robster970
Guest Author
- Messages
- 4,567
- Likes
- 1,390
Re: W/E 11th May - Results
I too have opinions about the scoring mechanisms but I did model your suggestion just out of curiosity on the weekly score sheet, some time ago.
https://docs.google.com/spreadsheet/ccc?key=0AnFF2Rblu36wdGRNQjNXTzVjdldJb2dXdjRVTGs5VVE#gid=2
You will see a tab called group stats and your suggestion of the cumulative (also taking into consideration whether people are + or - too) demonstrates that tar, who participates infrequently would win. My problem with this is that regular participants are penalised because their increased sample sizes demonstrate their long run forecasting capability. I am not sla@gging off tar here btw but he does not participate that frequently and therefore I believe his win is a statistical aberration rather than relating to a long running statement of performance.
So before trying to improve the scoring system, perhaps we should question what the purpose of the weekly comp is. Is it:
a) Demonstrate accuracy on a predictive basis for the week, alone (in which case the cumulative suggestion would be a good scoring system)
b) Represent a longer run forecasting capability over say a quarter or year (in which case the current or Pat's proposal would better suit).
c) Something else.
Food for thought lads.
However, one suggestion I'd like to make for the future which could be added in to competition is to do a distance based scoring system. So the idea would be that the winner of the quarter would have the lowest cumulative amount of points at the end of the quarter.
I too have opinions about the scoring mechanisms but I did model your suggestion just out of curiosity on the weekly score sheet, some time ago.
https://docs.google.com/spreadsheet/ccc?key=0AnFF2Rblu36wdGRNQjNXTzVjdldJb2dXdjRVTGs5VVE#gid=2
You will see a tab called group stats and your suggestion of the cumulative (also taking into consideration whether people are + or - too) demonstrates that tar, who participates infrequently would win. My problem with this is that regular participants are penalised because their increased sample sizes demonstrate their long run forecasting capability. I am not sla@gging off tar here btw but he does not participate that frequently and therefore I believe his win is a statistical aberration rather than relating to a long running statement of performance.
So before trying to improve the scoring system, perhaps we should question what the purpose of the weekly comp is. Is it:
a) Demonstrate accuracy on a predictive basis for the week, alone (in which case the cumulative suggestion would be a good scoring system)
b) Represent a longer run forecasting capability over say a quarter or year (in which case the current or Pat's proposal would better suit).
c) Something else.
Food for thought lads.