First off, the packets ranged in difficulty from over 600 points per 20 TUs (Virginia Tech--which was why I decided to use it as the "leadoff" packet) to under 250, but about 2/3 of them were within a standard deviation of the mean. The most interesting result, though, was that I tended to *consistently* underestimate the difficulty of the packets. Obviously, it goes without saying that I will do better, on the whole, in certain subjects, and will do worse in others. I did not expect, however, the net difference to be positive. Moreover, I tended to consistently underestimate difficulty by **80-100 points/20 TUs:** while my estimated scores averaged in the low 400s, the actual average scores were in the low 300s. [The average score I had in mind was somewhere around 380-420 points/20 TUs. This was predicated on a TU conversion rate of ~85%; the actual TU conversion rate at PB10 was ~72%, which I find disappointing.] In my mind, this does *not* represent a flaw in the system, but rather a discrepancy between my knowledge base and that of the circuit as a whole: it means, among other things, that I did not do quite as good a job as I would have liked of keeping the difficulty at a reasonable level. I tried to push all questions towards a common medium: I tried to make easy questions harder, and hard questions easier. For Penn Bowl 11, I will still do that to some extent, but I will focus more on reducing the high end of the difficulty spectrum than on raising the low end. --STI
This archive was generated by hypermail 2.4.0: Sat 12 Feb 2022 12:30:44 AM EST EST