Predicted Cut Line for Suscriber's Contest (1 Viewer)

[Dinsy Ejotuz]

Played with this a bit this morning and found that the only two things you need to predict the cut lines for the subscriber's contest are the percentage of entries being cut and the percentage of original entries still in the contest at the start of the week.

The number of teams on a bye each week and average score of the games played each week don't add anything. In fact, both make the predictions less accurate.

I don't know how to code this, so apologies for the bad formatting, but here's what the model spit out for Weeks 2-9. The first column is the predicted cut line, 2nd is the actual cut line and the last is the difference:

Week 2: 124.8........123.7........1.1

Week 3: 124.4........125.3........-0.9

Week 4: 104.3........106.8........-2.5

Week 5: 104.4........102.0........2.4

Week 6: 105.7........109.5........-3.8

Week 7: 109.2........101.2........8.0

Week 8: 116.3........121.5........-5.2

Week 9: 130.7........129.8........0.9

In Weeks 2-9 the model was within three points five times, and off by more than five points only once (Week 7, 8 points). So it appears to work. If it is right, the cut lines for Weeks 10-12 will look something like this:

Week 10: 104.9

Week 11: 120.4

Week 12: 155.6

I make no promises about the predictions, but hopefully it's better than guesswork. For the statistically inclined, the Adjusted R-squared for the model is .814.

 

[Bob N. Weaver]

This is very interesting to me and it looks like a fair amount of work/time went into it. Nice work.

If we compare Week 8 to Week 10, both weeks reduce the partipants by 25%.

I would think that the average score during a non-bye week (Week 10) would be higher than a bye week (Week 8), since the pool of NFL players would be larger which one can draw a non-zero score from. That being said, the predicted cut-off for Week 10 (104.9) is substantially less than the actual cutoff for Week 8 (121.5). Can you explain this to me. Thanks.

 

[Dinsy Ejotuz]

This is very interesting to me and it looks like a fair amount of work/time went into it. Nice work.If we compare Week 8 to Week 10, both weeks reduce the partipants by 25%.I would think that the average score during a non-bye week (Week 10) would be higher than a bye week (Week 8), since the pool of NFL players would be larger which one can draw a non-zero score from. That being said, the predicted cut-off for Week 10 (104.9) is substantially less than the actual cutoff for Week 8 (121.5). Can you explain this to me. Thanks.

I tried every combination of variables using:-- # of Teams on Bye Weeks-- Average Score for Games in a Week-- Percent of remaining teams being cut-- Percent of initial entries alive at the start of the weekBye weeks and average score weren't helpful -- I'm not sure why. I was surprised though.But according to the prediction, the fact that there are half as many people alive in Week 10 as Week 8 will lower the cut line score. I'm not sure, logically, why that's the case, but I think it's possible that as the competition has gone on every team is more likely to have significant overlap with the other remaining teams due to survivor bias. In other words, a higher percentage of surviving teams have Javon Walker this week than did prior to his demolition of the Steelers on Sunday. If you had him on your roster it was pretty hard to get cut in Week 9.And since there are fewer and fewer players that actually matter to a large number of contestents the liklihood that a bunch of them score in the same week goes down. If Frank Gore, Ben Watson and Willie Parker have bad weeks now it will lower the average score overall since so many teams have those guys. Previously teams with those guys might do badly, but there were plenty of other widely-held players to bump the average the other direction.On the other hand, it's always possible that the model is drek. It does happen sometimes.

 

[Bob N. Weaver]

Thanks for the response. I didn't consider the player ownership overlap. I hope the predicted cutoffs are accurate. It probably gives me some level of comfort (probably false) heading into Week 10.

Do you think another factor which may raise the cutoff going forward would be the overall scoring potential of the NFL players still owned versus the non-producers (e.g., Lamont Jordan, Randy Moss, etc) which are being weeded out?

 

[sinatravolta]

This is cool that you've come up with this but I don't see how the week 10 number can be correct.

I don't understand why player overlap should have an impact. Maybe on the standard deviation but not on the expected cut number.

I think it will be around 130 again. There are fewer people being cut than last week, but there are no byes to deal with and with every week that goes by, presumably the quality and consistency of the teams remaining is getting slightly better. There weren't any major injuries I know of to frequently owned players. There's really no way I can see it going below 120 unless it's just a freakishly low-scoring week or all the commonly held players like Watson, Gore, Parker, Glenn, etc. have bad weeks which is unlikely.