Did Oprah steal book sales with her reading club?

There is a new paper (pdf) from the excellent Craig Garthwaite, here is the abstract:

This paper studies the economic effects of endorsements. In the publishing sector, endorsements from the Oprah Winfrey Book Club are found to be a business stealing form of advertising that raises title level sales without increasing the market size. The endorsements decrease aggregate adult fiction sales; likely as a result of the endorsed books being more difficult than those that otherwise would have been purchased.  Economically meaningful sales increases are also found for non-endorsed titles by endorsed authors.  These spillover demand estimates demonstrate a broad range of benefits from advertising for firms operating in a multiproduct brand setting.
For one thing, it’s really hard to get people to read more.


I disagree with the premise. A knowledgeable celebrity endorsement, in effect, provides information to the market. This reduces risks to the purchaser, and should increase overall demand. Furthermore, if the endorsements lead to a reading as a new hobby, it could increase overall book demand.

So, is the author arguing that there would be more sales of books if there were less information. That counterfactual doesn't make sense...people would view buying a book as risky. Risk would reduce overall sales.

Moreover, if there were not an Oprah, there would be a librarian or the web, or Amazon reviews.

Shucks, you started so well Bill.

If I knew my favorite was going to come out I'd go to that movie this year and no other.

You may disagree with the premise, but your reasons for doing so are the exact issues that are in dispute. And while your reasons are not necessarily logically flawed, this study is using data to show that they might be incorrect, at least when it comes to Oprah's celebrity endorsements. Further, the author is not arguing that less information would equate to more sales. Not at all. He is arguing that this particular form of information might lead to fewer sales.

In fact, it could be construed as arguing that Oprah's celebrity endorsement is a form of negative information: Oprah is directing buyers to more difficult reading experiences, lowering their enjoyment. In so far as endorsements should direct to positive experiences, this is not information, but misinformation, or in your terms information that is leading to more rather than less risk. If this is in fact the case, then it is even consistent with your "more information should equal more sales" (When in fact more information, for any number of reasons, does not have to do so. It could lead to the same number of better informed sales. It could lead to fewer sales due to knowledge that acceptable purchases are not available. It could lead to indecision due to the number of choices the additional information presents.)

I suspect it could be even simpler. An endorsement goes to book readers. It does not go to non-readers. It says more "read this book" that it says "read a book, this one is good." So, it is easy to imagine that it is zero-sum within the book market. So, how could additional information reduce sales? Easily. If I'm looking for a good book in the absence of external information, I might have to browse and buy several books. Additional information from reviews or zeitgeist focuses attention on "The One" choice. Elimination of a lot of the tripe can be good for everybody except the tripe industry- another activity isn't prosperity example. However, the reason I would ignore Oprah is that I suspect her endorsements are more signaling than service, thus the overly difficult reading exercises. I don't need someone to raise the bar on my pleasure reading. Mostly though I suspect Oprah is just capturing the rents by building a toll bridge on the bandwagon of zeitgeist phenomena.

Much agreed


If Oprah is leading her readers to books they don't like, she won't be followed. If they have book lust, there will be other authors. As to the point of more choices leading to fewer sales, you make an argument for endorsements, not against them.


Your assertion that Oprah will not be followed if people do not like the books does not have to be true. People may purchase them for the signalling value of owning and being a part of the Oprah community, of having the book on display on your coffee table. Or any number of other reasons.

On endorsement, you are correct, but I never made an argument against them. An endorsement could either provide information necessary to reduce the indecision, or it could provide an additional option that increases indecision. Neither must always be the case, such endorsements will act differently on different buyers and for different products. I used it merely as a counterpoint to your assertion that more information should lead to more sales, not less. That does not have to be true. And, in this case, the "less" sales seems to be true based on the data on display.

Here is what I see as the problem with your argument: You are providing premises as general rules that make sense, "More information should lead to less risk. Less risk should not lead to fewer sales." But these are not unbreakable, universal rules. It is possible for particular examples to break them, and this study, with data, not just the rhetoric we're trading here, shows that the Oprah example may be one such case where this not-quite-universal rule fails. Oprah's recommendations may increase risk from the reading-enjoyment sense, but not from the purchase-satisfaction sense. So people follow her recommendation because they are satisfied with their purchases when they do so, but the reading-enjoyment side is not as high, which in some form reduces their global book purchasing even if their local purchase habits (purchase of Oprah books) rises. Or some other related phenomenon.

Some form of this is what the study has data to support. It could be wrong, but given the lack of unbreakable laws when it comes to consumer behavior, I'll not be convinced by a counter argument that does not at least address the data, and instead relies on a priori assumptions about how things should work.

See comment below re data deficiency.

Andrew, Jim: The issue is growth of market, not division of market.

If you believe that the persons influenced by Oprah's endorsements were dedicated avid readers before watching her show, then perhaps; but that would require a belief that Oprah's listeners consisted of avid readers.

If you do any marketing analysis, you would understand the difference between shifting a constant demand, and increasing overall demand. I have real doubts that increasing information and reducing risks of purchase reduces demand.

Ever read the piece on used cars and lemons?


"I have real doubts that increasing information and reducing risks of purchase reduces demand."

You have said this a few times, and it is certainly not an untenable position, but you have not supported it with respect to the counter evidence on display in this study. Your quoted assertion may be true in most cases, while this is an example where it proves to not be a universal rule of the effect of information on product demand & market size.

Let me try another example: Negative advertising against a local cable company reveals information about the Cable Co that causes many people to switch away from that service provider. Most people choose an alternative. Some few people do not. The information caused a reduction in market size.

How is Oprah's recommendation negative advertising?

I disagree with you, Bill. A knowledgeable celebrity endorsement provides information to the celebrity's market, not the total market. You obviously can't say that for every celebrity, but, all things being equal, Oprah is a celebrity with a big following.

Equating Amazon reviews and Oprah's book club doesn't work because of this. There is a factor of 'image'. Oprah, which has a very different reputation than a company like Amazon, is telling her viewers which specific books to read. Amazon, on the other hand, is telling all readers (which includes more than a segment like 'Oprah's readers'), which books are on sale, which ones have great reviews, which ones relate to a book they are viewing, etc.

In essence, what I'm saying is you are oversimplifying the two differences by saying Amazon, the web, and Oprah are all endorsements. Amazon is doing what they make money to do, as is the web. Oprah's show is not primarily about books, which means her investing part of her personal show-time in recommending books as a known public figure is defined as an endorsement. To further strengthen my first sentence, I'll quote the essay itself: "Half of all consumers in the survey reported that they only notice an endorsement if it is for a product category in which they are already interested, suggesting that endorsements lead to business stealing for many consumers (MEC Global, 2009)." That number, 50% is a large one. Essentially, out of a group of ten viewers, five admit they won't even notice your endorsement unless it's in a market they are already a part of. In advertising terms, that is half your impressions wasted. Now with only five people as opposed to ten who are at some level more likely to notice your endorsement, you have less people and therefore less of a probability to make a sale. You better hope those ten people are paying attention and like your endorser. That is why endorsements do not affect market size versus other venues of advertising.

Err excuse me, those five***

Also, I meant to say a market they are interested in, not "a part of."

Brooklyn, You might want to read Akerlof,s article on the markets for lemons.

Maybe Oprah was just trying to get people to read better books, increasing the consumers utility from reading.

If your definition of "better books" is "awful books," sure.

If books from Toni Morrison, Tolstoy, Garcia Marquez, Charles Dickens and William Faulkner are awful, you're right

" Economically meaningful sales increases "

What does that mean? Is it another ambiguous offshoot of "statistically significant"?

If I have $100,000 annual sales what is an "economically meaningful" increase? $100? $10,000?

I don't care two cents worth about the whole tempest-in-a-teapot thing, but I am offended by the word "stealing."

We could just as soon say "A recommendation from Tyler Cowen on Marginal Revolution is found to be a business stealing form of advertising." Doesn't sound so good that way, does it?

Well, in fairness to Tyler, it's the paper's use of the word he is quoting, not his own interpretation of the study.

Right, I did not mean my comment as a slur on Tyler. Rather, I was pointing out, tongue in cheek, that Tyler often recommends books. I enjoy reading his recommendations. Sometimes I take them. Because of that, I will have a measurable tendency (0%<X<100%) to read those, and thus a corresponding measurable tendency to avoid other competing books. But I would never characterize his recommendations as "stealing" anything from anybody. It is that terminology by Craig Garthwaite that offends me.

And yet true. I am proud of blunting the demand for lesser books.

Oh, but Andrew would say you are reducing the demand for all books, not lesser books alone, dear Oprah.

This makes a lot of sense. First, you have to consider Oprah's audience. They're probably not in the habit of reading certain types of literature, so the switch from Jennifer Weiner to Joyce Carol Oates could slow them down (which is not a judgment that one is better than the other. Personally, I'm not a big Oates fan.) So the rate of book consumption is lower, meaning the sales of books would be lower. The argument is favor of Oprah was that she got people reading, but all she really may have done is cause readers to switch their reading material.

Abstracts without effect sizes are annoying. How much did sales fall by? 0.1%? How accurately can they measure sales of books in the US?

I guess that's in that pdf somewhere!

go to page 27 & 28, exceprt:

"In the week following a Club endorsement, there was a statistically insignificant change of approximately 1.2 percent in the sales of adult books. In the eight weeks following an endorsement, there was a statistically significant 2.1 percent decrease in weekly adult fiction sales (p-value < 0.10). In the 12 weeks following an endorsement, weekly adult fiction book sales decreased by a statistically significant 2.5 percent (p-value <0.05). All of the estimates show greater sales decreases, suggestingthat a Club endorsement had a business stealing effect."

Why should anyone want to read more fiction? It's not like people are against the act of reading words, they spend plenty of time on the internet. Books get boring and redundant after about 10 pages, sitcoms stay funny the whole hour. "Educated people" spend all their time breathlessly pleading for people to read more. Might as well demand they still use a rotary phone.

I happen to be a bookscan power user, as I'm an acquisitions editor and have a strong economic incentive to understand the data. I also have an undergrad in Econ, this is my first post here.

The author has overlooked a MAJOR limitation in the dataset studied, bookscan does not include sales from Wal-Mart, Sam's Club, and most grocery stores, Let me suggest an alternative hypothesis, book sales are not reduced but rather are being pulled from retail locations that report to booskcan into those that do not.

I think it is likely that suggestible consumers who are sensitive to price and/or simple availability bought the suggested titles at Wal-mart or while grocery shopping. Wal-mart would have large quantities of the suggested books on hand, and discounted more heavily than almost all other booksellers. Grocery stores would carry the book when otherwise they would not. Because of the merchandising the suggested books saw, sales were pulled to retailers outside of bookscan's reporting network. Hence for the dataset the sales within bookscan were reduced, but overall book sales were not reduced.

I think this alternative hypothesis aligns more closely with known book industry dynamics and would be deserving of investigation.

+1 John. Stand up and take a bow.

John, thanks for your comments. I really appreciate them because this is an issue I thought a lot about when working on this project, and that I spoke to many people in the industry about. You make a valid point about the data that I clearly acknowledge in the paper. Bookscan is a panel of sellers representing the majority of booksellers in the United States, but they do not include certain retail locations. I think the lack of data from some of these sellers we face studying this market and many others.

However, I think that the alternate hypothesis you are proposing would have to be a bit more complicated than you currently state. Certainly, there could be increases in the sales only at non-Bookscan outlets that would not be picked up in the data (Note, however that Bookscan is estimated to represent roughly 75% of the market).

However, when we consider whether this hypothesis is driving the results in the paper, I think it is important to then examine the first half of the paper which analyzes the title-level and genre-level sales response to Winfrey's endorsements within the Bookscan data. In this section I find very large sales responses at the title-level, and I find sales declines in the romance, mystery, and action categories. I then demonstrate that these genre's are more popular in areas where Winfrey and the Club are more popular.

Therefore, for your alternate hypothesis to be driving the results, it would have to be that there is an entire population that is not reading, responds to Winfrey's announcement, and does so exclusively at these non-Bookscan outlets. Furthermore, there would need to be a shift in the purchasing behavior of regular participants out of Bookscan outlets that occurs several weeks after the endorsement and also varies by the difficulty of the endorsed title. Finally, it would have to be that this is also correlated with the geographic popularity of both Winfrey and her book Club. This is a fairly complicated alternate hypothesis.

In total, I want to be clear that I find Winfrey had a huge impact on the publishing industry. She caused readers to increase the difficulty of the books they read, and she may very well have changed the very types of books that were ultimately published. I just don't find compelling evidence that she caused marked increases in aggregate sales. Suggesting that, as Tyler says, its hard to get people to read more.

Craig, I have a suggestion, if I may? It would be nice if the Abstract or at least the Conclusion section had some mention of what the effect-size was. The first question that came to my mind was how much was the title level increase in sales and how much was the aggregate decrease.

I admit I have only scanned the paper, but I still haven't found the answer. Especially in relation to Johns point, how accurately do you think you are able to estimate sales figures and how does the increase / decrease in Oprah-influenced-sales compare.

page 27 & 28, come on!!! you can Ctrl + F on PDFs =)


The conclusion that Opera may not have increased aggregate demand for books, is a conclusion that I think is reasonable. However, I strongly disagree with “The endorsements decrease aggregate adult fiction sales; likely as a result of the endorsed books being more difficult than those that otherwise would have been purchased”

If you believe
1. readers are price sensitive
2. within a genre readers will more freely substitute one title for another, and
3. Bookscan reporting and non reporting locations are heterogeneous
The same results occur, without concluding that opera turned people off from reading. I find it far more believable people will save money by shopping at Wal-Mart and want to read romance, mystery’s, and thrillers that are “good” vs. being “turned off reading”

The locations that do not report data to bookscan are all very different types of book retailers, selling books is not their primary business. The non represented Wal-Mart, "club stores", and grocery stores all tend to sell a given book at a much higher discount than the traditional bookstores.The non bookscan reporting retails also happen to a customer set that tilts very heavily towards romance, mystery, and action titles. If you go look at the books on sale at Wal-Mart you’d clearly see that what they sell is highly focused in those three categories.

If you are not measuring all channels, you have a bias problem.

Furthermore, I would guess that Oprah books are more likely to be carried in the channels (Wal-Mart, etc.) that are not covered by the survey, so you have both a difficulty in measuring substitution, or net changes in readership.

One of Oprah's choice was Tolstoy's 800-page Anna Karenina. It seems reasonable that reading this took a long time on average, discouraging purchases of shorter, easier modern books for some time.

But it likely also moved a lot of incremental copies of Tolstoy at unscanned Wal-Marts.

Maybe it's just my prejudice, but I have such a hard time imagining an Oprah watching audience reading Tolstoy.

Kevin Drum quotes from the original paper to the same conclusion as Sailer:

"Following an endorsement, the sales of classics rose by 3.5 percent []. In contrast, there were statistically significant decreases for mysteries and action/adventure novels. Romances also saw a sales decline....

Club selections were longer and more difficult than the bestselling titles in the genres that were popular among consumers likely to respond to the endorsement. Assuming that longer and more difficult books will take more time to read, the difference in estimated grade level combined with the genre-level sales shifts help explain the pattern of aggregate sales declines in the main results....Taken together, these estimates suggest that the difficulty of the endorsed titles contributes to the aggregate sales decline."

Tyler's lead makes this all sound like a bad news for Oprah but more people reading Garbiel Garcia Marquez and Tolstoy and less reading romance and adventure novels? Sounds like she's earned two gold stars.

How can you say anything if you do not measure all channels.

Note also you are equating sales to time spent reading.

What are the sales figures for books you can get for free at the library or free (like tolstoy) where the copyright expired and you load the free epub to an e reader.

If you are measuring sales, and the universe includes books that are not sold or are free, you have a data problem.

By the way, one measure of the effectiveness of endorsements, measured in the library channel, is to look at the number of library books of a given title, and the number of holds following endorsement.

Most other library data is confidential.

Another one I've noticed lately is used bookstores. I wondered why there were so many copes of Bret Lott's Jewel around, but none of his other books. Then I found out Jewel was an Oprah book.

CG says

"Therefore, for your alternate hypothesis to be driving the results, it would have to be that there is an entire population that is not reading, responds to Winfrey’s announcement, and does so exclusively at these non-Bookscan outlets."

But surely this is far more likely for the group that buys books at Walmart or Costco? It's quite reasonable to suppose that reading for them substitutes for other non-reading activities (like buying a DVD or a video game) and that gains in book buying (e.g. Anna Karenina or Joyce Carol Oates) more than offset the lost sales to non-Oprah titles.

Or perhaps for customers who buy at wal-mart and costco see books as goods that can be more easily substituted?

"I just don’t find compelling evidence that she caused marked increases in aggregate sales."

If the objective of some people is to get more people into reading, celebrity endorsement it not the way. Celebrity endorsements may make a lot of people feel really good, SPWL style, but with no measurable results Thanks a lot sir.

Ps. when I was a horny teenager I got caught by Henry Miller and Jim Ballard, since the I've read lots of more diverse material. Maybe reading is not boring, teacher's (education status quo) recommendations are boring. Take it with a grain of salt, this is only 1 measurement dataset.

People who are already readers simply don't have much time left.
I already buy more books than I can possibly read.

Andrew, You assume what you claim to prove with your statement that Oprah's statement only reaches and influences existing readers, and does not reach and convert non readers.

What evidence?

very very good ı like nice web site

Comments for this post are closed