538 is good but I don't care. It's simple. Win tomorrow.
Bracketology
- Status
It was an advertisement for an ESPN service (Strength of Record) written by an ESPN analyst. Don't put too much stock in it.
I'm confused with USC. I never get to watch the PAC 12 so I'm uninformed but how are they on the bubble with a 23-8 record?
Because we live in opposite world and in opposite world Syracuse is in with an 18-14 record.
Cheez
- Sun Prairie WI
2-5 vs. top 50 (wins are UCLA and SMU, both top 20)
5-7 vs. top 100
bad loss at Arizona State (RPI 131) on Feb 26
SOS 78
Non-Conference SOS 144
I would have them in, but they cannot lose this game. Washington is #200 in RPI right now
all numbers per ESPN because I was too lazy to look elsewhere
Thanks. I got a kick out of that one :thumb: The whole Syracuse thing blows my mind.
Cheez
- Sun Prairie WI
Cant help but think that if you put USC's name on Syracuse's resume, they would be easily out.
For sure haha
Wow. USC almost had an epic collapse. That game was way closer than the final score.
danielb927
Orange Krush Class of 2013
- Rochester, MN
I agree, and took a look at 538's model plus another one that is favorable to the Illini right now, Dance Card.
TL;DR:
Most at-large picks are easy, and models that claim things like "95% of at-large bids predicted correctly" are really only doing slightly better than a coin toss on the true bubble teams. Dance Card seems to vastly overpredict their confidences, but has Illinois barely on the right side of the bubble and generally gets 2/3 of true bubble teams right. 538 seems to have a reasonable idea of their confidence on bid likelihoods and gives Illinois a 1.5% chance, but the data they present is only enough to say that they may get around 70% of the true bubble teams right. The only thing that can be said for sure is that we're in the bubbly zone that's very difficult to predict. And, with less confidence, it appears we're probably more likely to miss out than be selected, as it stands now.
The nitty gritty:
538 says we're a 1-2% chance with their model, but without info on how successful it was in testing it's impossible to know whether it's useful or total crap. All they say is that using Strength of Record (SoR) alone gets you 90% agreement with the committee's picks, which as I'm about to show, means nothing. For comparison's sake, Dance Card (where we're a 70% chance) claims a model that's gotten 95% of at-large picks correct the last few years. As I'll show, this also means very little.
Here's the problem with those 90% and 95% stats: most anyone (or metric) can predict most of the at large bids without fail. Last year, for example, 31/36 at-large teams (86%) were correctly predicted by almost every bracketologist in the bracket matrix. Throw out those freebies, look at the remaining teams, and you'll see that Dance Card is hitting about 2/3 and SoR alone is giving you 1/4, which is even worse than a coin toss.
What's really needed are statistics that show if your predicted bid chances agree with reality: do teams with a 0-10% chance really get bids 0-10% of the time? A quick check makes me think Dance Card, in particular, is highly overconfident on their reported bid chances. They assign the probabilities 100%, 100%, 100%, 99%, 70% to their last 5 teams in (the non-freebies); if those were accurate, they'd expect to get 94% of the freebies and 99% overall, not the 95% they've achieved historically. 538 is a bit more believable. Their last 5 in have 86%, 76%, 66%, 63%, and 62% chances, so you'd expect them to get about 70% of the non-freebies right. If that's true, it translates to about 96% of all at-large bids and is similar to dance card's actual performance. Without further data it's impossible to say whether the 538 model can tell you anything more than "70% likely to be in" or "70% likely to be out."
To be fair most colloquial statistics fail to compensate for correlations in their samples. Either way, we could talk probabilities till we are blue in the face, at the end of the day we can only cheer for our team. I.L.L!!!
P.S. anything that assigns 100% probability to something is either terribly flawed in their methods or rigged.
P.S. anything that assigns 100% probability to something is either terribly flawed in their methods or rigged.
JFGsCoffeeMug
BU:1 Trash cans:0
- Chicago
I thought DC isn't a probability of whether we will get a bid, but a number reflecting the percentage of teams with a similar resume who have made the dance historically. To the extent past results are not predictive, the number loses value.
danielb927
Orange Krush Class of 2013
- Rochester, MN
Good point, you're probably right. I admittedly went from recollection on what thise numbers meant.
That's correct.
It may not be intentional, but the site is a bit misleading in this regard. A lot of people will just see "chance of bid" and a percentage and assume it's a predictive probability.
danielb927
Orange Krush Class of 2013
- Rochester, MN
Sure, but to be fair, how else should they interpret it? It's not like for every current team there's a corresponding past team with the same exact profile, so it's already being predictive of the chance of a "similar" team in past years. IMO not much separates that from being an attempt at predicting the current year, plus the headline for the percent column says "chance of bid" to boot.
Edit: and one more thing, they have the cut line included and presumably update it (and the chance of bid numbers) if a spoiler steals an auto-bid. Even more reason to consider it a predictive probability for the current season.
Last edited:
#2 SOS and #1 non conference SOS along with 5 top 50 wins. Will be interesting if they become the first 15 loss team to get in. I think they still need to win a game or two.
foby
- Bonnaroo Land
With our luck, we get handled easily by M. And so far, it appears that will happen.
Dan
Admin
- Status