Having working with the raw data, it was pretty fantastic. Segemented by industry/business size, handled issues with multiple programming languages or companies where one section used one language and another used other ones, etc. We even knew which tools and add-ons were used for which languages and which compiler on each platform (i.e. how many commercial shops using C++ targeting linux are using gcc vs. icc?).
But that data was also stunningly expensive. My marketing friends tell me that accurate market data always is.
1) I use more metrics and
2) I don't go around making statements about how, from one month to the next, someone displaced someone else, or climbed into a certain ranking, or things like that.
3) I let people reweight the chart based on the metrics they like.
I think the numbers LangPop comes up with are pretty good, but by their very nature are a bit fuzzy.
Also, maybe add GitHub and StackOverflow as data sources?
GitHub and StackOverflow started out really biased in terms of their communities - GitHub with Ruby, and StackOverflow with Microsoft languages. Do you think they've sufficiently lost that bias?
I'm not sure I even trust that... I do lots of bindings from OCaml to C, and whereas I consider them to be OCaml projects, GitHub sees they're more C by LOC and counts them as C.
The confidence of the SuperCollider programming language has been set to 80%. The reason for this is quite funny. In order to prove that the TIOBE index can be manipulated easily, Adam Kennedy created an empty Perl library called Acme-SuperCollider-Programming. This was to boost the unknown programming language SuperCollider by adding it to Perl's popular library archive CPAN. Now 20% of all +"SuperCollider programming" come from this artificial library.
Yes, quite funny indeed, but not for the reasons they think. Just to show that Tiobe methodology is a joke. I don't understand how people take this seriously.