Separate names with a comma.
Discussion in 'NWSL' started by cpthomas, Mar 12, 2019.
THE FINAL WEEK
NC 3 NJ 1
ORL 1 SEA 2
UTA 2 HOU 1
POR 2 WAS 0
And ... we have Final Standings for the regular season prediction part of the contest. Congratulations to blissett, who held on to a lead through the last bunch of weeks of the season and comes out as the winner by a good margin. And, congratulations to McSkillz for being top scorer for the last week.
Detailed breakdowns for this part of the contest are coming soon, as well as "per game predicted" results.
I managed to keep a 7 points margin over the second-placed competitor, who happened to be the robot, the rival I was most eager to beat since the beginning of the competition (in the name of the humans! ), as you all know, guys. It's been a long ride, but looks like I made it.
* unassumingly bows to the crowd *
Congratulations to @McSkillz, both for beating me by one point this week and for turning out being my closest human competitor for the season. You stumbled a little in the central phase of the competition, but you were mostly challenging me all weeks.
Edit: not sure if my win was mostly due to doing much better than i did in 2018 season or to other competitors, including the robot, doing much worse, but what I can say is that, after 2 or 3 years of this game, I finally managed to make predictions while totally ignoring any allegiance to teams I could have been rooting for. This kind of detachment surely paid in gold.
Can't wait to see that!
In other words, you became like a robot?
Well, don't forget that I went so far as predicting a 6-0 score. Not very robotical.
More seriously: I tried to approach the prediction in a more scientific way, if you want, but i wouldn't call it a purely "statistical approach", since I was considering many different factors that you always claimed are basically ingrained into the stats, although I never completely agreed with you about that. Things like (to randomly throw around some of them): injury report, NT duty during the World Cup, apparent state of form of some players, psychological reactions to the trend of results or to the sense of urgency that was dictated week by week by the rankings, weariness from many consecutive games including mid-week ones (we know the schedule of a 9 teams' league is very hectic, with busy periods followed by odd breaks), and so on...
Basically, if you take the weather as a metaphor, I'd say the statistician tries to predict it mostly through the past series of data and basically ignoring the actual current conditions, while I try to do the latter, looking around for any batch of available data about the current status quo (both those that can be expressed by numbers and those that can't but can somehow give a likely relevant piece of info to a human brain).
If you want, we can say that, when someone asks to the both of us what the weather will be tomorrow, in your case you rush to a book of old data and plunge into it, while in my case I open the window and look around. Not guaranteed that I'll do better (on the contrary, you demonstrated that normally only the occasional human-oddity do better than the statistical robot), but it's obvious that, for my part, I like it better this way.
Here are final "per game predicted" results. Once again blissett came in on top, but after that there were position change with Smallchief and McSkillz stepping into 2nd and 3rd places. The tables also break results into the different ways competitors could score so that you can see who scored their points where. I have one big table on my computer, but for posting here I've split it into two. The total results columns are on the right of the second table.
And, here's a little table that shows were our competitors, as a group, scored points. The row at the bottom is a possible revised scoring system for next year, based on how different it is to score in each of the scoring categories.
And, here are the percents of possible points scored results for this year and then for last year. These are for all games, so they don't take into account whether competitors predicted all or fewer than all games:
And, for the last of the regular season data, here are the final regular season game location and goals scored data:
Wow super fascinating! Thanks @cpthomas! It was fun, sometimes frustration took its toll and I wondered if there would be any movement up or down the ranks, but ultimately this was a great way to keep track of the NWSL season and added another fun dimension to the league since I won't get a NWSL team in my city until I'm 80 years old.
Well, although history is not made with "ifs", these data seem to show that, hadn't @Smallchief and @McSkillz missed some batches of prediction, there could have been two other humans able to beat the robot: that's quite re-assuring, compared to the idea that only a precious few humans can do that; 3 humans out of 12 being able to beat the robot makes for a whopping quarter of the field!
(Although, as we're going to see below, it's more the robot stinking big time this season than the human competitors doing much better than usual)
I am happy to see that I did worse (only 6th-placed) in the category that should actually be the easier one to predict (goal scored for one team) but that, as I sometimes said (but I admit the subsequent arguments against me were very valid), it's the one that I hate most and that makes less sense to me!
It's interesting that, when you parse scored points per categories, it appears that I only topped the "Correct Goal Differential" one, while I've been #2 in other two categories and just #6 (as I already noticed) in another one.
Of course, though, I was good enough in the overall predictions that it didn't matter that I didn't top every single category.
Well, as @cpthomas had noted earlier this year, it's interesting that I got better by a mere 1,5% (41,7% to 43, 2%), while the robot did worse by a massive 5,9% (46,9% to 41,0%)!
Had the robot kept the pace from the last season, it would have beaten me by far (but, on the other hand, in that same case we don't know what the score of the other competitors would have been, including mine: theoretically, we could have been all better by the same margin because everyone would have correctly predicted the "easy" games from this year that were upsets instead; or, of course, the robot could have correctly predicted games that everyone else was going to miss. Who knows?).
I hadn't noticed that, but apart from the robot and the "cyborg" (cpthomas and BSbraintrust), only two players fullfilled all of the needed 108 predictions of the season: me and @BlueCrimson!
To be fair, some of the others had quite marginal misses (107/108 for @Tapas&Fútbol, 106/108 for @McSkillz), but we've seen that, although I keep being 1st-placed in "score per game predicted" category also (thus showing that it's not been a victory coming from being just regular in not missing any prediction), those missed predictions, as marginal as they were, could have changed the final outcome for the rest of the chart, since positions were so close in the pack that was chasing me.
This is were we just have to be "robotic" too, in my opinion: the robot has many advantages over humans already; let's not forget to predict, thus giving it even more edge!
Something's not right. I might be wrong, but as I remember I predicted every game
Aargh, you're right. I missed your predictions for the four games from July 19-21. When I add them in, you stay at #8 in the overall predictions with 122 points and move from #9 to #8 in the score per game predicted with 1.130. Sorry, and good catch!
Well, I am happy that there were at least 3 humans with a 108/108 rate of predictions!
@cpthomas, what now? I suppose you're going give to us the scores from the ladder prediction we made before the beginning of the league? What about play-offs prediction? I guess the one we made along with the ladder-prediction will stay and will give some points to us if correct, but I suppose we're going to make new predictions for the actual games, aren't we?
Can you remind us of the rules? What can we predict exactly and in which format? (Score, extra-time, penalty shootout...)
We've completed the regular season games prediction part of the competition, with blissett as the CHAMPION!
There are two other parts of the competition. For one part, the Ladder Predictions, we already did the predictions way back before the first game of the season. For the other part, the Playoff Predictions, we do predictions as the playoff games proceed, similar in timing to what we did for the regular season games.
Since we are needing Playoff Predictions this week, here are their scoring rules (from the first post on this thread):
Playoff Predictions (to be posted as the playoffs progress, with each game prediction to be posted before that game begins):
8pts - Exact score predicted (W/L)
8pts - Exact score predicted (D) WITH correct advancing team
6pts - Exact score predicted (D), wrong advancing team
4pts - Exact score predicted (D) at 90min but not 120min WITH correct advancing team
3pts - Exact score predicted (D) at 90min but not 120min, wrong advancing team
5pts - Correct GD predicted (W/L)
4pts - Correct GD predicted (D) WITH correct advancing team
3pts - Correct GD predicted (D), wrong advancing team
2pts - Correct GD predicted (D) at 90min but not 120min WITH correct advancing team
1pt - - Correct GD predicted (D) at 90min but not 120min, wrong advancing team
3pts - Correct W/L prediction
2pts - Correct advancing team if you predicted W/L by 1 goal but match was D
1pt - - Correct advancing team if you predicted W/L by 2+ goals but match was D
+1pt - Not exact score, but correct goals scored for one team
6pts - correct prediction of the Final game MVP
If I remember correctly, this means all we have to do is follow our usual pattern of simply predicting the goals for each team in each game -- except for predicting the Championship game MVP. The scoring, however, is more complicated that we're used to. (It's all programmed into my computer, I just haven't looked at the scoring system in a long time.)
THE SEMI-FINAL GAMES ARE THIS COMING SUNDAY, SO WE NEED PREDICTIONS BEFORE THOSE GAMES BEGIN!
Here are the rules for the Ladder Prediction part of the competition. Part of that competition already has scores, but we won't have all the scores until the playoff games are done. Tomorrow, I'll post the scoring for this part of the competition, as well as cumulative scores to date.
Ladder Predictions (to be posted before the first game of the season begins):
6pts - Perfect prediction of a team's end position on the ladder
3pts - 1 position off of a team's end position on the ladder
1pt - - 2 positions off of a team's end position on the ladder
6pts - Correctly picking the group of 4 playoff teams
4pts - Picking 3/4 of the group of 4 playoff teams
2pts - Picking 2/4 of the group of 4 playoff teams
1pt - - Picking 1/4 of the group of 4 playoff teams
4pts - Correctly picking the Shield/Spoon winners
2pts - Your Shield/Spoon winner is within 2 positions
2pts - The real Shield/Spoon winner is within 2 positions on your ladder
2pts - For each correct advancement you put through the playoffs
1pt - - Correct method of advancement (regulation, OT, or PKs) for the semifinals
2pts - Correct method of advancement (regulation, OT, or PKs) for the final/title
6pts - Making one ladder prediction and never editing it
-1pts - For every week up to wk12 that's been played before your prediction is finalized
By the way, shouldn't Lunatica actually move from #8 to #9? She's already #8, in the chart I see.
And, if that's not enough, here's a series of tables that show the Ladder Prediction scoring, other than the part for the playoff games results. They're in a long table that runs across the page, so I've broken the table into sections so you can see how contestants did for each team. The totals are in the first table, and the details by team are in the subsequent tables. The cyborg -- the collective "us" -- killed it!
You're absolutely right, she dropped from #8 to #9. I'm going brain dead!
And, last for today but not least, here is the combined score table for the Regular Season Predictions and Ladder Predictions parts of the competition, to date. Blissett is sandwiched between the robot and the cyborg! Lunatica is the high-scoring human for the Pre Season Ladder predictions, including outscoring the robot (as CoachJon did also).
Well, as a human between non-humans, I really feel like a sandwich man.
I'm not surprised at all to see the cyborg at the top of the preseason ladder predictions. In theory, any one person is "bad" at predicting things, but on average a bunch of predictors working together will get things fairly accurate. And this general rule works much better on "grouped" predictions (i.e. crowd size or team score after a full season of games) than on more individual items/events (e.g. a dice roll or the score of a single game).
Can you recommend any reading material on grouped predictions? I'd be very interested to read something on it.
Oh, I don't personally have anything thorough on it. It's a general statement that I've heard both from an occasional class discussion (once in a math class and once in a social sciences class) and from some documentary programs on TV.
If you look at the "Wisdom of the crowd" page on Wikipedia, it mentions that the process works best on point estimation of a continuous quantity, with the idea that any individual guess includes stochastic noise (about the real value) that can be eliminated with the average of many guesses (thus isolating the real value).
The most famous example of the phenomenon comes from a county fair 1906, but it's been known since ancient times apparently (Aristotle, etc.). There was also a fairly popular book written about a decade ago.
Some important factors in what makes the "wisdom of the crowd" correct are the relatively continuity of the guessed number, the independence of the guesses, and the diversity of the guessers. If you fail strongly in any of those three categories, you're introducing some sort of bias into the guess (or, in the case on noticeably noncontinuous distributions, just simply not framing the problem well).
My use of "grouped predictions" probably wasn't the best choice of words when I was trying to talk about relatively continuous quantities where you might be able to assume normal-like distributions. (That is, you can still look at discrete situations like the number of candies in a jar - there are enough of them inside that precision doesn't mean much.) The Wikipedia page on the topic is fairly thorough with references listed throughout it, and a quick Google search gives a few nice results too: