|
Post by hiller 87 on Nov 11, 2012 0:31:48 GMT -6
While I do feel like the current "NHL style" standings are much better than the old way we used to determine seedings and such for Nationals, I feel like there are a couple other alternatives that we should at least discuss, possibly even more.
Felix of WKU has come up with a formula that gives points based on wins. It's best described this way- a win over a team with a .750 win percentage would be worth more than a win over a team with a .250 win percentage. Teams can lose points by losing games, but losing to tough teams doesn't penalize you as much as losing to bad teams (hopefully I'm doing this justice for you Felix, if you want to explain in more detail that'd be awesome!)
His standings look like this right now:
1 JMU 4.152 2 UK 4.088 3 MSU 4.047 4 GVSU 3.469 5 Kent 3.313 6 SVSU 2.885 7 CMU 2.177 8 MBI 2.031 9 UMD 1.888 10 DePaul 0.981 11 WKU 0.931 12 TU 0.719 13 BGSU 0.692 14 OSU 0.131 15 Miami 0.031 16 VCU 0.000 17 EMU -1.531 18 LCC -1.531 19 NSU -1.531 20 RIT -1.531 21 UN-L -1.531 22 UNT -1.531 23 UofL -1.531 24 UWP -1.531 25 WIU -1.531
JMU is in first place due to the fact that they beat UK (who I believe has the best win percentage) and forced OT against Kent (you get points for OT in this system, just like the current system.)
Andy Lieblich of MSU has come up with a RPI system (similar to what they use for the NCAA Basketball tournament) that works like this- It's your team's win percentage x .25+ your opponent's win percentage x .50+ your opponent's opponent's win percentage x .25. This makes strength of schedule a bonus, so not only is winning important but beating good teams is beneficial.
The RPI standings as of right now:
# 1 [0.5752590731186622, 'MSU'] # 2 [0.5582531307977736, 'UK'] # 3 [0.5488737161531279, 'JMU'] # 4 [0.5342592592592592, 'GVSU'] # 5 [0.5244858031119823, 'KENT'] # 6 [0.5035117943281209, 'SVSU'] # 7 [0.5030618686868686, 'WKU'] # 8 [0.49209023734885804, 'OSU'] # 9 [0.48331611570247934, 'UMD'] # 10 [0.471425693101225, 'CMU'] # 11 [0.4452400278293135, 'BGSU'] # 12 [0.4366488868274582, 'DEPAUL'] # 13 [0.43449463118580767, 'TU'] # 14 [0.4064989177489178, 'MIAMI'] # 15 [0.38876855287569567, 'VCU'] # 16 [0.3633928571428572, 'MBI']
MSU is #1 in this formula due to playing the hardest schedule and having the second best win percentage.
Both sides have their pros and cons, and both would need a little tweaking in my opinion to be effective (for example, they need to figure out how to deal with Saturday of Nationals for seeding for Nationals on Sunday) but they're both really good formulas.
I was just curious as to see the rest of the league's opinions on these possible alternatives to the current rankings, or if anyone else has any ideas of possible alternatives.
|
|
|
Post by WKU-Perrone-76 on Nov 11, 2012 10:31:11 GMT -6
Thanks Sam!
The breakdown for my formula is:
A win against a team with a 1.0 WLP (win-loss percentage) equals 4 points A win against a team with a WLP of .75-.999 = 3.5 points A win against a team with a WLP of .5-.749 = 3.25 points A win against a team with a WLP of .25-.499 = 3 points A win against a team with a WLP of .001-.249 = 2.75 points A win against a team with a WLP of 0 = 2.5 points
Then you add:
An OT-L (Overtime Loss) against a team with a 1.0 WLP = 1.05 points An OT-L against a team with a WLP of .75-.999 = .975 points An OT-L against a team with a WLP of .5-.749 = .95 points An OT-L against a team with a WLP of .25-.499 = .925 points An OT-L against a team with a WLP of .001-.249 = .9 points An OT-L against a team with a WLP of 0 = .875 points
Then from that total number you get, you subtract the losses:
A loss against a team with a WLP of 1 = -1.25 points A loss against a team with a WLP of .75-.999 = -1.375 points A loss against a team with a WLP of .5-.749 = -1.5 points A loss against a team with a WLP of .25-.499 = -1.625 points A loss against a team with a WLP of .001-.249 = -1.75 points A loss against a team with a WLP of 0 = -2 points
Then you divide all that by TGP (Total Games Played)
So the very basic formula is:
((Wins+OT-Ls)-(Losses))/TGP
But within those Wins, OT-Ls, and Losses would be the wins against teams with WLP of 1, >.75 but <.999, etc
Also instead of showing a negative value for those who have lost more games than they've won, it's adding the absolute value of the lowest ranking team everyone. The only problem is for those who haven't played a game yet, it gives a value error, so I put a rule in saying a team who hasn't played a game would show the lowest ranking team's value, giving them a negative number, and would put the lowest ranking team who has played a game at 0. Does that make sense?
|
|
|
Post by mgonzo46 on Nov 11, 2012 19:40:08 GMT -6
A couple things that I'm confused by with your system, Felix (I'm not trying to be negative, I just want to understand the concept behind it a bit more): based on JMU's ranking, you count their loss to Kent State as an OT loss, but not a loss. What's the reasoning behind this? Next, how did you decide on the values for everything? It seems a bit arbitrary to me. I think the current system is easily exploited to gain more points, as is seen with Kent State (nothing against you guys, and I'm not saying the top ranking isn't deserved). It would work only if there was a better defined season with a uniform number of games, or at least close to it. As it stands right now, it's biased towards teams that are able to schedule more games. It penalizes teams like us, K-State, North Texas, and Northwestern State for being isolated from other teams. As far as the two alternatives offered so far, I prefer the RPI, personally, as it goes one level deeper in analysis, albeit a very primitive look at it. While Felix's system does track strength of schedule, it doesn't factor in the opponents' SoS. A college football example of this is Texas A&M. With Felix's system, Texas A&M's wins over Alabama and Louisiana Tech would be considered equal as they are both 9-1, but using RPI, we see that Alabama's schedule is tougher than Louisiana Tech's schedule (.587 OWP vs. .317). One important thing to think about factoring in to your calculations (if you haven't already) is the opponent's results against the team. For example, JMU's only loss is to Kent State. Kent State should get the credit for beating an otherwise undefeated team, so it might be beneficial to take out all instances of JMU vs. Kent State in the opponents' record portion of both teams' rankings. It better rewards teams for handing a team their only loss, and penalizes teams for giving a team their only win.
|
|
|
Post by WKU-Perrone-76 on Nov 11, 2012 23:17:18 GMT -6
Don't worry, I'm not offended haha.
The reason I decided not to have Overtime losses count as actual losses and thereby subtracting points is because our current system positively reflects OT-Ls with a one point reward. I could change it really quick to subtracting those same amounts only reversed (a loss against a WLP of 1 = -.875, WLP of .25-.4999 = -.95, WLP of 0 = -1.05, etc) the results would then be:
UK 4.088 MSU 4.047 JMU 3.840 GVSU 3.469 Kent 3.313 SVSU 2.617 CMU 2.177 MBI 2.031 UMD 1.888 DePaul 0.981 WKU 0.931 TU 0.719 BGSU 0.692 OSU 0.131 Miami 0.031 VCU 0.000 EMU -1.531 LCC -1.531 NSU -1.531 RIT -1.531 UN-L -1.531 UNT -1.531 UofL -1.531 UWP -1.531 WIU -1.531
With the original values, if I count an OT-L as both a loss and a OT-L, meaning the number for the loss would still be subtracted but the OT-L value would still be added, the rankings change to:
UK 4.088 MSU 4.047 JMU 3.923 GVSU 3.469 Kent 3.313 SVSU 2.688 CMU 2.177 MBI 2.031 UMD 1.888 DePaul 0.981 WKU 0.931 TU 0.719 BGSU 0.692 OSU 0.131 Miami 0.031 VCU 0.000 EMU -1.531 LCC -1.531 NSU -1.531 RIT -1.531 UN-L -1.531 UNT -1.531 UofL -1.531 UWP -1.531 WIU -1.531
As far as the values, I really have no reasoning behind them. I just kind of threw out some numbers and they seemed to work. They are very arbitrary.
|
|
|
Post by WKU-Perrone-76 on Nov 13, 2012 21:50:49 GMT -6
I did something ridiculous, but hell, maybe it'll work. I put all of the different ranking systems we've used over the years (win-loss percentage, hockey ranks, Andy's RPI, and my system) side-by-side, then created a rule comparing where a team ranks on each system. For example: UK ranks 2nd in my system, 1st in Win-Loss Percentage, 2nd in hockey-style ranks, and 2nd in Andy's RPI; I add those together and divide by four giving them a collaborative ranking of 1.75. In the instance of a tie of this system (like MSU and JMU ended up), I gave the team who played the most games a higher ranking as a reward for doing more. This is what the collaborative rankings ended up looking like.
1 UK 1.75 2 MSU 2.5 3 JMU 2.5 4 Kent 4 5 GVSU 4.5 6 SVSU 5.75 7 CMU 7.75 8 UMD 8.75 9 WKU 10 10 MBI 10.25 11 DePaul 10.5 12 TU 12.25 13 BGSU 12.5 14 OSU 13 15 Miami 14.25 16 VCU 15.75
|
|
|
Post by hiller 87 on Nov 14, 2012 11:42:13 GMT -6
I know everyone hates the BCS but I'd say that's a good ranking system Felix. The two proposed ranking systems and the two systems we've used in the past all have pros and cons, but mixing these could work pretty well.
|
|
|
Post by mgonzo46 on Nov 14, 2012 12:17:28 GMT -6
I agree with Sam. A "BCS-style" ranking system would probably be the best way to do it. As long as we don't rank teams using a poll. As long as it's composed solely of quantitative measurements, and not qualitative, I don't think we'll have the same problems as the BCS, but that's just my own opinion.
|
|
|
Post by Zigmister on Nov 14, 2012 12:55:20 GMT -6
I agree with Sam. A "BCS-style" ranking system would probably be the best way to do it. As long as we don't rank teams using a poll. As long as it's composed solely of quantitative measurements, and not qualitative, I don't think we'll have the same problems as the BCS, but that's just my own opinion. Standings by polls... that was indeed a dark period in the NCDA. I think that the simple explanations have some flaws that could readily bring up seeding problems in the Nationals tournament bracket. If we're already making it more complicated by deepening the analysis, why not involve a good number of deep statistics and take the average of all the number of factors? I like to think that the a fundamental part of the NCDA is playing dodgeball. What ever supports playing games, win or lose, competitive or non competitive; i'll favor. But of course I'll support whatever gets decided, and we'll code to spec. Tradition dictates that the Nationals host implements whatever they think best, so what say you UK?
|
|
|
Post by kentuckybrown on Nov 14, 2012 13:40:31 GMT -6
One of the guys here at UK has been messing around with all of the different suggestions and looking to come up with something and present it to our team tonight.
Whatever we decide to do we are looking at doing what is the fairest for all teams so...we want everyone's input!
What we are looking at doing for Nationals will probably be to have six pools for Friday/Saturday play. Then on Sunday we will do a ratio of Saturday's play and your regular season record. I remember that last year there was a lot of discussion as to what the ratio should be. So, I will probably go back and look at the last few years worth of data before I make a decision. We will try to be as transparent and open to suggestions as possible.
|
|
|
Post by Zigmister on Nov 14, 2012 13:50:39 GMT -6
I remember that last year there was a lot of discussion as to what the ratio should be. So, I will probably go back and look at the last few years worth of data before I make a decision. We will try to be as transparent and open to suggestions as possible. Last year's spreadsheet has a number of seed ratios in different sheets. in case you needed a lazylink.
|
|
|
Post by mgonzo46 on Nov 15, 2012 3:10:34 GMT -6
I'm going to play around with my own rankings so I can be cool like the rest of you guys. It's going to be the system that is used to calculate the International Rugby Board (IRB) rankings, with a couple of slight variations. The system is a "Points Exchange" system, meaning each match results in the transfer of rating points from the losing team to the winning team. The amount of points transferred is based on the result of the match and the difference between the ratings of the two teams, and have some bonuses for margin of victory and home-court advantage (if any). I was going to type up an explanation, but I figured it's easier to use theirs and explain my changes. Obviously, we don't have the possibility of a draw, so I've eliminated that from the possible calculations. The way I'm calculating the ratings change is: RatingsChange=.1*(RatingsGap)+GameResultIndex (+1 for a win and -1 for a loss). This replicates the two lines in the graph they show. As far as the "weighting factors," I'm using a win of 3 or more points (as 15 points in rugby is at least three scores, like in football) to add a weighting factor of 1.5x, and Nationals Finals results could be doubled like RWC Finals are. This system does eliminate the OT loss bonus seen in other iterations of rankings, but mainly because it's difficult/impossible to reward a team for this without penalizing the winning team. I'm going to start all teams at 40 (that's what new teams in the IRB are given). The good thing about this system is that it can carry over from season to season to track rankings at the beginning of a season when some teams may not have played games yet. I'll work on it over the next week (read Thanksgiving) and post the up-to-date rankings once I'm done.
|
|
|
Post by mgonzo46 on Nov 19, 2012 21:57:35 GMT -6
Just finished my rankings through the UK Invitational.
1. UK - 46.46 2. JMU - 45.48 3. Kent - 45.22 4. MSU - 44.89 5. GVSU - 42.65 6. CMU - 40.54 7. SVSU - 40.18 8. EMU - (40.00)* 8. KSU - (40.00)* 8. LCC - (40.00)* 8. NSULA - (40.00)* 8. RIT - (40.00)* 8. UNL - (40.00)* 8. UofL - (40.00)* 8. UWP - (40.00)* 8. WIU - (40.00)* 17. MBI - 39.66 18. WKU - 38.86 19. UMD - 38.41 20. DePaul - 37.27 21. TU - 37.17 22. Miami - 37.12 23. BGSU - 36.82 24. VCU - 35.11 25. OSU - 34.20
The teams with asterisks following their rating are yet to play a game. I know it's weird to have them ranked above the teams below them, but that' just the way the system works out, as odd as it may be. Checking them up against the combined rankings with this added in (more on that later), it seems to fit with the rest of the data, with no team being more than a couple places away from their spot in the "BCS-style" ranking. An interesting thing with this system is that it takes into account the point in time where a team wins/loses to another team. For example, the thing the kills Moody in the RPI is DePaul's losses at the MSU Invite lowering their SoS. However, because Moody defeated them before this, they got a bigger bump in the rankings than they would have post MSU Invite in my system. Another interesting thing is the boost Kentucky is getting from their 3+ point wins, especially the last one against Kent. Without the multiplier, Kent would be ahead of JMU and possibly be #1.
I re-did the combined ratings with my system included (removing the teams with no games played), and here are the results:
1. UK - 1.6 2. JMU - 2.6 2. MSU - 2.6 4. Kent - 3.6 5. GVSU - 4.6 6. SVSU - 6.0 7. CMU - 7.4 8. UMD - 9.0 9. MBI - 9.6 9. WKU - 9.6 11. DePaul - 10.4 12. TU - 12.2 13. BGSU - 12.8 14. OSU - 13.0 15. Miami - 13.8 16. VCU - 14.4
An alternate version of this, omitting each team's best and worst ranking, looks almost exactly the same, but with Moody jumping Maryland.
|
|
|
Post by hiller 87 on Nov 19, 2012 23:07:49 GMT -6
a few things I don't really like about that system:
1. It penalizes teams for hosting tournaments. While you should be rewarded for hosting things, you definitely shouldn't "lose" points because you're going out of your way to host things. 2. With the rankings continuing from year to year you could get some really weird things. I'm not a fan because some schools have very drastically changed results. For example, MSU was #2 in the standings with a 14-6-1 record in 2010-2011, last year we were not so good (#5 but 11-9-0). WKU is another good example, as they went from #5 in 2010-2011 to #13 in 2011-2012. OSU has gone from 9-2 to 8-8 to 0-5. 3. Margin of Victory isn't really the greatest thing for our league- see Nationals 2010 at BGSU. 4. Teams that haven't played yet should just be omitted from the rankings or something, because say GVSU doesn't play a game until Nationals one year, and SVSU plays 50 games, but they lose more than they win, GVSU shouldn't be above SVSU.
just my opinion though.
|
|
|
Post by WKU-Perrone-76 on Dec 3, 2012 15:26:29 GMT -6
Hiller, do you have an updated RPI after all the games this weekend?
|
|
|
Post by hiller 87 on Dec 17, 2012 21:31:27 GMT -6
|
|