Results of the Forecast Experiment (October 2, 2003)
The results...
--posted by TangoTiger at 04:04 PM EDT
Posted 7:57 a.m.,
October 3, 2003
(#1) -
Anonymous
.
Posted 10:02 a.m.,
October 3, 2003
(#2) -
Greg Tamer(e-mail)
Conclusion? Trust the numbers you see (which is enough for hitters), and fill-in the information missing (which is the case with pitchers, be it health or mechanics). Any extra nuance that you find just doesn't have the impact you'd hope. None of the 3 groups dominated the others. This was about as close to a draw as you'd expect.
But what if we did this again next year and the results are different? In other words, how conclusive is this?
Posted 11:08 a.m.,
October 3, 2003
(#3) -
Alan Jordan
I did paired t-tests and none of them were significant because of the small sample size. A couple marginal around p<.15 I think, which hints that with a larger sample some might have been significant.
I would suggest a larger sample of hitters and pitchers next year, say 40 each.
Of course even if you only do 30 total next year, the results can be combined for a larger sample size.
Posted 11:57 a.m.,
October 3, 2003
(#4) -
tangotiger
(homepage)
A similar experiment was done a few years ago at the above link. In this case, rather than having 165 people picking on 32 select players, there were about 30 people picking on 125 or so players.
There were 3 good reasons I did not do as that study did:
1 - No interest in duplicating the work
2 - Much easier to get 100+ people to participate if I ask them to do less work
3 - By selecting the 32 players who were most inconsistent, it's here we'll see the variations in picks. The other say 100 players that I would have added would have left little variation. If you've got a guy with a 4.20, 3.98, 4.12 ERA, there'll be very little difference in what any group will pick. So, you can increase the sample size, but little variance will be added.
Alan, can you send me an email?