Why would there be error bars or confidence levels on this graph? It's actual reported sales data, not an estimate.
Reality is of course more complex than book sales alone would indicate. But just as book sales of Perl in 1996 told me that people were under-valuing the language -- it was the number one computer book in revenue of 1996, but the language got no coverage in the computer trade press -- led me to launch my Perl conference and other Perl activism, and eventually my open source activism in general, so now, the same data set tells me that Perl is in something of a decline.
As I said in the article, the story is far from over. And in fact, you can see the reversal in the Java slope that happened with the release of Java 1.5. So a new release will bring new life to the book market.
The data is the data. Make of it what you will.