blog dds

2006.10.11

Web Page Hits, Amazon.com's Sales Rank, and Actual Sales

Over the past three years I've been collecting the amazon.com Sales Rank for my book Code Reading: The Open Source Perspective, and (lately) also for its sequel Code Quality: The Open Source Perspective. Yesterday I mapped the sales rank to actual sales, and correlated those with significant events and hits on the book's web page.

The book's hourly sales rank over the last three years isn't particularly enlightening.
Code Reading amazon.com sales rank
There is huge variation, and averaging the sales rank over longer periods is clearly nonsensical. To map the amazon.com sales rank to the book's weekly sales I used data from an interesting paper by Erik Brynjolfsson and his colleagues. In the paper, titled Consumer Surplus in the Digital Economy: Estimating the Value of Increased Product Variety at Online Booksellers, the authors outline how they mapped the amazon.com sales rank to actual sales by repeating an experiment by two other researchers. First, they obtained actual weekly sales data for a book with stable sales from a publisher. Then, within ten minutes they bought six copies of the book from amazon.com and noted the book's sales rank before and after. Using regression analysis they found that

log sales = 10.526 - 0.871 * log rank

Using that formula, I converted the hourly sales rank into an estimated weekly sales figure, which I could then meaningfully average over weekly periods. The resulting chart made a lot more sense.
Code Reading weekly sales
Surprisingly, when combined with amazon.com's market penetration the sales figures also matched the sales numbers I get from my publisher in their royalty statements. The big spike on March 2005, coincides with the book's review on slashdot. It was apparently the only event that made a significant impact on sales.

Finally, I examined the relationship between hits on the book's web page and the sales figures. Charting the weekly hits on the book's index page did not produce a result that matched the patterns of the sales.
Code Reading web page hits
However, when I plotted the two charts on top of each other, I could see that often the two numbers matched in step. As you can see, an additional 50-100 web page hits often appear together with 1-2 additional book sales.
Code Reading sales and web page hits
On the other hand, large increases in web page hits don't always translate to sales. I traced the large spike on February 2004 to a reader comment in a slashot story concerning kernel development. More than 1000 hits didn't results in any visible change in that week's sales.

Read and post comments, or share through   


Creative Commons License Last modified: Wednesday, October 11, 2006 10:20 am
Unless otherwise expressly stated, all original material on this page created by Diomidis Spinellis is licensed under a Creative Commons Attribution-Share Alike 3.0 Greece License.