Success Drivers of Fiction Books: An Empirical Analysis of Hardcover and Paperback Editions in Germany

Abstract

Despite consumer's widespread interest in reading, enjoying, and buying books, very little research has considered the critical success factors of books. This article focuses on similarities and differences between success factors when selling fiction books sequentially in hardcover and paperback form. Using a large dataset gathered in cooperation with a leading German market research institute, this work estimates a seemingly unrelated regression model and finds that key marketing considerations—such as popular authors (stars), special genres, publisher strengths, and book cover designs—have different (and sometimes conflicting) influences on sales of the same book title depending on the edition format.

Notes

Christina Schmidt-Stölting and Eva Blömeke are post-doctoral researchers at the Institute for Marketing and Media, University of Hamburg, Germany. Michel Clement is a Professor for Marketing and Media Management at the Institute for Marketing and Media, University of Hamburg, Germany.

¹In comparison, the Association of American Publishers (2008) estimates a net sales volume of $25 billion for the United States.

²The general lack of research on marketing books has already been mentioned by Compaine (1978, p. 198).

³We thank an anonymous reviewer for calling this issue to our attention.

⁴Hennig-Thurau, Houston, and Walsh (2006) also built their work on cognitive categorization theory (e.g., Cohen & Basu, 1987) because, in the movie market, it is common to consume the same product twice but in two different versions (e.g., watch a movie at the cinema and later on DVD). For the book industry, this is not applicable because books are usually not consumed more than once.

⁵As a robustness check, we also estimated a model in which we include the hardcover sales in the paperback equation (see Appendix).

^a Sales multiplied with price—hardcover (HC): M = 330,390.83, SD = 599,769.55; paperback (PB): M = 293,590.02, SD = 538,292.22.

*GfK = Gesellschaft für Konsumforschung (market research company).

⁶This dataset is comparable to the BookScan audience measurement system, which measures book sales for the U.S. book publishing industry (Andrews & Napoli, 2006).

⁷We chose the timeframe from 2003 to 2006. Here, we include one full year of sales for each book to capture all potential seasonal effects. Thus, a book released in December 2005 was followed until December 2006. We cover 4 years for which we include all books that were within our sampling restrictions.

⁸We explicitly set this restriction to focus on the effects of the marketing mix on both release editions (hardcover and paperback). However, this leads to a selection effect as a result of choosing books that are sequentially released. We note that this might lead to potential biases because hardcover books that sell very poorly may not be followed by paperback versions.

⁹In only two cases in our sample was a paperback version published before the 1-year time limit.

¹⁰Based on an interview we conducted with Frank Sambeth (Chief Operating Officer, Random House, Germany) in February 2008, we have set this sales limitation lower than his initial statement of at least 5,000 books to reduce the bias introduced by ignoring less successful books. We note that this selection process eliminates our chances of identifying the success drivers of “low sellers” (books with sales < 1,500 per year) or analyzing the reasons for unexpectedly low-selling books—information that would be of some interest to managers. We acknowledge this limitation, which holds true for many empirical studies (e.g., Elberse & Eliashberg, 2003), but we chose 1,500 books as an appropriate limitation to generate a full sample of all fiction books that have been released as hardcover and paperbacks within the considered time period. This threshold includes all books that received at least some attention from the management team in Germany; with average sales of < 30 books per week—thus, our sample includes somewhat less successful books.

¹¹We use a logarithmic transformation of the total sales for both editions to achieve a more favorable data structure with less extreme values (see Danaher, 2002).

¹²Because we want to measure the success of fiction books, we only include biographies that are fictional biographies like “The Bookseller of Kabul” (Asne Seierstadt, 2003) or “The Clan of the Changs—The Story of a German and Chinese Family” (Benno Kroll, 2003). We exclude books that are classified in the main category “biographies.”

¹³ Table 2 indicates slight differences between hardcover and paperback genre classifications. These differences likely exist because the publishing house is not always the same for the two versions. Licenses are often sold to a different publisher, which may reclassify the book as another genre. In our dataset, we observe a publisher change in approximately 60% of cases and a change in genre in approximately 11% of all cases.

*GfK = Gesellschaft für Konsumforschung (market research company).

*GfK = Gesellschaft für Konsumforschung (market research company).

¹⁴Using four raters to measure the attractiveness of the cover and title is, of course, not representative. However, it is very difficult to obtain a representative sample of raters to rate 1,206 different covers. Thus, we rely on ratings given by four student raters following other studies that used, for example, two PhD candidates as raters (Eliashberg & Sawhney, 1994) or three students to classify movies into genre categories (Eliashberg, Hui, & Zhang, 2007). This approach reflects the classic trade-off between the complexity of content analysis and the need for representative samples.

¹⁵We used scales ranging from 1 (not at all/not attractive at all/poor) to 5 (very much/very attractive/very good). The “interest in title” measure is based on a single-item scale, whereas “cover” is measured with a multi-item scale across the three dimensions of attention, attractiveness, and fit.

¹⁶Our sample contains 11 participants who rated the celebrity status of 559 authors (raters were men and women, with a mean age of 38 [SD = 15]). Other studies also use small samples to control for variables when data is not available elsewhere (Eliashberg & Sawhney, 1994). For the movie industry, star power is often measured with regard to user behavior on the Internet (e.g., the Starmeter at Internet Movie Data Base; see Elberse, 2007), whereas we cannot rely on similar data for the book industry. We double-checked the data using Google™ and Wikipedia to learn about the authors' background and found high face validity for our method.

¹⁷Note that we focus on licensed books from abroad that enter the German market.

¹⁸Note that due to the fixed book price agreement in Germany, retailers are strictly limited in terms of (price) promotions. Therefore, variables such as in-store price promotions or special offers (e.g., “two-for-one”), which would be relevant in other countries, do not play an important role in the German book market.

¹⁹We note that Amazon reviews can be biased if the author and related peers post reviews (we thank a reviewer for bringing this issue to our attention). We analyzed the number of reviews and found that the hardcover (paperback) books in our sample received, on average, 6.6 (11.9) reviews, with a standard deviation of 10.8 (19.6). The maximum number of reviews is 126 (228).

²⁰We used the Breusch-Pagan test of independence (Breusch-Pagan, 1960) to test if the error terms of both equations are correlated. High significance confirms that application of SUR model is appropriate.

*p < .1.

**p < .05.

***p < .01.

²¹We also estimated a model in which we include hardcover sales as an independent variable in the paperback equation (see Appendix). The results reported in the Appendix show the indirect effect of hardcover sales on paperback sales, as well as the total effects on paperbacks (see also Albers & Clement, 2007).

²²Including hardcover sales in the paperback equation leads to R ² = .67 (generalized least squares) and .70 (ordinary least squares; see the Appendix).

²³Except the genre of “other”; here, we could not identify a significant difference because of too few observations.

²⁴Because the covers' attractiveness proved to be a significant variable, we decided to undertake some additional exploratory steps. We selected 50 of the books rated most attractive and 50 of the books rated least attractive and compared their covers side by side. The difference between the two groups is remarkable. The 50 most attractive are densely designed, with vibrant colors, whereas the 50 least attractive are sparsely designed, with a great deal of white space. It appears that this could be an area of promising additional research that could prove to be very important to the field of book cover design in the publishing industry.

²⁵We thank an anonymous reviewer for calling our attention to this issue.

²⁶We analyzed the diffusion patterns of the books and found that the release of the paperback usually occurred at the end of the diffusion process of the hardcover edition. Thus, we conclude a causal relation, such that an earlier release of the paperback should increase its sales.

*p < .1.

**p < .05.

***p < .01.