For a few dollars more (II)

The “gold standard” for establishing causation is a randomized experiment in which the researcher can vary the independent variable randomly and can measure the resulting changes in the dependent variable. For example, a publisher could randomly divide its titles into different groups, and then delay the e-book release of some groups by one week, some by two weeks, some by three, and so on.

Unfortunately, these sorts of randomized experiments are extremely difficult to engineer for a host of reasons, among them the unsurprising fact that authors and agents object to having the work on which their livelihood depends become the subject of an experiment that may well cause their sales to drop. Indeed, we tried for several months to work with the publishing house mentioned at the beginning of this series of articles to design a randomized experiment, but in the end we couldn’t overcome concerns from the publisher’s authors and agents about how such an experiment might affect their sales.

If a randomized experiment isn’t feasible, the next best option is a naturally occurring event that stimulates the characteristics of the randomized experiment. And in 2010 just such an event happened. The publishing house we had been working with to design the experiment got into a pricing dispute with Amazon that culminated on April 1, when the publisher removed all its books from Amazon’s Kindle store. Amazon was still able to sell the publisher’s hardcover titles, just not the Kindle titles.

Amazon and the publisher settled their differences fairly quickly, and on June 1 the publisher restored its e-books to Amazon’s market and returned to its previous strategy of releasing its hardcovers and e-books simultaneously. Table 3.1 summarizes how the publisher’s e-book releases were delayed during the dispute. A quick glance shows that the resulting e-book delays are close to the delays that might occur in a randomized experiment. Books that were released in hardcover format in the first week of the dispute (April 4) were delayed on Kindle format by eight weeks (June 1) as a result of the dispute.

Likewise, books that were released in hardcover format the week of April 11 were delayed in Kindle format by seven weeks after the hardcover release, and so on for books released on April 18 (six weeks), April 25 (five weeks), all the way to books released the week of May 23 (one week). More important, neither the timing of the event nor the release schedule during the event was driven by the expected popularity of titles, and thus sales of undelayed books should provide a reliable measure of what sales of the delayed books would have been had they not been delayed. In this case, all we had to do to test how delaying e-books affected sales was to compare the sales of delayed books (those released during the dispute) to the sales of undelayed titles (those released shortly before and after the dispute).

But before discussing what we learned from analyzing the data, let’s examine the economic rationale for why publishers have separate hardcover and paperback releases in the first place. Why make paperback customers wait nearly a year for their book? Why not release the hardcover and paperback versions at the same time? Why have two different versions at all?