Separate names with a comma.
Discussion in 'Women's College' started by Soccerhunter, Nov 5, 2018.
I can't win it, but I did get the final twosome right!
OK. The results are in for the 5th round.
FSU's win over Stanford really has simplified things all around. We have the winners in both the contest divisions!
First, over on the Machine/Statistical side it is all done as none of the entries picked UNC or FSU to win the championship.
The winner of the Machine/Statistical side is the Massey Ratings with a score of 144. In second place is cpthomas' Iteration 5 RPI with a score of 141. The three point difference is attributable to one game... Massey called Penn State in the Second Round while the Iteration 5 ARPI called South Carolina. Otherwise the scores would have been identical. Only one point back in third place is the NCAA ARPI. If the NCAA machine had called Texas Tech over Princeton it would have tied Iteration 5's score. Basically these three systems are in a virtual tie.
There would have been a different winner if Georgetown had won over UNC. The United Soccer Coach's Poll entry would have lead the pack by 5 points, but UNC prevailed and the three top entries gained 8 points to secure the lead. (THe Score listing is below.)
In the Hunan Division, only 2 entries out of 25 picked FSU over Stanford. What this did was to seal the top two entries. No matter what happens, on Sunday the WINNER of the 2018 NCAA Division 1 Big Soccer Prediction Contest is SOCCERLOL and the RUNNER-UP is MILLENNIUM, and THIRD PLACE will be DMTHOMAS49..
Here is how it shakes out. Right now Soccerlol has 140 points, Millennium has 138, and Dmthomas49 has 129 points. If FSU wins on Sunday these scores will not change. However, if UNC wins, the order is the same, but Soccerlol will have 148 points and win by ten over second place Millennium.
It's the fourth and fifth places that will be influenced by who wins on Sunday. If FSU wins, Tom81 will achieve a fourth(/fifth) place tie with Wendy's with 127 points. If UNC wins, then Stubifer will be in 4th with a score of 128 and Wendy's will end in 5th with 127 points.
Congratulations to SOCCERLOL! We will await your acceptance speech and insight to your brilliance (especially if UNC wins.)
5th Round Full scores follow:
MACHINE / STATISTICAL DIVISION
144 Massey Ratings
141 CPT's 5 Iteration RPI
140 NCAA RPI
133 United Soccer Coaches
107 Top Drawer Soccer
80 NCAA Non-Conf RPI
119 Tom 81
114 Kurt Kline
105 Glove Stinks
Nice, and thanks for keeping score Soccerhunter.
Congrats to SoccerLol, who be laughing all the way to the proverbial bank with his newly claimed bragging rights.
Also,, congrats to all who participated; this isn't nearly as fun without stiff competition. Until next season, guys/gals, its been real.
The machine/statistical division, at its top, came out in exactly the order my long term statistical studies said they should have; (1) Massey, (2) 5 Iteration ARPI, (3) NCAA ARPI. Pretty interesting.
If I understand this correctly, if FSU wins on Sunday, then Soccerlol will have defeated the machine/statistical division, with its top 3 finishing ahead of all the other humans. If UNC wins, the machine/statistical division will occupy positions 1 and 2 and will tie with Soccerlol for #3. This is about what's expected when humans are competing with the machines/stats.
With Massey being the machine/stats winner, for those who came in behind it, it would be interesting to see where the human competitors differed with it and to know why the humans made their decisions where there are differences. Maybe some time after Sunday, I'll see where the differences were and ask some questions of posters that differed about why they made their decisions.
Soccerhunter, I'm wondering about the total for the Non-Conference RPI. Here's what it had for the third round on (I'm copying and pasting from its posted predictions):
For some reason, I think the total you had after the third round isn't right and also isn't thereafter. You correctly have 56 after round 2, but in round three it had 3 correct (Stanford, FSU, and Baylor) which would have given it 18 for a total of 74. In the quarters, it had 2 right (Stanford and FSU) which would have given it 16 for a total of 90; and in the semis, it had 1 right (FSU), which would have given it 8 for a total of 98. It's still a poor total, but it's pretty close to TDS and I think exposes how woefully deficient TDS's prediction was.
FSU beating Stanford really tanked my bracket but I'm good with that because I am happy for FSU for making the title game.
I read an article this past week, though I can't recall where the article was, about the Stanford men and women and how their chances looked good for both repeating the titles of last year. It was more of an informative piece but it came off with a small hint of braggadocio - almost like the chickens were counted before being hatched. It pleases me that neither team will repeat. The men were knocked out by Akron, 3-2 in regulation, last night.
Yikes! I'm too tired to look into it at the moment, but I'll double check what went on next week. An initial glance looks like one of the adding formulas in a cell in the Non Confrerence column in the spreadsheet got corrupted and I didn't notice.
Oh, I've never had that happen to me. Except about 1,000 times. It's near miraculous that we can land a spacecraft on Mars and have it send back pictures. I'm sure they have computer programs that monitor their computer programs and detect programming errors all on their own. Just like you and I do ... or, maybe we don't .... Oh well, we do have fun!
PS - I really miss cbg
Hey tie for 4th place for me is not bad!
I'd like to thank all the little people in my life that made this possible...
I've entered this thing twice. Once winning it two years ago by picking USC to take home the trophy. And losing it miserably this year, but correctly picking the noles to walk away with it. I'll take that!
It's all official now, Soccerlol edges out Millennium by two points to take the 2018 DI WOmen's NCAA soccer tournament prediction contest on Big Soccer. Congratulations Soccerlol!
As noted earlier, in the Human Division Tom81 and Sweepsit were the only two entrants to pick FSU to take the Championship and so the results for our prediction contest are as follows:
1st Place: Soccerlol 140
2nd Place: Millennium 138
3rd Place: Dmthomas49 129
4th Place: Tom81 127
4th Tie: Wendy's 127
Thanks to all of the good natured entrants!
The full results are listed below followed by the Machine/Statistical Division and some comparative notes.
HUMAN DIVISION FINAL RESULTS
127 Tom 81
114 Kurt Kline
105 Glove Stinks
The Machine/Statistical Division scores remain the same with the exception of a computer (operator) error which corrupted the score for the non-conference RPI entry. The score for that entry is 104. Comment follows the report of these scores.
MACHINE / STATISTICAL DIVISION FINAL SCORES
144 Massey Ratings
141 CPT's 5 Iteration RPI
140 NCAA RPI
133 United Soccer Coaches
107 Top Drawer Soccer
104 NCAA Non-Conf RPI
In comparing the scores from previous years, some things stay the same and some things change. One thing that stays the same is that the Machine/Statistical division always out-scores the Humans on the average. It happened again this year by a whopping 24 point average difference (104.28 to 128.17) while in previous years the average disparity was closer with 7 points in 2017 and 9 points in 2015 for examples.
This year for the first time the M/S division top score was higher than the top Human. (Usually the top human picks outliers and beats the Machines even as on the whole humans do worse.) Also, last year the Non-COnference RPI came in first, while this year that entry came in last in the division. (I'll let cpthomas explain that!)
In general, this contest is a good lesson on gambling. Gambling is a statistical game which if approached dispassionately yields predictable results (good or bad.) We humans play hunches and favorites, and on the average get handily beaten by the "house".
Thank you Soccerhunter for setting all this up : ) Couldn't find another bracket challenge on the internet besides this one. Also my men's bracket complete busted so I was glad I still had one iron in the fire. Thought UNC would pull it out but ah...
I look forward to doubling down on my hunches for next year only to finish in the bottom five and then proceed to change my bracket name to socceridk.
Yes, and actually the top two in the M/S division were higher than the top Human and #3 in the M/S division tied #1 Human. Massey, my 5 Iteration ARPI, and the NCAA's ARPI all had very similar predictions. Only the Non-Conference ARPI tanked and even it came in ahead of 7 Humans and was close to several others.
The Non-Conference ARPI, for a team's games, uses only their non-conference games. That's a very small sample and, in my opinion, far too small a sample to reliably rank teams. I think in some years it will do very well and in others it will do very poorly, all basically a matter of luck. Based on the NCAA's explanation of why they have the NCARPI, they believe it gives a better read of conference strength than the ARPI. They're right, but the cost of getting better conference to conference comparisons, they get much poorer team to team comparisons. And, Massey and my 5 Iteration ARPI pretty much take care of the conference to conference problem, so if the NCAA were interested in rating reform either of those systems would be better.
I'd be really interested in hearing from participants about which games they missed the results on and why they think they got it wrong. I've been reading up on why humans make predictions that don't match with what statistics predict, so here are some possibilities:
Law of Small Numbers -- this is the tendency to use too small a data sample as a basis for a prediction. As in, X team lost to Y team, and that means Y team is better. It doesn't.
Law of Recency -- this is the tendency to use the most recent results as a basis for a prediction.
Law of the Gut -- this is a tendency some have to trust their gut feeling more than the statistical evidence in front of them. From another sphere, this is as in the Donald's recent statement that when it comes to whether the Fed should increase interest rates, he has a gut feeling and he trusts it more than all those metrics.
Loyalty -- it's hard to say the team I love is going to lose.
Wanting to be the big winner -- this is a gambling thing. I don't really care if I come in way down on the list, I'm going to make some predictions that are contrary to the stats because if I'm right I might be the big winner and that's what I really want.
Getting stuck on pre-conceptions -- this is a tough one, I think.
The Machines correctly predict game results only about 73% of the time (or 79% if ties are decided by PKs). That may not seem very good. But it's a lot better than most humans do. So, it may be fair to say, as WWC_Movement did, that the ARPI "sucks royal." But, as you can see if you compare WWC-Movement's 108 points as compared to the ARPI's 140, sucking royal is a whole lot better than he did.
I have a theory about a way for humans to participate in a contest like this and have the best chance of winning. The theory says: Don't make your predictions until you've seen what the machines are predicting. Then, take a close look at those predictions and see if you have particular knowledge that indicates some of them are wrong. Where you don't have that kind of evidence, use the machines' predictions. Where you do, make your own predictions. But, if you're making your own predictions, be sure you're not falling into any of the above traps.
Thanks Soccerhunter for doing this one more time this year. It's fun, and interesting. Next year, one or the other of us will get you some relief.
PS - Next year, I think I would put United Soccer Coaches and TDS in the Human Division. They make predictions as groups of humans, and I think that would give a better sense of how humans can do. The USC group did very well, which is interesting to me. The TDS group did pretty poorly.
On your PS regarding United Soccer Coaches and TDS...
My understanding is that the voting for the UCS weekly national polls for college soccer is that they have empaneled a representative group of a conference representatives (28 members for the D1 women) who vote weekly. (Theyalso have regional ranking committees representatives for each region in the country.) In any case these reps should be fairly close to keeping a hand on the pulse of their confernece or region.
TDS is a private profit-making business running a website. I have no clue as to how they came up with their bracket, but I can imagine how it might have gone. Glove Stinks sent me their bracket and I was amazed that they left two lines blank. In the first round the left the Texas Tech vs Princeton winner line blank, and in the second round (since they picked Rutgers to beat Duke in the first round) they left the Lipscomb vs Rutgers) line blank too. This is very puzzling, but it implies some disorganization.....So I imagine that one employee who is their college women in-house guru put together a bracket and while he may or may not have showed it to a few other staff members, it was basically a single person operation. --If I am right, that would make it very different than a poll of head coaches. (I just can't imagine a busy staff in a lean media organization formally organizing a committee to get together various brackets to represent the organization.)
For the record, I assume that most readers of this thread will know that United Soccer Coaches does not put out any brackets. To make their entry, I take their latest poll (after the regular season and conference tournaments) which includes a rank order of about 40 to 42 teams (including the "also receivingvotes" line at the bottom.) Consequently I have the data to fill in a bracket only needing 32 favored teams to start the process and go through the bracket advancing the higher ranked team. (If a game pair in the first round does not include one of the 40 rank ordered teams in the poll results, I go back a week or maybe two and find all the teams I need.) Fortunately, the NCAA selection is good enough that the overlay with the Coaches poll listing (40+ teams) pretty much solves any problems.
Very interesting and it fits with something I learned this year. For at least the last couple of years, what the Women's Soccer Committee does as it approaches the end of the season is start figuring out who the Top 45 teams are. The Top 45 is a reasonable number for them to use. There are 33 at large selections, and of the Top 45, a good number are going to be automatic qualifiers. If there are 12 AQs in the Top 45, then they have their at large selections (45 - 12 = 33). Ordinarily there are fewer than 12 AQs in the Top 45, so using the Top 45 gives them a few teams more than they will give at large positions to. For example, if there are 10 AQs in their Top 45, they will have a list of 35 teams from which to fill 33 at large slots.
So, if the USC coaches are ranking 40+ teams, this puts them very close to what the Committee is doing.
For the Committee, there's one danger with this process that I think they would need to be careful about. Even at the end of the season, things happen that can require a re-evaluation. If the Committee goes into the last week of the season with a group they've identified as the Top 45, they need to not to be psychologically attached to that group so that they can entertain changes to it based on the last week's results.
There is also the law of trusting in a program and/or coach. The good ones really know how to get the best out of their teams in the big games and usually have excellent game plans.
Also, good defensive teams usually have better odds of beating good offensive teams IMO.
Look I don't know anything about statistics or machines or any of that stuff(though I was forced to take an entry-level university statistics/probability course as a pre-req for grad school at UCLA and it bored me to tears so perhaps I have zero passion in the field) but one thing I want to challenge the stats on is that a well knowledgable human being who has been following every wsoccer program in supreme detail for years would probably do a better job than statistics. For one thing, in our very small cluster group on bs, we are just too small of a sample size to say, "well I believe machines did better than bs posters and therefore lets generalize the whole population of wsoccer people and say they won't beat machines" You know what? At the end of the day, no matter what team is listed as supposedly winning a particular round against another team, you will always have unpredictable variables that will come into play such as fatigue and rest time, weather, location of match, heckling and/or passionate audiences, a player catching the flu, another player losing a relative to a heart attack, another player having a psychotic breakdown in the middle of the night, random birds or a giant meteor crashing at the very spot the keeper is standing at, etc.(the variables are infinite you get my point) The point is, you can't calculate that stuff into Artificial Intelligence and assume they will make a better choice than a human. Personally, I won't follow any machines, I go it my own way on the brackets without much thought and if I do well, it's out of luck... just like everything else in this World. Anyway, cpthomas, you are certainly a master at this stuff more than any of us and I appreciate the work you do because I read it and try and understand it myself(not terribly well mind you but I do try because I want to understand something that seems too complicated, I just don't believe humans do worse, it depends on what humans are doing the predicting. IF you rely on me to predict well, then there's going to be a huge issue because I'm not going to be knowledgable, I'm only a fan, not an expert. But if you asked Anson Dorrance to give you a bracket and he had all the time in the World(which he doesn't) he could beat a machine.
Regarding your first point, I agree completely at least in evaluating how teams will do over the long term. I think the historic results of a coach/program don't get enough weight in a lot of peoples' thinking, as compared to who the particular players are this year. Which players have come and gone, of course, matters, but gets too much weight, I think, as compare to the weight given to the coach/program.
Regarding your second point, I don't know whether defense is more important than offense, but I do know that to win a College Cup a team must be very good defensively. I haven't checked this year's numbers yet, but prior to this year, no team ever has won a College Cup giving up more than 0.7 goals per game over the course of the season.
I completely agree that your "well knowledgeable human being who has been following every wsoccer program in supreme detail for years would probably do a better job than statistics." I think that's pretty much what the Las Vegas odds makers do for the areas in which they book bets. How this all works is something I've been interested in. One of my questions is, how much better do they do than the "machines"?
I consider Chris Henderson to come closer than anyone else to being your "well knowledgeable" human, looking at the entire array of NCAA Division I teams. And I consider the coaches in a conference to come closer than anyone else to that, looking at the teams in their conferences. That's why, this year, I tracked my "machine's" simulations of where teams would finish in their conference standings at the end of the regular season, as compare to where CH and each conference's coaches predicted they would finish. The coaches, on average, had their standings within 2.03 positions of where the teams actually ended up. CH was at 2.20. My machine was at 2.24, using what I consider to be a very crude program. Those differences are marginal. And, I think that's likely ordinary -- the differences among legitimate statistical programs are pretty marginal and the improvement that our "well knowledgeable" human can achieve beyond the machines also is pretty marginal. On the other hand, now and then a human will do significantly better, due to luck.
PS - I think all of this relates to how different people think about "climate change."
Hmmmm, I see what happened now. I had the year wrong.
Duke actually wins the National Title in 2028, according to the almanac.
Damn small print.
SoccerHunter, did you ever find a volunteer to take over the annual prediction contest?
I've PM'd Soccerhunter that I've got my computer programmed to do it, so I'll go ahead and bite the bullet. Piggybacking on his work over the last bunch of years -- "I" is Soccerhunter -- here's the intro post, followed by a more detailed post on the rules:
Hello WoSo Fans!
Once again it's just about time to set up the annual NCAA Prediction Contest for women's DI soccer.
For those of you who haven't participated, here are the basics.
For the last fiveyears, we had two divisions: (1)The Human Division where about 25 of us WOSO fans entered our best bracket, and (2) the Statistical / Machine Division where we had 6 entries last year. Usually, the very top humans beat everyone (including the computers) but the Machines beat the humans on the average.
I've received positive feedback on the system of scoring for the last five years that I've done this, so I'll suggest using that system once again. Recall that its major feature is to underweight a little bit the big payoffs for calling the semi final and final games, and emphasize more points in the second and third rounds to reward savvy picks in those "upset rounds" leading up to the elite 8. If you can call the elite 8 you're going to be in high-level company with a great shot at taking the contest. But if you get every other pick wrong except the the eventual finalists then you will end up in the middle of the heap. In other words, if you pick Stanford and and UNC to play for the 2019 championship, you might have called the winners, but if you got all of the other picks wrong, other contestants who correctly called most of the 56 games leading to the elite 8 will have a big lead on you even if they miss on the two golden teams.
In the next post, I'll set out the specific rules. (Same as last year, repeated.)
cpthomas as the inheritor from Soccerhunter