Its very nice to be top, but also somewhat disappointing as I expected Kinoshita's RD/pts ratio to be a bit higher, although he has still brought in a decent amount of points. If Hatanaka keeps overperforming, maybe I wont have to worry about that though.
Pretty awful season start in TDU, but I'm not that concerned. Of course it really does suck that we scored less than last season with a much better line-up, but it can happen if PCM decides to be a bitch.
Eritrea's next and I'm very positive Roglic will deliver more on his own than our entire team did in TDU. Thanks for the update roturn.
Apparently Matt is interested in this, too, so I will update the current correlation of the different predictions with the ranking updates throughout the season. The metric I will use for this is correlation coefficient of [Predicted Ranking Positions] and [Points in Ranking Update].
Spoiler
I use the points instead of the ranking positions for the updates because i think it is more fair to rate predictions. Close ranking battles are rated more fair than e.g. the pure difference of positions.
Example:
2016 PT Final Ranking
1
[img]pcmdaily.com/files/Micros16/tif.png[/img]
Team I-Gen - Festina
8315
2
[img]pcmdaily.com/files/Micros16/bec.png[/img]
Becherovka - Petrof
6666
3
[img]pcmdaily.com/files/Micros16/vsa.png[/img]
Vesuvio - Accumalux
6641
1. Prediction:
1
[img]pcmdaily.com/files/Micros16/tif.png[/img]
Team I-Gen - Festina
2
[img]pcmdaily.com/files/Micros16/vsa.png[/img]
Vesuvio - Accumalux
3
[img]pcmdaily.com/files/Micros16/bec.png[/img]
Becherovka - Petrof
2. Prediction:
1
[img]pcmdaily.com/files/Micros16/bec.png[/img]
Becherovka - Petrof
2
[img]pcmdaily.com/files/Micros16/tif.png[/img]
Team I-Gen - Festina
3
[img]pcmdaily.com/files/Micros16/vsa.png[/img]
Vesuvio - Accumalux
Both predictions have just swapped two teams with each other from the final ranking but Imo the 2nd prediction should be rated much better than the 1st because the difference of Vesuvio and Becherovka turned out much smaller than the difference between I-Gen and Vesuvio.
A value of 1 would be a perfect prediction, a value of 0 is what you'd expect with a complete random prediction and the highest positive value indicates the most accurate prediction.
I know this is way too early to take anything out of this but anyway here we go:
Spoiler
I've included a second "main prediction" for users that posted two different predictions if different methods were used (e. g. one formulaic and one subjective) but if you prefer just one per manager then I'm fully okay for it (Knockout PCT Prediction would be my prediction in that case).
#1*
Prediction
1.
0,56011
Knockout Round Table Expert Prediction
2.
0,55959
matt17br
3.
0,54557
RoundTable Statistical Outcome
4.
0,52862
Knockout PCT Prediction
5.
0,49576
Tastasol
6.
0,47868
Jandal Round Table Expert Prediction
7.
0,47718
Croatia Round Table Expert Prediction
8.
0,38942
SN's Slightly Mathematical PCT Preview
9.
0,34703
Marcovdw's OVR Final Ranking
10.
0,30615
Bushwackers
11.
0,28555
Abishek Round Table Expert Prediction
12.
0,27890
Croatia's QCA prediction
13.
0,22185
Atlantius
14.
0,21715
Abisheks PCT Rankings - Co-efficient Based
15.
0,09125
Tamijo's Mathestimat
*PCT Ranking Update #1 - correlation with points
One of the key differences here right now is where teams have Grieg ranked. Those who have them as a top 3 team are near the top of the prediction while those who have them 11.-18. are further down. Also having Eritel, UBS and Newton ranked low or Meiji and Isostar near the top 5 helps a lot.
I also collected a few minor predictions to compare:
Spoiler
The names might be slightly confusing so a short explanation to that:
- The main predictions are highlighted in the darker blue.
- "+" Indicates a minor published ranking that was presented in combination with another prediction. e.g. Part of the calculation process or using a different weighting of factors.
- "#" indicates that i did not use the predicted rank but a different score like "expected points (e.g. Abisheks History Based one) " / "artificial Team-OVR " (e.g. Marcovdw) or "Average Finishing Position" (Knockout's Prediction). In theory these should be more accurate than the "normal" predictions.
- ° indicates a ranking that was not actually predicted by anyone but uses other things to predict the rankings:
°2016 Rankings = Taking the finishing order of last season's ranking as prediction -> 1. Newton ... 26. Haute Route
°Longest Existing Team = The theory is that longer existing teams have an advantage over newer teams because they could let their team grow over multiple years so I i thought it would be interesting to see whether that would have a chance against any proper predictions: 1. Orange, 2. Eritel, ...
°Median/Average Prediction: Used for these are all Main Predictions
If i forgot any prediction or you want to see any other metric (average age, number of letters in username, number of owned riders in the team ...) tracked, tell me
I really hope it's a good thing and not a bad one that my predictions have been some of the most spot-on so far! Definitely would have tweaked a few things looking at my prediction now, but I still mostly agree with my final rankings despite the first month of races.
Shonak wrote:
What's your university studies, knockout? Cycling statistics & winning probalities? Either way, some pretty neat calcs _O_
Close enough. I think the proper translation is Industrial Engineering. I'm supposed to write my bachelor thesis these days so I try to find all the different possible ways to procrastinate - currently, I'd say that I'm quite succesful with that
matt - pretty sure it's either a meaningless or slightly positive sign for your prediction. I hope it's the positive one looking at my own correlation coefficients at this time
If you have the time to do it, it would be nice to test it in last year's ranking, so we have an example of how the "predictions ranking" should look like at the end of the season
Aquarius97 wrote:
If you have the time to do it, it would be nice to test it in last year's ranking, so we have an example of how the "predictions ranking" should look like at the end of the season
For the final PCT ranking:
Spoiler
Used those that MARS collected in this screenshot:
