I’ve decided that you are probably one of those people who show their appreciation of a free gift by taking mild potshots at the giver. So, thank you very much for your kind words.

The reason I changed to Python was simple. I am not familiar enough with Perl; I am familiar with Python. This is not an attack on Perl, but merely a quirk of history that I once needed to learn Python, and to date, I have not needed to learn Perl.

There are several reasons that I didn’t publish the code. Each individual reason isn’t strong, but together they made a case.

1) It is not my default position to publish code, unless there is a reason. It is my default position to not publish code, unless there is a reason. I see no reason here.

2) The code isn’t what this article is about. The code isn’t interesting. The conclusion about Stack Overflow demographics is what it is about. The rest (including the offer to make the code available) is for scientific credibility about the conclusion, not because all software should be free.

3) I am not willing to offer support for the software. If I made it available, I was likely to get support calls.

4) Yes, the code is bad from a code smell perspective. I was in a rush, and I didn’t spend the time to make it clean, readable, reliable, maintainable, efficient, (etc.) I didn’t even spend the time to package it up. I don’t want to associate my name to the source-code. The code was excellent from the perspective of one user, in a rush, for one-time use.

5) I felt a little uncomfortable about scraping someone-else’s site without permission rather than using the API (which they said was coming, but hadn’t published at the time). If I made it freely available, I would be encouraging others to do the wrong thing too.

6) I didn’t think there would be any real demand. (In fact, I have had a total of two requests, and one person wrote back to say it wasn’t what they were looking for.)

Scatterplots are great for sparse data-sets.

However, this one is very clumpy, with some co=ordinates on the scatterplot having over a hundred data-points. (e.g. there are 137 26-year-old users with a reputation of 1)

If there was an occasional 2 or 3 duplicates, you can put the number of duplicates at the co-ordinate instead of the cross, but that loses impact when the numbers get higher.

I have sometimes used the technique of adding a teensy amount of randomness to get illustrative clusters of crosses around the real data-point, but that stops making sense around 10 duplicates.

I don’t have any other techniques for representing this data meaningfully on a scatterplot. Let me know if you have any ideas.

Re: Colour Scheme

Yes, I agree, but I knew if I didn’t get it out by yesterday morning, I wouldn’t get a chance to complete it for several days, so I didn’t finesse the charts as much as I would like.

As a member of the Epoch Generation, of course I believe that age has a tendency towards surprisingly high intelligence, knowledge and communication skills which results in a higher reputation score. However, I strongly suspect that this is just an artifact within the normal bounds of the variability of the data.