Fun With Predictions - An interactive CFP predictor
FiveThirtyEight has a great CFP predictor. I'm no 538 fanboy (though one couldn't tell from my last two posts), but this is pretty cool.
https://projects.fivethirtyeight.com/2022-college-football-predictions/
Go Blue!
November 10th, 2022 at 3:21 AM ^
Click the top 8 as winning then click Tennessee and TCU losing this week
Our chances at winning the conference are 28%
Then click on TCU winning
Our chances of winning the conference got to 33%
Im confused.
November 10th, 2022 at 3:38 AM ^
Maybe some variable for common opponents/strength of record gets adjusted?
November 10th, 2022 at 3:40 AM ^
Maybe more half-baked 538 do-do.
November 10th, 2022 at 6:07 AM ^
tru dat, and all based on the whatcha-ma-call-it thingy.
November 10th, 2022 at 7:38 AM ^
Punt: whatcha-ma-call-it thingy
Counterpunt: dad gum thingamajig doodad
November 10th, 2022 at 8:34 AM ^
I hope the Punt/Counterpunt authors see this and incorporate it into this week's article
November 10th, 2022 at 7:23 AM ^
My wife blames 538 for 2016. Statistics and models can tell you a lot but when there is a 95% chance of something happening, fate can still pick the 5%.
November 10th, 2022 at 8:33 AM ^
Interesting. Most people blame Hillary for 2016.
November 10th, 2022 at 8:55 AM ^
Your wife is wrong. 538 put the odds at about 1 in 4 on election day in 2016. Everyone else was in 95 to 99 percent range.
I'm no Nate Silver fan, but he was closer than anyone else.
November 10th, 2022 at 9:09 AM ^
Goldman Sachs suspended trading in Mexican Pesos around 10AM on Election Day 2016. That was the most accurate predictor of all. Money talks, bullshit walks.
November 10th, 2022 at 9:29 AM ^
Agree with your general sentiment but that's not even accurate. He had Trump's odds at closer to 1 in 3. Which, when you look at sports, a team with a 33% chance to win wins *all the time*
Silver and his crew are easily the best politics stats people out there, whatever the criticisms of them may be.
November 10th, 2022 at 7:03 PM ^
I like their statistics and models as an exercise. Sometimes they are very insightful. But sometimes they leave out data that is available and relevant to make their point.
November 10th, 2022 at 7:03 PM ^
I like their statistics and models as an exercise. Sometimes they are very insightful. But sometimes they leave out data that is available and relevant to make their point.
November 10th, 2022 at 8:48 AM ^
The way that happens is this: the model works by running 10000 different simulations of the rest of the season. Percentages are based on the number of times a specific outcome occurs within the set of simulations where the specific outcome you select occurs. When you click on as many specific results as OP did the number of simulations where that specific set of results happened is much lower than 10000. With this lower sample size random chance plays a much bigger role and in the simulations with TCU winning the 30% chance we beat Ohio State came up in favor of us in more simulations than in the set of simulations with TCU losing, just by random chance.
November 10th, 2022 at 1:10 PM ^
Exactly this. The analysis was done via Monte Carlo simulation and to your point, clicking all those scenarios reduces the data set to a relatively small number of scenarios and then toggling back and forth between TCU winning or losing just cuts those scenarios into two even smaller sets.
It's like flipping a coin ten times and saying, well I got heads four of the first five and only two of the second five, what gives? Arbitrarily splitting scenarios post hoc, doesn't change the underlying probabilities. Even at large samples you'll see some variation, but not significant.
Here, the difference between 28% and 33% is not significant.
If the analysis was done using Bayesian first principles, then the Michigan conference win percent wouldn't change (as expected) because you'd assume it's independent of the TCU outcome which it mostly is.
November 10th, 2022 at 10:12 AM ^
Lies, damn lies, and statistics
November 10th, 2022 at 3:21 AM ^
Looks like every OOC game affects teams conference championship winning % dramatically.
November 10th, 2022 at 6:57 AM ^
Why even include Texas here? In what world would they have a 1/3 shot at making the playoff if they win out? 538 has a lot of stuff with poor validity like this
November 10th, 2022 at 8:09 AM ^
Is "poor validity" anything like "unverified voracity?"
I keed. That would be a terrible name for a column.
November 10th, 2022 at 8:00 AM ^
Poor validity is the spice of life, I once heard.
November 10th, 2022 at 8:20 AM ^
Horrible to know that if we win out, Clemson has a 66% chance of making the playoffs. Ugh. I wouldn't mind playing them in the first round of the CFP, but you know that isn't happening.
November 10th, 2022 at 8:25 AM ^
You're looking at the wrong column. They have a 66% chance of winning the conference, not making the playoff.
November 10th, 2022 at 8:44 AM ^
Whew, thank God. You're right. I feel better now.
November 10th, 2022 at 8:43 AM ^
Can we just play 538 as a number together the whole Mgoblog family? Pitch in 5 dollars and we will never ever miss a post from 538 again.