B1G Tournament Forecasting Thread

Status
Not open for further replies.
#1      

danielb927

Orange Krush Class of 2013
Rochester, MN
By request, here's a thread to collect statistical forecasts for the rest of the season. I have my laptop running these nightly but haven't automated the process of making the charts to share yet, so for now I'll try to get to updates once or twice a week!

All of the models I run use KenPom's rankings as their basis. I run the remainder of the season 100k times with three scenarios — normal, no tiebreakers (so any ties are resolved by a 50-50 coin flip), and equal strength (where every game remaining is a tossup).

As of the morning of Feb. 5, here's the forecast for B1G Tournament seeds (normal case). Teams are ordered by mean expected finish, which is why IL is above PSU, for example. Usually that choice doesn't change things much at all, but this is a very messy season.

1580911096515.png


One other note — in these charts, "0%" refers to outcomes that happened less than 0.5% of times, but at least once, while a blank refers to outcomes that happened zero times in 100k simulations.
 
Last edited:
#2      

danielb927

Orange Krush Class of 2013
Rochester, MN
By contrast, here's what the seeding projections would be if every game from here out were a toss-up. This gives a sense of how clear the B1G race is due to just the standings and length of season remaining since, in this model, each team is expected to go .500 the rest of the way.

The most obvious difference from the chart above is the 13/14 seeds. Nebraska and Northwestern are almost guaranteed the bottom two seeds, but those odds are only 70% and 87% here. This tells us that, in some sense, the "depth of the hole" that those teams have dug themselves — even if they started playing on par with the rest of the conference, they'd still be more likely than not to wind up with a bottom 2 seed.

1580911543441.png
 
#3      
By request, here's a thread to collect statistical forecasts for the rest of the season. I have my laptop running these nightly but haven't automated the process of making the charts to share yet, so for now I'll try to get to updates once or twice a week!

All of the models I run use KenPom's rankings as their basis. I run the remainder of the season 100k times with three scenarios — normal, no tiebreakers (so any ties are resolved by a 50-50 coin flip), and equal strength (where every game remaining is a tossup).

As of the morning of Feb. 5, here's the forecast for B1G Tournament seeds (normal case). Teams are ordered by mean expected finish, which is why IL is above PSU, for example. Usually that choice doesn't change things much at all, but this is a very messy season.

View attachment 5632

One other note — in these charts, "0%" refers to outcomes that happened less than 0.5% of times, but at least once, while a blank refers to outcomes that happened zero times in 100k simulations.
This is really cool. Thanks for doing this. I'm not sure I like your sorting, though. If Penn State has a 39% chance of finishing in the top 2, and Illinois only has a 34% chance, it seems like PSU should be ahead of Illinois. Is Penn State being penalized for the very remote chance of finishing 13th?
 
#4      
First off, thanks for doing this. I enjoy this stuff tremendously.

I see a pretty big difference in the KP odds and the 50/50 odds for PSU and Maryland.
I also note that they do not play each other in the remainder of the season.
Here are the sums of the top 4 seed probabilities for the 5 schools
The center column is using KP odds, the right column is based on 50/50

Maryland 82 73 (pretty big drop)
Illinois 66 71
PSU 66 50 (big drop)
MSU 64 67 (only gain)
Iowa 55 51

Again, I'm not sure what to make of it, but thought it was interesting.

I'll take Illinois have a 2 in 3 chance of a double bye !
 
#5      

danielb927

Orange Krush Class of 2013
Rochester, MN
This is really cool. Thanks for doing this. I'm not sure I like your sorting, though. If Penn State has a 39% chance of finishing in the top 2, and Illinois only has a 34% chance, it seems like PSU should be ahead of Illinois. Is Penn State being penalized for the very remote chance of finishing 13th?

That's right — when you're figuring a mean, the tails of a distribution can have a big effect, so that 0-2% higher chance for all of the 8th-13th seeds really drags down their average. It's exceedingly close, though. IL, PSU, and MSU have mean projected seeds of 3.75, 3.80, and 3.90, respectively.

Anyhow, I tend to agree with you that it doesn't look "right" somehow and will probably shift to something else for sorting.
 
#6      

danielb927

Orange Krush Class of 2013
Rochester, MN
One thing I'm interested in is a way to quantify how "clear" the conference race is at some point in a season. So here's an attempt at that — two metrics I'll call "True Clarity" and "Perceived Clarity".

To compute these, I took the standard deviation of each team's possible finish distribution (a row in the charts above). Here's what that looks like for the runs from this morning:

1580964496727.png


The "most unclear" scenario I can think of would be a case where all teams had equal, 1-in-14 chances for all seeds. In this case, the standard deviation of their distribution is just over 4. So we can define the "clarity" as:

Clarity = 1 - (average Std. Dev) ÷ 4.03

Based on that metric, I can think of two versions: "true" clarity, where the distributions don't take into account the strengths of the teams, and "perceived" clarity, which uses a rating system to forecast the games.

I haven't had time to look at many data points yet, but for the two dates I've post-processed so far this season:

January 13 (about 1/4 way through the season)
True Clarity - 8%
Perceived - 33%

Today (just over halfway)
True Clarity - 39%
Perceived - 50%

This seems to fit with what we've seen the last few weeks. Back on Jan 13, 10 teams were within a game of each other with 15 to play. Now there's more like a 3-game spread and only 8-9 games left. But during that time, the teams have (I think) squeezed together a bit in the rankings, so our "best guess" is closer to random than it was before.

What do you all think? Does 40—50% feel about right for how "clear" the B1G race is right now?
 
#7      
By request, here's a thread to collect statistical forecasts for the rest of the season. I have my laptop running these nightly but haven't automated the process of making the charts to share yet, so for now I'll try to get to updates once or twice a week!

All of the models I run use KenPom's rankings as their basis. I run the remainder of the season 100k times with three scenarios — normal, no tiebreakers (so any ties are resolved by a 50-50 coin flip), and equal strength (where every game remaining is a tossup).

As of the morning of Feb. 5, here's the forecast for B1G Tournament seeds (normal case). Teams are ordered by mean expected finish, which is why IL is above PSU, for example. Usually that choice doesn't change things much at all, but this is a very messy season.

View attachment 5632

One other note — in these charts, "0%" refers to outcomes that happened less than 0.5% of times, but at least once, while a blank refers to outcomes that happened zero times in 100k simulations.
You should report your “average finishes”. I assume there are hidden decimals, but using the rounded integers you reported gives:

Maryland average finish 2.83
Illinois average finish 3.77
PSU average finish 3.8
MSU average finish 3.86
Northwestern average finish 13.6

Also, Torvik doesn’t have as much of a gap between illinois and Maryland.http://www.barttorvik.com/conodds.php?conf=B10
 
#8      

Hoppy2105

Little Rock, Arkansas
@danielb927 This stuff is amazing. Thanks so much for turning the stat machine back on! I’m sure it’s a somewhat time intensive project.

As to your question on true and perceived clarity, I think your metric is capturing it pretty well. With about 9-11 games left for each team, the picture should definitely be clearer while still taking all the variables into account. (Teams left to play, KenPom, road vs home)
 
#9      
By contrast, here's what the seeding projections would be if every game from here out were a toss-up. This gives a sense of how clear the B1G race is due to just the standings and length of season remaining since, in this model, each team is expected to go .500 the rest of the way.

View attachment 5633
I like this, so thanks...I may choose not to look at this again all season pending the outcome of the next 2 games...win them both & I would think we become the favorite, lose them both & not so much.

So question on this one...if every game is 50-50, with both us and MD identical records, shouldn't our numbers be the same as MD, same for other teams with identical records at this point and this chart look exactly like the current standings? Maybe the better example is IA/PSU/RU, where PSU is 1 game ahead in the loss column, but splits IA/RU in the ave. ranking.

What am I missing?
 
#10      

danielb927

Orange Krush Class of 2013
Rochester, MN
I like this, so thanks...I may choose not to look at this again all season pending the outcome of the next 2 games...win them both & I would think we become the favorite, lose them both & not so much.

So question on this one...if every game is 50-50, with both us and MD identical records, shouldn't our numbers be the same as MD, same for other teams with identical records at this point and this chart look exactly like the current standings? Maybe the better example is IA/PSU/RU, where PSU is 1 game ahead in the loss column, but splits IA/RU in the ave. ranking.

What am I missing?

Good question. I think the difference is probably coming from the fact that the 50-50 scenario includes tiebreakers, and those tiebreakers aren't 50-50 anymore given the results that have already happened. Since we lost our first game against Maryland, I'm guessing we lose the tiebreaker more often than not — any scenario where we lose tomorrow but still tie them, they'll come out ahead.

Regarding the Iowa/PSU/Rutgers split, this model was run at the start of Feb 5 before Iowa's loss, so Iowa/PSU were both at 7-4 and Rutgers was at 7-5. I would expect Penn State to be clearly ahead of the other two in the 50-50 scenario now.
 
#11      

danielb927

Orange Krush Class of 2013
Rochester, MN
You should report your “average finishes”. I assume there are hidden decimals, but using the rounded integers you reported gives:

Maryland average finish 2.83
Illinois average finish 3.77
PSU average finish 3.8
MSU average finish 3.86
Northwestern average finish 13.6

Also, Torvik doesn’t have as much of a gap between illinois and Maryland.http://www.barttorvik.com/conodds.php?conf=B10

I'll start including that in the spreadsheets going forward! You can see from my post this morning that those averages are about right. KenPom has Maryland all the way up at #10 versus #17 in T-Rank, and us at #24 vs #20, so I'm assuming that (plus our home game tomorrow) accounts for the difference. I absolutely choose to believe Torvik for the next 24 hours!
 
#12      
You should report your “average finishes”. I assume there are hidden decimals, but using the rounded integers you reported gives:

Maryland average finish 2.83
Illinois average finish 3.77
PSU average finish 3.8
MSU average finish 3.86
Northwestern average finish 13.6

Also, Torvik doesn’t have as much of a gap between illinois and Maryland.http://www.barttorvik.com/conodds.php?conf=B10

I did a quick check, could not find it. Does Torvik do a home vs. away or is it just vanilla, combined reporting?

There is a rather pronounced difference between stats at home vs. away.
 
#13      
Good question. I think the difference is probably coming from the fact that the 50-50 scenario includes tiebreakers, and those tiebreakers aren't 50-50 anymore given the results that have already happened. Since we lost our first game against Maryland, I'm guessing we lose the tiebreaker more often than not — any scenario where we lose tomorrow but still tie them, they'll come out ahead.

Regarding the Iowa/PSU/Rutgers split, this model was run at the start of Feb 5 before Iowa's loss, so Iowa/PSU were both at 7-4 and Rutgers was at 7-5. I would expect Penn State to be clearly ahead of the other two in the 50-50 scenario now.
Thanks, the fact that you have tie breaker scenarios make sense & is obviously what I was missing... makes it a level deeper than I realized, nice work. Let’s beat MD tonight & then MSU & I expect we will all like how this looks
 
#14      

danielb927

Orange Krush Class of 2013
Rochester, MN
Update from this morning:

Perceived Clarity: 48% (down 2% after the Iowa loss @Purdue!)
True Clarity: 39% (unchanged)

1581084031051.png
 
#16      

danielb927

Orange Krush Class of 2013
Rochester, MN
Would love to see us play ourselves into a completely clean set of 11-14 boxes tonight!
 
#18      

Epsilon

M tipping over
Pdx
The next four games are going to decide a lot for us regarding standing, probably the most important stretch left as far as first place aspirations go.
 
#19      
Its a very exciting time for the whole conference and we are certainly right in the thick of things.
After wanting to be relevant for soooo very long, this is absolutely fantastic.
Wish I could teleport back to SFC tonight.
Hope the crowd gives me chills tonight
 
#20      

danielb927

Orange Krush Class of 2013
Rochester, MN
Does a win tonight do that??

We'd be 9-3, while Michigan is in 12th at 4-7, so it certainly would still be possible for the 12th place team to jump us. It would be harder for all 11 teams to jump us (since they're mostly playing each other), so I don't know if we'd mathematically "clinch" or not with a W tonight — my guess is no. That said, we might still get to a less than 1-in-100k chance, which would do it as far as these charts are concerned!
 
#22      

danielb927

Orange Krush Class of 2013
Rochester, MN
Update after last night's loss.

Conference Clarity: 49% Perceived; 39% True

1581179468885.png
 
#23      
Still with a 57% chance of a top 4 seed and next best is 34%.
I would like to own that top line, but this spot is not too bad.
Now . . . .
A win on Tuesday would certainly come in handy
 
#24      

danielb927

Orange Krush Class of 2013
Rochester, MN
Just looking at the standings this morning, I have a feeling the conference picture is even less clear than it was at the start of the weekend! Will try to get some updated charts posted tonight.
 
#25      
Still with a 57% chance of a top 4 seed and next best is 34%.
I would like to own that top line, but this spot is not too bad.
Now . . . .
A win on Tuesday would certainly come in handy

73.4% chance at a Top 4 Seed per Torvik with the next being Rutgers at 42%. If we win vs MSU I'd expect that gap to grow. If we lose this game vs MSU I'd expect the Rutgers / Illinois percent to be approximately equal. Need to get a W against MSU and make progress on securing at Top 4 finish.
 
Status
Not open for further replies.