Random Effects of Ratings on eBook Sales. What's luck got to do with it?

Well... the answer may well be: a lot! This fact emerged from experiments carried out by sociologists Duncan J. Watts, Mathew Sagalnik and Peter Dodds which Watts described in his 2011 book, “Everything is Obvious* (*Once You Know the Answer)”. Their work focuses on online markets. This post of mine is based on Robert H. Frank’s article on this subject that appeared in The New York Times of August 5, 2012.

Franks describes the researchers experiments thus: They invited subjects to a temporary web site called Music Lab.  The site listed 48 recordings by little-known indie bands. The subjects could download any of the songs free if they agreed to give a quality rating after listening. The average of these ratings for each song then became a proxy for the song’s quality in the rest of the experiment. Importantly, since each subject saw no information other than the name of the bands and the songs, their ratings of each were completely independent of reactions of other subjects.

What does this have to do with marketing your book on say---Amazon?
Think of song ratings in this experiment as the rating that readers give to your book newly added to Amazon.

The result of the first phase of the experiment was that the independent ratings were extremely variable. Some songs got mostly high marks or mostly low marks, but a substantially larger number received distinctly mixed reviews.

The second phase of the experiment differed from the first phase in that two new pieces of information were added to each band and song title: 1) how many times each song had been downloaded by others, and 2) the average rating it had received so far. Eight separate sessions followed this protocol.

This phase is exactly what happens when readers consider whether or not to buy a book on Amazon based on early reviewer ratings.

The result of this second phase was that this social feedback caused a sharply higher inequality in song ratings and download frequencies. The most popular songs were a lot more popular, and the least popular songs were a lot less popular than the same songs rated in Phase 1. Also, each of the eight sessions which involved discrete subsets of subjects, displayed enormous variability in popularity rankings. An example cited in Frank’s article highlighted the results for a song called “Lockdown” by the band 52 Metro. “Lockdown” was ranked 26th out of 48 songs by the subjects who rated songs in the first phase without any information other than the name of the song and the band. But in the second phase which added the ratings given by other subjects in the group, “Lockdown” achieved a ranking of #1 in one of the eight groups, and #40 in another!

This is the crux of the issue when a potential buyer considers the ratings given by earlier reviewers and sees nothing but 3's or less for example. What are the chances that s/he will buy?

So, if “Lockdown” (in the middle of the objective ranks of quality produced in Phase 1) happened to experience the dynamics of the group that ranked it #1 with the social feedback information added it would have been miraculously transformed into the best song of the lot.  If it experienced the dynamics of the group that ranked it #40, it would be the worst.  Presumedly, sales is much affected by perceived quality. What dynamics, you may say? Well, the dynamics of the sequencing of the ratings, that is, do the early ratings reflect favorably, unfavorably or neither, for instance. If the sequence of ratings is completely random, a big if, it means that luck is everything.

So, if early reviewers are ecstatic about your book, you have a much better chance of making the sale when subsequent potential buyers consider it based on these high rankings. Nothing new about this.

As Frank points out, “the most striking finding was that if a few early listeners disliked the song, that usually spelled its doom.  But if a few early listeners happened to like the same song, it often went on to succeed. In their experiments, the sociologist researchers showed how feedback could be a vitally important random effect. Any random differences in the early feedback we receive tend to be amplified as we share our reactions with others. Early success —even if unearned — breeds further success, and early failure breeds further failure. The upshot is that the fate of products in general– but especially of those in the intermediate-quality range – often entails an enormous element of luck.  Chance elements in the information flows that promote success are sometimes the most important random factors of all.”

The lesson here would seem to be that independent of how good your book is, it's fate in the Amazon market place is determined largely by lots of early ratings that just might be extraordinarily bad for no other reason than the fact that of all the ratings your book will receive, the early bad ones begot more bad ones. Of course, the opposite could happen too; good early ratings that inflate the perceived quality of your book.





.