales of both formats.

Interactions with Pre-release Reviews and Genre Category
To further examine how the effects of the delay of digital releases vary across different types of books, we test whether two exogenous demand-related characteristics, pre-release buzz and the genre of a book, moderate our treatment effects. 8 Note that the number of pre-print-release Amazon reviews is not driven by sales after the book's release; hence, it is an exogenous variable that is not directly affected by whether 8 In untabulated results, we examine two other exogenous moderators that are related with the popularity of a book's author and thus may also be good measures of anticipated demand. Without access to data on each author's prior sales, we utilize the number of prior books as a proxy of productivity and the number of customer reviews as a proxy of book sales. We first collect the list of prior books published by each author in our sample and then collect all the customer reviews received by those prior books on Amazon before our study period. We construct two variables, AuthorPriorBooks i (number of prior books by the author of book i) and AuthorPriorBookReviews i (number of Amazon reviews on all prior books by the author of book i), and conduct similar analyses outlined in this section to test the interactions between the Experiment variable and these two variables. We do not find any significant interactions from these analyses. These findings suggest that the reviews on the quality of a book itself (i.e., pre-release buzz) seem to play a more important role than the profile of the book's author in the event of a digital release delay. the book's digital version is delayed. To conduct these two analyses, we add the interaction terms of Experiment × Pre-releaseReviews and Experiment × Fiction to the list of explanatory variables , respectively, and then estimate the model for the same four dependent variables.
In Table 7, we investigate how the level of pre-release buzz moderates the effects of the delay of digital releases. In our sample of 182 books, the average number of pre-release Amazon reviews is 5.7, 9 and the minimum and maximum of pre-release Amazon reviews are 0 and 243, respectively. The coefficient estimates on Experiment × Pre-releaseReviews imply that the effect of the delay of digital releases on print sales does not vary for books with different levels of pre-release buzz (Columns 1 and 2), but the effect on digital sales varies for books with different levels of pre-release buzz (Columns 3 and 4).
Specifically, the coefficient estimate on the interaction term in Column (3) is 0.310 and statistically significant at the 5% level, while the coefficient in Column (4) is 0.395 and statistically significant at the 1% level. Since the main effects of Experiment on digital sales in Columns (3) and (4) are negative, a positive and significant interaction effect implies that the negative effect of the treatment is weaker for books with more pre-release buzz and stronger for books with less pre-release buzz. In other words, delaying Kindle releases reduces the digital sales of books with less pre-release buzz more than it does for other books.
To put these coefficient estimates into their proper economic perspective requires more effort than for interactions in linear models. For nonlinear models, such as negative binomial regressions, there is no single economic interpretation of the interaction effect, as it varies for different combinations of the two interacting variables (Hilbe 2011, Appendix A). In Table 8, we report the effects of the delay of digital releases on digital sales at different levels of pre-release buzz. We observe a decreasing trend in the magnitude of the treatment effect as the number of pre-release reviews increases from 0 to 5. However, the change in the treatment effect is nonlinear in the levels of Pre-releaseReviews. Note that according to the main results from Table 6, the average effect of the delay of digital releases is -43.8% on digital sales after print release and -48.2% on digital sales after Kindle release, respectively.    Table 9 are statistically insignificant, although the main effects of Experiment and Fiction remain largely similar as in Table 6. These results suggest that the effects of the delay of digital releases do not vary between fiction and nonfiction titles.

Digital Sales in other Channels
Although our results suggest that there is no substitution from Kindle sales to online or offline print sales when the Kindle release is delayed, it is still possible that consumers may purchase ebooks from other digital channels. A potential confounding factor in our experiment is that Apple's iBookstore opened in early April 2010 and included content from this publisher. One could argue that at least part of the drop in Kindle sales for the experiment group could be attributed to the opening of the iBookstore if the decrease in sales came from consumers who substituted from the (unavailable) Kindle channel to the iBookstore.
We first note that the market share held by iBooks in our study period (Carnoy 2010, Hoffelder 2015 is unlikely to explain the drop in ebook sales that we observe. In addition, iBookstore purchases can only be viewed on Apple Mac and iOS (e.g., iPhone, iPod Touch, iPad) devices, reducing the potential market segment that could make the tradeoff between the Kindle and iBookstore. We also note that we have already controlled for this effect in part by adding a set of time dummy variables as this event is likely to have a systematic shock on the demand of all books.
However, to directly evaluate the extent to which consumers switch from Kindle to iBooks in the event of Kindle release delay, we obtained this publisher's sales on the iBookstore for the control and experiment titles in our sample. With these data we run a regression similar as in Table 6 but replacing the dependent variable with weekly iBooks sales. This test would reveal whether there is a significant difference in the iBooks sales between the control and experiment groups.  Robust standard errors are in parentheses. **Significantly different from zero, p < 0.01. * p < 0.05. Time dummies are included in the estimation.

Discussion
Our research analyzes how the availability of a product in a digital channel impacts sales in physical and other digital channels -in our case in the digital and physical channels for books. This question is important to both managerial and academic audiences. From a managerial standpoint, content owners across a variety of industries are making decisions about whether, when, and how to include digital 10 We further note that our analyses in Table 10  products in their existing set of distribution channels, and where to place these products into their existing product release cycles. These decisions are particularly salient for book publishers, many of whom have experimented with releasing ebook versions in between the (high margin) hardcover release date and the (lower margin) paperback release date. From an academic perspective, our work adds to a growing literature analyzing the impact of new digital distribution channels on physical sales.
We analyze this question using a novel "natural experiment" where a publisher stopped releasing new ebook content to Amazon for a period of about two months. Because this publisher still released print copies of their titles to Amazon during this period, the net effect of this event is that books released during this timeframe were delayed on the electronic channel relative to the print channel by between one to eight weeks.
Our results show that delaying the publication of the ebook relative to its print version causes a large and persistent decrease in digital sales. This negative impact on digital sales is more pronounced for books with less pre-release buzz. However, the effects of the digital release delay do not vary with the genre category (fiction vs. nonfiction) of the book. Contrary to the common belief in the industry that delaying digital releases increases hardcover sales, we find no significant increase in print sales in both online and offline channels.
These results point to the possibility that, in general, consumers may be relatively more tied to their mode of consumption (physical or digital) than they are to a particular product. Said another way, when facing new digital channels, publishers and other media firms have frequently conceptualized the consumer's decision process as being driven by product choice first and then channel choice. This conceptualization is seen in the frequent strategy to delay digital availability as a way of retaining physical sales. Our results suggest that in general consumers choose their channel (digital versus physical) first, and then restrict their choice set to the products available in that channel. To be clear, our results do not suggest that digital channels will not cannibalize aggregate physical sales in the long term; they certainly will. Rather, we believe our results suggest that as a managerial question, given that digital channels exist and that some consumers have preferences for these channels versus physical channels, refusing to provide books and other products in digital channels is unlikely to result in increased short-term physical sales for a particular title.
Our results are not without limitations. First, and most notably, our results are based on the assumption that the books released immediately before and after our "experiment" period are good controls for the books released during the experiment. We believe this is true based on the empirical tests above and based on conversations with the publisher in question. However, we cannot conclusively rule out the possibility that unobserved differences between the control and experiment groups may be driving some of our results.
Second, our results are situated within a particular market (books) and at a particular stage of development of that market. Ebook adoption has experienced rapid growth, from 9% of all book sales in 2010 (Association of American Publishers 2010) to 20% in 2015 (Association of American Publishers 2016), interaction between these two channels has increased in importance in the publishing industry. Our results suggest that as the ebook market grows, it is likely that delaying ebook releases will result in more substantial decreases in digital sales without significantly improving print book sales. Nonetheless, it would be useful for future research to investigate how the cannibalization between print books and ebooks evolves at different stages of ebook adoption.
Finally, our natural experiment is slightly different from officially delaying the digital release because no specific future release time is set. However, in our context because consumers face more uncertainty regarding whether the digital release would ever be available, they are more likely to switch to other formats or channels than they would be if they knew for certain when the digital content would be available. Because we still observe no substitution from Kindle to print or iBooks sales, we believe our result of lost digital sales is likely to hold when the future release date is known to consumers.