08-25-2022, 04:07 PM
(This post was last modified: 08-26-2022, 01:55 PM by dude_man. Edited 1 time in total.)
Team Luck in S36
determined by a crappy logistic regression
What is happening?
A long (long) time ago in this league, I was a fairly frequent sim tester. This was with the old sim, where it took about 2-3 minutes to run 500 tests. In those early days, it was pretty quickly discovered that you could export the league file and get all of the simulated results in an easy to filter spreadsheet - allowing you to get averages for many statistics such as yards gained and average points scored - but importantly allowing you to very easily determine your approximate win percentage.
This still exists today, obviously, though it is much more complicated in the new sim. But in those old days, the ease of using an auto-generated spreadsheet after a few minutes meant more time to build some quick statistical models that modeled the rest of a team's entire season. I've published these results in the past, but seeing as I did not bother to buy the new sim they haven't been relevant since the before times.
More recently, I've been intrigued by attempting to get a model to predict sportsbook bets to help me be better informed when making these bets. I haven't gotten anywhere with this project (and probably won't ever), but part of it required me to build a fairly bare bones game predictor. So, as any good programmer would, I went to Google, searched "predict NFL games with Python", and got my clipboard ready to copy and paste from the first link I could find.
The TL;DR of the programming is that I imported data from the index (which I already have access thanks to my past work on SurrenderPunts and WolfieBot Dashboards - I do plan on making these public as soon as I figure out how to safely do so), cleaned up some of the data, and ran a logistic regression on it to see what values affected a team's likelihood to win the most.
My overall data set was made up of all regular season and postseason games between S27 and S35, split 3/4 to 1/4 for training and testing. When I first ran it, it actually returned some nonsense results - most notably that being home was actually a slight disadvantage. Even though the model was fairly accurate at predicting the results of the test dataset, this felt wrong to me. So I committed a cardinal sin and started changing the inputs to get a more reasonable output - which I did! Does this taint the entire article? Sure, but this is just a money grab really so who cares.
In the end, these were the most important factors (according to the crappy regression model) to a team winning or losing:
Anyways, when I ran this model on the test set, it came back with about an 88% accuracy score, so I'm pretty happy about that I guess? Again, this is predicting the result based on the stats from that game - this isn't anything world breaking. If anything, it's interesting to see that 12% of the games had the statistically "worse" team winning.
After this, I went and ran this model on the S36 season, using the season's collected data as inputs - so this is thoroughly a retrospective look at the season - to try and determine win probabilities for each team against each other team, home and away. These are certainly not definitive win probabilities, but they do somewhat come up with reasonable numbers (with two big exceptions).
Win Probabilities and Power Rankings
Here is the overall win probability table - the percentages show the likelihood that the home team (the top row) wins against the away team (the left column). I believe in you all to figure out the away team's likelihood of winning.
(larger size here)
A few things to quickly point out - the model really likes Chicago and really hates Berlin and New York. Looking at the standings (I must admit here that I did not pay attention to the league last season) this seems to make sense. Chicago went 13-3 while Berlin and New York both finished last in their conference. New York has a slight edge at home over Berlin, but Berlin actually seems to be better? Most of the other teams seem to be fairly middle of the pack, but looking at the averages (which I will call the power rankings) show our two interesting outliers:
POWER RANKINGS - normalized from 0 to 1
The model looks mostly at turnover difference, location, passing attempts, and rushing attempts (the last two in an inverse way). Let's look at each of these and try to parse what's going on:
Team Luck
So, you came here to see if your team was lucky or unlucky. I think we can pretty quickly stash Yellowknife and New Orleans into the unlucky bin seeing as the model has them winning a lot but neither hit double digit wins. Honolulu and Sarasota might make the lucky bin. How do all of the teams fare?
Austin Copperheads
Record: 7-9
Projected Wins: 7.017
Difference: -0.017 wins
The Copperheads finish right on the mark here, going pretty much exactly as predicted.
Arizona Outlaws
Record: 9-7
Projected Wins: 8.931
Difference: +0.069 wins
Just like the Copperheads, Arizona pretty much performed as expected by the model.
Baltimore Hawks
Record: 7-9
Projected Wins: 6.954
Difference: +0.046 wins
Once again, the model does a good job at predicting what Baltimore ended up at. At this point you might be thinking - hey this seems way too accurate! Can you predict the future? Well, no - this model is purely a retrospective model, taking the stats from that season and seeing who should have won based on them.
Berlin Fire Salamanders
Record: 2-14
Projected Wins: 3.390
Difference: -1.390 wins
The first team with a mild discrepancy - based on the games played, the model thinks Berlin should have won 3-4 of their games, but the Fire Salamanders came out with only 2 wins. Stash them in the unlucky bucket.
Chicago Butchers
Record: 13-3
Projected Wins: 13.383
Difference: -0.383 wins
I wouldn't really call the Butchers unlucky - they were still pretty close to the model's expectation - but it's hard to not look at the 3% chance of going undefeated and feel like an opportunity was missed. This might be a fun future project to see which team was the most likely to go undefeated...
Colorado Yeti
Record: 5-11
Projected Wins: 6.529
Difference: -1.529 wins
It's probably fair to throw the Yeti into the unlucky bucket, having fallen a win and a half short of the model's expectation. It's not helping them get into the playoff discussion at 6-7 wins, but it still stinks.
Honolulu Hahalua
Record: 11-5
Projected Wins: 9.681
Difference: +1.319 wins
It feels wrong to put the Ultimus champions in the lucky bucket, but the model sure thinks they got off easy. 9 to 10 wins is still very respectable and playoff bound, but the Hahalua pulled off 11.
New Orleans Second Line
Record: 7-9
Projected Wins: 11.207
Difference: -4.207 wins
By far the unluckiest team in the league, at least according to the model. Again, I didn't pay attention to any of the games this season, so I couldn't tell you if the Second Line were just good at hitting the marks the model likes or if they were genuinely unlucky, but this is a huge difference. The model gives them barely a 2% probability of only getting 7 wins, yet here they are. 11 wins would have been good enough to tie them for first in their conference, but instead they finished second to last.
New York Silverbacks
Record: 5-11
Projected Wins: 2.814
Difference: +2.186 wins
It feels weird to call the last place team in the ASFC lucky, but here we are. The Silverbacks should have won 3 games according to the model, but they managed to get 5. Perhaps if they were a little less lucky, they'd have the first overall pick of the draft.
Orange County Otters
Record: 7-9
Projected Wins: 6.479
Difference: +0.521 wins
I don't think it's fair to put the Otters in the lucky bucket, but with their predicted win total being about 6.5 forces them to be slightly lucky or unlucky regardless.
Philadelphia Liberty
Record: 8-8
Projected Wins: 6.642
Difference: +1.358 wins
It's probably fair to call the Liberty lucky - not lucky enough to get into the playoffs but certainly lucky enough to compete for it. Baltimore arguably should have been ahead of them (just look at the point differentials!), but the luck of the Liberty pushed them ahead.
Sarasota Sailfish
Record: 12-4
Projected Wins: 9.343 wins
Difference: +2.657 wins
Sarasota are the only other team to fall outside the bounds set by the probability curve's full width half maximum - finishing over two and a half wins above expectation and competing with Chicago for home field advantage. The model places them much more in the middle of the pack - still making the playoffs, but having to fight for it. They were certainly lucky to not have to deal with that.
San Jose SaberCats
Record: 10-6
Projected Wins: 9.206
Difference: +0.794 wins
The SaberCats outperformed expectations by about a win, making them slightly lucky. The model still thinks they are playoff bound, but only just over Arizona.
Yellowknife Wraiths
Record: 9-7
Projected Wins: 10.381
Difference: -1.381 wins
Our protagonists from earlier this article, the Wraiths weren't really as unlucky as I thought coming in - certainly not as unlucky as the Second Line. That being said, the Wraiths should be kicking themselves for not finishing the job.
Luckiest Teams
Determined by Wins over Expected:
Least Expected Results
The logical conclusion to this is to see which games least followed the model's expectations - which games were the biggest upsets. We will look at the top 10 games that bucked the trend.
10. Week 13 - Baltimore @ Yellowknife
Result: Baltimore 38 - 20 Yellowknife
Loser's Win Probability: 73.03%
The model loves Yellowknife and doesn't particularly care for Baltimore - but I guess the hawks didn't care for the model either. The Hawks came out and dominated this game, and it was never in doubt.
9. Week 13 - Austin @ New Orleans
Result: Austin 30 - 27 New Orleans
Loser's Win Probability: 75.13%
The scoreline says "Austin squeaked away", but They were up by 10 points with under 2 minutes to go. The Second Line had more rushing yards, but TE82's 400 passing yards helped win them the day.
8. Week 14 - New York @ Orange County
Result: New York 26 - 23 Orange County
Loser's Win Probability: 79.14%
One of quite a few New York upsets in this list, the Silverbacks helped crush any ideas Orange County might have had about squeaking into the playoffs with this win. A combination of Ian Cole's 4 field goals and Regina Ferraro's pick six helped keep the Otters at bay.
7. Week 16 - Orange County @ New Orleans
Result: Orange County 21 - 18 New Orleans
Loser's Win Probability: 80.18%
New Orleans once again falls short of expectations in a game that ultimately meant nothing for playoff placement - the Second Line fell behind 21 early and failed to use their 18 point comeback in third quarter to win the game. It was a truly abysmal day on offense for both teams, as neither topped 400 yards and combined for 4 total turnovers.
6. Week 11 - San Jose @ New York
Result: San Jose 31 - 38 New York
Loser's Win Probability: 81.01%
Look at those Silverbacks, stunning the world again. After storming off to both a 21-0 lead and a 31-10 lead, the Silverbacks completely blew it with 2:38 remaining in the game. However, a 9 yard touchdown pass from Malcolm Savage to Nacho Macho Man after a clutch drive (and two SJS penalties) helped New York seal the win.
5. Week 15 - Chicago @ Philadelphia
Result: Chicago 27 - 30 Philadelphia
Loser's Win Probability: 81.63%
A rare loss for the Butchers, their comeback from 30-0 down fell just short. It is very impressive that they managed to score 27 points in just 12 minutes, but it was too late.
4. Week 6 - Yellowknife @ Berlin
Result: Yellowknife 27 - 34 Berlin
Loser's Win Probability: 82.94%
Berlin shocks the world by taking down the underperforming Wraiths - the Fire Salamanders dominated through the air for the big upset and once again hurting the Wraiths chances.
3. Week 1 - Baltimore @ Chicago
Result: Baltimore 36 - 24 Chicago
Loser's Win Probability: 86.04%
The Hawks walked into Chicago in week 1 with all the confidence in the world, and they put on a show. After taking the lead in the 3rd quarter they never looked back - but perhaps the Butchers used this loss to dominate for the rest of the season?
2. Week 16 - New York @ Arizona
Result: New York 36 - 30 Arizona
Loser's Win Probability: 87.16%
New York with another big upset - taking down playoff contender Arizona in the last week of the year. The Silverbacks dominated through the air and ground, and despite their two turnovers and 11 penalties they came back from a 16 point 4th quarter deficit to bring the game to overtime. Michaelangelo McTurtle's touchdown run in overtime helped dash any thoughts the Outlaws might have had at a home playoff game.
1. Week 1 - New Orleans @ New York
Result: New York 31 - 24 New Orleans
Loser's Win Probability: 91.41%
The biggest upset happened in Week 1? We wouldn't have known it then, but the model insists this was the least likely result from the season. Despite a non-existent rushing attack, New York managed to take a 31-17 lead in the 4th quarter and held on long enough to keep the Second Line away. Even with their overall performance, this is the Second Line's 3rd big loss on this list and New York's fifth (!) surprise upset in the top 10.
Well, there you have it! I hope this was at least interesting reading for you. I wrote this model way back in May or June and then promptly ignored it while I had a very busy summer, but I'm glad I could spend the day writing this up so that work didn't go to waste. I wouldn't take the results pumped out from this as gospel, but they do seem kind of reasonable given the circumstances.
determined by a crappy logistic regression
What is happening?
A long (long) time ago in this league, I was a fairly frequent sim tester. This was with the old sim, where it took about 2-3 minutes to run 500 tests. In those early days, it was pretty quickly discovered that you could export the league file and get all of the simulated results in an easy to filter spreadsheet - allowing you to get averages for many statistics such as yards gained and average points scored - but importantly allowing you to very easily determine your approximate win percentage.
This still exists today, obviously, though it is much more complicated in the new sim. But in those old days, the ease of using an auto-generated spreadsheet after a few minutes meant more time to build some quick statistical models that modeled the rest of a team's entire season. I've published these results in the past, but seeing as I did not bother to buy the new sim they haven't been relevant since the before times.
More recently, I've been intrigued by attempting to get a model to predict sportsbook bets to help me be better informed when making these bets. I haven't gotten anywhere with this project (and probably won't ever), but part of it required me to build a fairly bare bones game predictor. So, as any good programmer would, I went to Google, searched "predict NFL games with Python", and got my clipboard ready to copy and paste from the first link I could find.
The TL;DR of the programming is that I imported data from the index (which I already have access thanks to my past work on SurrenderPunts and WolfieBot Dashboards - I do plan on making these public as soon as I figure out how to safely do so), cleaned up some of the data, and ran a logistic regression on it to see what values affected a team's likelihood to win the most.
My overall data set was made up of all regular season and postseason games between S27 and S35, split 3/4 to 1/4 for training and testing. When I first ran it, it actually returned some nonsense results - most notably that being home was actually a slight disadvantage. Even though the model was fairly accurate at predicting the results of the test dataset, this felt wrong to me. So I committed a cardinal sin and started changing the inputs to get a more reasonable output - which I did! Does this taint the entire article? Sure, but this is just a money grab really so who cares.
In the end, these were the most important factors (according to the crappy regression model) to a team winning or losing:
- Turnover Difference. If your opponent had more turnovers, you were much more likely to win.
- Location. Home is better - this was kind of artificially selected, but the data backs it up.
- First Downs. Make more first downs? More likely to win. Give up fewer first downs? Less likely to win. Seems like common sense, right?
- Passing Attempts. Interestingly, the more your team passed the ball, the less likely you were to win. The opposite goes for your opponent - the more they pass the ball, the less likely they win. I originally thought this has to do with having to pass more to catch up, but...
- Rushing Attempts. The same pattern was there for rushing attempts. I don't know what to make of this, but it's there?
Anyways, when I ran this model on the test set, it came back with about an 88% accuracy score, so I'm pretty happy about that I guess? Again, this is predicting the result based on the stats from that game - this isn't anything world breaking. If anything, it's interesting to see that 12% of the games had the statistically "worse" team winning.
After this, I went and ran this model on the S36 season, using the season's collected data as inputs - so this is thoroughly a retrospective look at the season - to try and determine win probabilities for each team against each other team, home and away. These are certainly not definitive win probabilities, but they do somewhat come up with reasonable numbers (with two big exceptions).
Win Probabilities and Power Rankings
Here is the overall win probability table - the percentages show the likelihood that the home team (the top row) wins against the away team (the left column). I believe in you all to figure out the away team's likelihood of winning.
(larger size here)
A few things to quickly point out - the model really likes Chicago and really hates Berlin and New York. Looking at the standings (I must admit here that I did not pay attention to the league last season) this seems to make sense. Chicago went 13-3 while Berlin and New York both finished last in their conference. New York has a slight edge at home over Berlin, but Berlin actually seems to be better? Most of the other teams seem to be fairly middle of the pack, but looking at the averages (which I will call the power rankings) show our two interesting outliers:
POWER RANKINGS - normalized from 0 to 1
- Chicago (13-3) - 1.000
- New Orleans (7-9) - 0.793
- Yellowknife (9-7) - 0.759
- Honolulu (11-5) - 0.658
- Sarasota (12-4) - 0.634
- San Jose (10-6) - 0.608
- Arizona (9-7) - 0.592
- Baltimore (7-9) - 0.477
- Philadelphia (8-8) - 0.436
- Orange County (7-9) - 0.392
- Austin (7-9) - 0.381
- Colorado (5-11) - 0.353
- Berlin (2-14) - 0.069, nice
- New York (5-11) - 0.000
The model looks mostly at turnover difference, location, passing attempts, and rushing attempts (the last two in an inverse way). Let's look at each of these and try to parse what's going on:
- Turnover difference. The Wraiths rank 11th in the league at +5, while the Second Line are 9th with a differential of -2. These are not good. Consider Honolulu (2nd, +11) and Sarasota (1st, +12) and you can see why I am extremely confused.
- Location. I looked at who had the best "homefield advantage" (who has the biggest boost in win probability at home versus the same team away) - Yellowknife were 1st with a 10.63% boost at home, while New Orleans was 6th with a boost of 9.78%. Honolulu were 8th with a 9.44% boost and Sarasota were 10th with an 8.57% boost. The top team (Chicago) is 13th with a boost of 6.93%.
- First Downs. This might be the strongest source of the outliers. Yellowknife was 5th in first downs (257 first downs) while New Orleans was 1st (275 first downs) in S36. Honolulu was 14th (234 first downs) and Sarasota were 11th (245 first downs) - I guess they were scoring too fast? Opposing team first downs are a bit more of a wash - Yellowknife had a league best 221 given up, New Orleans was 4th (242 first downs), Honolulu was 14th (277 first downs) while Sarasota was 3rd (239 first downs).
- Passing Attempts. Yellowknife was 8th in terms of opponent passes (573 attempts against) and 11th (488 attempts) in terms of their own attempts. New Orleans was 6th in opponent passes (558 attempts against) and 14th (431 attempts) in their own attempts. Honolulu was 14th (634 attempts against) and 10th (714 attempts) respectively while Sarasota was 7th (559 attempts against) and 2nd (491 attempts) respectively. This might be another big addition to the rankings as the lower your own pass attempts, the better the model sees you - the Wraiths and Second Line are near or at the bottom of the league.
- Rushing Attempts. This might be the weirdest one I see. Yellowknife are 1st and New Orleans 2nd in opponent rushes - meaning they were the two teams least rushed upon. The model prioritizes more rushes against you, so this doesn't make much sense. They also were the teams that ran the ball the most (4th and 1st respectively), also counter to what the model wants. Granted, rushing attempts affect the model much less than the rest of these factors.
Team Luck
So, you came here to see if your team was lucky or unlucky. I think we can pretty quickly stash Yellowknife and New Orleans into the unlucky bin seeing as the model has them winning a lot but neither hit double digit wins. Honolulu and Sarasota might make the lucky bin. How do all of the teams fare?
Austin Copperheads
Record: 7-9
Projected Wins: 7.017
Difference: -0.017 wins
The Copperheads finish right on the mark here, going pretty much exactly as predicted.
Arizona Outlaws
Record: 9-7
Projected Wins: 8.931
Difference: +0.069 wins
Just like the Copperheads, Arizona pretty much performed as expected by the model.
Baltimore Hawks
Record: 7-9
Projected Wins: 6.954
Difference: +0.046 wins
Once again, the model does a good job at predicting what Baltimore ended up at. At this point you might be thinking - hey this seems way too accurate! Can you predict the future? Well, no - this model is purely a retrospective model, taking the stats from that season and seeing who should have won based on them.
Berlin Fire Salamanders
Record: 2-14
Projected Wins: 3.390
Difference: -1.390 wins
The first team with a mild discrepancy - based on the games played, the model thinks Berlin should have won 3-4 of their games, but the Fire Salamanders came out with only 2 wins. Stash them in the unlucky bucket.
Chicago Butchers
Record: 13-3
Projected Wins: 13.383
Difference: -0.383 wins
I wouldn't really call the Butchers unlucky - they were still pretty close to the model's expectation - but it's hard to not look at the 3% chance of going undefeated and feel like an opportunity was missed. This might be a fun future project to see which team was the most likely to go undefeated...
Colorado Yeti
Record: 5-11
Projected Wins: 6.529
Difference: -1.529 wins
It's probably fair to throw the Yeti into the unlucky bucket, having fallen a win and a half short of the model's expectation. It's not helping them get into the playoff discussion at 6-7 wins, but it still stinks.
Honolulu Hahalua
Record: 11-5
Projected Wins: 9.681
Difference: +1.319 wins
It feels wrong to put the Ultimus champions in the lucky bucket, but the model sure thinks they got off easy. 9 to 10 wins is still very respectable and playoff bound, but the Hahalua pulled off 11.
New Orleans Second Line
Record: 7-9
Projected Wins: 11.207
Difference: -4.207 wins
By far the unluckiest team in the league, at least according to the model. Again, I didn't pay attention to any of the games this season, so I couldn't tell you if the Second Line were just good at hitting the marks the model likes or if they were genuinely unlucky, but this is a huge difference. The model gives them barely a 2% probability of only getting 7 wins, yet here they are. 11 wins would have been good enough to tie them for first in their conference, but instead they finished second to last.
New York Silverbacks
Record: 5-11
Projected Wins: 2.814
Difference: +2.186 wins
It feels weird to call the last place team in the ASFC lucky, but here we are. The Silverbacks should have won 3 games according to the model, but they managed to get 5. Perhaps if they were a little less lucky, they'd have the first overall pick of the draft.
Orange County Otters
Record: 7-9
Projected Wins: 6.479
Difference: +0.521 wins
I don't think it's fair to put the Otters in the lucky bucket, but with their predicted win total being about 6.5 forces them to be slightly lucky or unlucky regardless.
Philadelphia Liberty
Record: 8-8
Projected Wins: 6.642
Difference: +1.358 wins
It's probably fair to call the Liberty lucky - not lucky enough to get into the playoffs but certainly lucky enough to compete for it. Baltimore arguably should have been ahead of them (just look at the point differentials!), but the luck of the Liberty pushed them ahead.
Sarasota Sailfish
Record: 12-4
Projected Wins: 9.343 wins
Difference: +2.657 wins
Sarasota are the only other team to fall outside the bounds set by the probability curve's full width half maximum - finishing over two and a half wins above expectation and competing with Chicago for home field advantage. The model places them much more in the middle of the pack - still making the playoffs, but having to fight for it. They were certainly lucky to not have to deal with that.
San Jose SaberCats
Record: 10-6
Projected Wins: 9.206
Difference: +0.794 wins
The SaberCats outperformed expectations by about a win, making them slightly lucky. The model still thinks they are playoff bound, but only just over Arizona.
Yellowknife Wraiths
Record: 9-7
Projected Wins: 10.381
Difference: -1.381 wins
Our protagonists from earlier this article, the Wraiths weren't really as unlucky as I thought coming in - certainly not as unlucky as the Second Line. That being said, the Wraiths should be kicking themselves for not finishing the job.
Luckiest Teams
Determined by Wins over Expected:
- Sarasota Sailfish (+2.657)
- New York Silverbacks (+2.186)
- Philadelphia Liberty (+1.358)
- Honolulu Hahalua (+1.319)
- San Jose SaberCats (+0.794)
- Orange County Otters (+0.521)
- Arizona Outlaws (+0.069)
- Baltimore Hawks (+0.046)
- Austin Coppperheads (-0.017)
- Chicago Butchers (-0.383)
- Yellowknife Wraiths (-1.381)
- Berlin Fire Salamanders (-1.390)
- Colorado Yeti (-1.529)
- New Orleans Second Line (-4.207)
Least Expected Results
The logical conclusion to this is to see which games least followed the model's expectations - which games were the biggest upsets. We will look at the top 10 games that bucked the trend.
10. Week 13 - Baltimore @ Yellowknife
Result: Baltimore 38 - 20 Yellowknife
Loser's Win Probability: 73.03%
The model loves Yellowknife and doesn't particularly care for Baltimore - but I guess the hawks didn't care for the model either. The Hawks came out and dominated this game, and it was never in doubt.
9. Week 13 - Austin @ New Orleans
Result: Austin 30 - 27 New Orleans
Loser's Win Probability: 75.13%
The scoreline says "Austin squeaked away", but They were up by 10 points with under 2 minutes to go. The Second Line had more rushing yards, but TE82's 400 passing yards helped win them the day.
8. Week 14 - New York @ Orange County
Result: New York 26 - 23 Orange County
Loser's Win Probability: 79.14%
One of quite a few New York upsets in this list, the Silverbacks helped crush any ideas Orange County might have had about squeaking into the playoffs with this win. A combination of Ian Cole's 4 field goals and Regina Ferraro's pick six helped keep the Otters at bay.
7. Week 16 - Orange County @ New Orleans
Result: Orange County 21 - 18 New Orleans
Loser's Win Probability: 80.18%
New Orleans once again falls short of expectations in a game that ultimately meant nothing for playoff placement - the Second Line fell behind 21 early and failed to use their 18 point comeback in third quarter to win the game. It was a truly abysmal day on offense for both teams, as neither topped 400 yards and combined for 4 total turnovers.
6. Week 11 - San Jose @ New York
Result: San Jose 31 - 38 New York
Loser's Win Probability: 81.01%
Look at those Silverbacks, stunning the world again. After storming off to both a 21-0 lead and a 31-10 lead, the Silverbacks completely blew it with 2:38 remaining in the game. However, a 9 yard touchdown pass from Malcolm Savage to Nacho Macho Man after a clutch drive (and two SJS penalties) helped New York seal the win.
5. Week 15 - Chicago @ Philadelphia
Result: Chicago 27 - 30 Philadelphia
Loser's Win Probability: 81.63%
A rare loss for the Butchers, their comeback from 30-0 down fell just short. It is very impressive that they managed to score 27 points in just 12 minutes, but it was too late.
4. Week 6 - Yellowknife @ Berlin
Result: Yellowknife 27 - 34 Berlin
Loser's Win Probability: 82.94%
Berlin shocks the world by taking down the underperforming Wraiths - the Fire Salamanders dominated through the air for the big upset and once again hurting the Wraiths chances.
3. Week 1 - Baltimore @ Chicago
Result: Baltimore 36 - 24 Chicago
Loser's Win Probability: 86.04%
The Hawks walked into Chicago in week 1 with all the confidence in the world, and they put on a show. After taking the lead in the 3rd quarter they never looked back - but perhaps the Butchers used this loss to dominate for the rest of the season?
2. Week 16 - New York @ Arizona
Result: New York 36 - 30 Arizona
Loser's Win Probability: 87.16%
New York with another big upset - taking down playoff contender Arizona in the last week of the year. The Silverbacks dominated through the air and ground, and despite their two turnovers and 11 penalties they came back from a 16 point 4th quarter deficit to bring the game to overtime. Michaelangelo McTurtle's touchdown run in overtime helped dash any thoughts the Outlaws might have had at a home playoff game.
1. Week 1 - New Orleans @ New York
Result: New York 31 - 24 New Orleans
Loser's Win Probability: 91.41%
The biggest upset happened in Week 1? We wouldn't have known it then, but the model insists this was the least likely result from the season. Despite a non-existent rushing attack, New York managed to take a 31-17 lead in the 4th quarter and held on long enough to keep the Second Line away. Even with their overall performance, this is the Second Line's 3rd big loss on this list and New York's fifth (!) surprise upset in the top 10.
Well, there you have it! I hope this was at least interesting reading for you. I wrote this model way back in May or June and then promptly ignored it while I had a very busy summer, but I'm glad I could spend the day writing this up so that work didn't go to waste. I wouldn't take the results pumped out from this as gospel, but they do seem kind of reasonable given the circumstances.