Need some help here...(ESPN Win Probablilty)
Combing through this AM as I am gearing up for basketball season and went game by game through our Wolverines' schedule and need some understanding on how the #22 team in America is projected to be a sub .500 team?
vs Purdue Fort Wayne) (78%)
vs Eastern Michigan (83%)
vs Pittsburgh (51%)
vs Ohio (72%)
vs Jackson State (93%)
vs Virginia (43%)
vs Kentucky (18%)
@ Minnesota (36%)
vs Lipscomb (92%)
@ North Carolina (25%)
vs Central Michigan (89%)
vs Maryland (56%)
vs Penn State (61%)
@ Michigan State (27%)
@ Iowa (20%)
vs Northwestern (65%)
@ Maryland (26%)
vs Minnesota (69%)
vs Purdue (49%)
@ Penn State (32%)
@ Northwestern (36%)
vs Ohio State (48%)
vs Nebraska (67%)
vs Indiana (39%)
@ Wisconsin (25%)
vs Michigan State (53%)
@ Rutgers (25%)
vs Wisconsin (54%)
@ Illinois (23%)
@ Indiana (16%)
Not counting the extra game in Brooklyn, my math comes out to 14-16...
The only logic I can come up with is our unproven roster and an overvaluing of home court advantage.
November 8th, 2022 at 8:37 AM ^
That stuff is meaningless nonsense until it gets some actual data. Even more meaningless nonsense for teams with as much roster turnover as we had.
November 8th, 2022 at 10:35 AM ^
Word. Beat EMU
November 8th, 2022 at 11:54 AM ^
One game at a time for me. The 3 pt shooting was off last night and many were open shots. There were many flashes of bigger things to come with this team so let's just think more about beating EMU next.
November 8th, 2022 at 12:12 PM ^
It is a bracket filled out before any games were played!
November 8th, 2022 at 8:50 AM ^
Your multiplication of probabilities incorrectly assumes each game is independent of the others, which isn't true. In reality, the games are correlated with each other.
An example relevant to election day. Let's say the Dems have a 15% chance to win the house and 50% chance to win the Senate. These aren't independent. If the Dems with the House, they have a >99% chance to win the Senate, not 50%.
Similar correlations exist in the games above.
November 8th, 2022 at 9:04 AM ^
Separate sporting events are about as close to independent trials as you can get. Certainly, when modeling them, you treat them independently because the correlation is impossible to estimate. Every season projection system you’ll find works the same way.
OP: your analysis has several flaws;
1 - ESPN May not consider Michigan to be #22.
2 - it’s hard to win on the road — the #40 team is likely favored to defeat the #22 team in its home building.
3 - Michigan has a tough schedule this year.
4 - BPI may not be the gold standard. I prefer KenPom, which, coincidentally, also ranks Michigan #22 at the moment. KenPom estimates Michigan’s record as 19-11.
November 8th, 2022 at 9:28 AM ^
Separate sporting events are about as close to independent trials as you can get.
Yeah, right. Next you'll probably try telling me that the 'hot hand' isn't real. 😊
November 8th, 2022 at 9:50 AM ^
The original reply is right in that each game UM plays will impact the win probability of all future games. For example, this week Michigan has a 97% chance to win against Nebraska. That win probability was not 97% pre-season. However, assuming the pre-season BPI is accurate for every team in college basketball, then yes the BPI p-50 projection is a losing record.
But... FPI/BPI are not as predictive as SP+/KenPom/Torvik. BPI has Michigan #87, not #40. KenPom has UM at #22, Torvik at #28, and the AP poll at #22. Torvik projects UM to go 20-10, KenPom projects 19-11. 19/20 wins seems about right.
November 8th, 2022 at 10:56 AM ^
“Your multiplication of probabilities incorrectly assumes each game is independent of the others”
Just fyi, to get the probability of two independent events both happening you multiply the probability, and this works only if the events are independent.
But to get the expected number of wins, you add the probabilities. And this works regardless of whether or not the events are independent.
November 8th, 2022 at 8:58 AM ^
The problem is your assumption that ESPN win probabilities are worth a shit. They aren't. These are bad even with a bunch of games already played that gives their system some data. With 0-1 games being played, they are entirely worthless.
November 8th, 2022 at 12:07 PM ^
Exactly. I noticed that ESPN only had Michigan with a 78% chance of winning which is not even close to what 15.5 point favorite would have. I'm looking at Torviks probabilities and they have us as an 18 point favorite over EMU and that's a 95% win probability. We had more like a 93% chance against IPFW, not 78%.
And Torvik has us projected to go 20-10 which sounds about right (12-8 in conference).
November 8th, 2022 at 9:02 AM ^
More accurately, 14.7-15.3. However, keep in mind that the "#22 team in the country" is coming from a very different source than ESPN's projected records. If you look at ESPN's BPI, which is how they actually calculate the probabilities, it has Michigan at #87. I have no idea how that's calculated, but I'm guessing it undervalues potential while overvaluing last year's results, at least compared to how poll voters vote and how KenPom calculates ratings.
November 8th, 2022 at 9:10 AM ^
Of course the derivatives guy is concerned with initial conditions 🙄
Nice concise summary of the difference.
November 8th, 2022 at 9:22 AM ^
it's also ignoring the fact that all of this will change once the models start getting more data
preseason models are worth about as much as preseason polls
November 8th, 2022 at 9:24 AM ^
If I’m not mistaken, Michigan made it to the Sweet Sixteen last year. If ESPN is overvaluing last year, it would seem necessary to rank them higher than #87, say #20 or higher. This stuff is all gibberish at this time of the year, so I would suggest the OP not lose any sleep over any of this. I would also caution him to be more circumspect when looking at statistics, lest he fall for some ponzi scheme or bad real estate investments.
November 8th, 2022 at 9:40 AM ^
I would rank ESPN the #22 sports network in the country.
November 8th, 2022 at 9:52 AM ^
What about ESPN Deportes?😂
November 8th, 2022 at 10:20 AM ^
# ocho
November 8th, 2022 at 11:40 AM ^
I've never understood that name; doesn't "deportes" just mean sports in Spanish? Completely redundant...
November 8th, 2022 at 10:04 AM ^
ESPN is #22...in Georgia.
November 8th, 2022 at 9:46 AM ^
there is no sports predictive stat more useless than the ESPN win probability. it is utter horse doodoo
November 8th, 2022 at 9:50 AM ^
...and if one steps into the ESPN Barnyard, you're bound to get stuck over your knees in shit.
November 8th, 2022 at 9:49 AM ^
I believe you may have gotten a bit off track when you referenced data from eSECpn.
This applies to all revenue sports - predominantly, football and basketball.
November 8th, 2022 at 10:22 AM ^
i'll help ya out ..... don't use espn
should solve your probelms
November 8th, 2022 at 10:29 AM ^
ESPN probabilities over index returning production for basketball. These will get better as the season goes on.
November 8th, 2022 at 11:17 AM ^
ESPN uses there own ranking system - the BPI. The BPI is EXTREMELY down on Michigan compared to places like Kenpom or Bart Torvik. They currently have Michigan ranked 84 in the country. This is why you're seeing the win % above.
https://www.espn.com/mens-college-basketball/bpi/_/view/overview/page/2
I honestly think the answer is somewhere in the middle. Michigan looks good on offense, but their rebounding is shaping up to be HORRENDOUS. I was already worried and IPFW rebounded 31% of their misses last night. IPFW is one of the worst O rebounding teams in the country.
November 8th, 2022 at 11:39 AM ^
Okay, so.. last year, we had an opening game and ran it up and looked like National title contenders. then the season happened and no.
So i refuse to take much from last night. However, what I saw was that we have far more depth at guard, though very inexperienced. But at least some bodies.
Also, they jacked up 300 3ptrs. By default, we are going to give up some rebounds. Our guards are young and dont know they need to block out on the perimeter. The B10 will punish them for that and theyll learn.
November 8th, 2022 at 11:38 AM ^
we have a 26% chance to win At Maryland? But a 26% chance at UNC? K.
Maryland sucks. Put that game at 60+% and put the UNC at 0%.
I dont think this is about how daunting the B10 is, as much as its a knock that they dont think we're any good and they dont have a clue yet
November 8th, 2022 at 11:48 AM ^
All I know is ESPN gave my fantasy team a 23% chance of winning going into MNF and I pulled it off and won by 5.
November 8th, 2022 at 12:17 PM ^
Didn't realize how rough that last month of the season is. That's a grind.
November 8th, 2022 at 1:13 PM ^
The probabilities are all 0 or 100. We're either going to take an L or get a W. Every game. Just win them one at a time.
November 8th, 2022 at 1:43 PM ^
Tell us you don't know what "probabilities" means without telling us in so many words.
November 8th, 2022 at 4:03 PM ^
It was a comment on overthinking statistical prognostication based upon an imperfect predictive model, which is far more imperfect in the preseason. This is one place where media or coaches are a more reliable indicator.
The predictive model will improve as actual current-year data accumulates. And the "probabilities" will fluctuate greatly as that happens.
It's a tough schedule, but it's not really a good use of the mathematical instrument to say, "this indicates that we're likely to be sub-500."
Weird, pejorative response, sir.