I was looking at Virginia Tech's FPI win probabilities for each game in 2016 (via ESPN) and thought it might be fun to track as the season progresses. Who doesn't like over-analyzing data between games? I'll try to update my spreadsheet weekly, but may need a reminder along the way.
In 2015, the FPI was 78 percent accurate in FBS-FBS games.
I thought some of the numbers were interesting and it seemed like the FPI gave Tech little chance to win the marquee ACC games this year, namely UNC and Pitt. I felt like the Hokies could take one of these games. Going purely off of the percentages, FPI predicts Tech to win 5 games (really, Syracuse?) which is lower than what myself and most others expect (I'm predicting 7, maybe 8 regular season wins if we can steal a game or two).
Discuss, complain, etc. I apologize for the legend colors - my eyes' rods and cones don't work the same way as "normal" peoples', but I tried to get things close to school colors for ease.
Comments
Didn't the FPI have LSU at No1 and Oklahoma at 2?
I stopped reading when I saw VT had a less than 50% chance to beat Syracuse.
As much as I think Tech will handle Syracuse.
I could see Syracuse being a darkhorse this year for the ACC. Their QB looked good yesterday.
You mean like a darkhorse to not finish last in the Atlantic?
More like a darkhorse to finish 7th....
...unless they beat Louisville and become this years Houston team and Babers becomes the new prince of coaching in which I will pretend that I saw it coming...
The orange D is being sent to the glue factory.
I'm a dumbass
Week 2 update from our friends at the Worlwide Leader in
SportsUseless Statistics:Tech is now a 7-5 football team! Thanks to the Syracuse and Duke losses to the ACC Atlantic co-leaders (saying Wake Forest leads the Atlantic feels dirty).
The ECU "upset" seems to hurt the chances in Week 4 a bit, and I really don't see how UVA is still below 101%.
Hooray football!
Looks like they're predicting 5-7 to me....
5-7 last week. moved up to 7-5 this week.
4 of the 7 wins are a 50-60% likelihood though.
So we are still going to beat Liberty?
Time will tell.
Some things (can) never change.
Spoiler Alert: FPI numbers look pretty good after the BC win. Still projected to go 7-5, though.
Ef 'em. We are a one loss team until proven otherwise.
i mean yea that is how that works...
Anybody know how the FPI is against the spread? Asking for a friend...
Ummm, how could that formula only give us an 80 for the BC game? Were we supposed to win 102 to -9?
I suppose they deducted because we couldn't keep our scoring in check. Won by too much. /s
Week 3 Update. Thanks for letting the cat out of the hypothetical bag, Chazz.
The folks from The Acronym located in everyone's second least-favorite Bristol do, in fact, have our Hokies finishing up at 7-5 despite the complete dismantling of Boston College. We can expect a similar shellacking to be handed to the fighting Broncos (coincidentally blue and orange, and mind-bottlingly still less than 100 percent chance of victory) and have an increased odds of winning each game.
We're coming for you, Louisville.
Username: piercepj
Skill: Just coined the hashtag of the 2016 ACC season
#ComingForYouLouisville
My bad. Couldn't wait for Monday.
Wait, what?
More odds and probabilities!
After an exciting ACC slate in Week 4, Virginia Tech has a greater chance to win each game except for UVA (plummeting by 0.8 percent) and Duke (down 0.1 percent). Duke's upset in South Bend make the ND game look much more manageable, but I do expect Brian Kelly to turn them around at least a little bit. A bye week in Week 5 will take the updated probabilities out of Tech's hands and should set up a spectacular matchup with UNC. FPI says the UNC game is the toughest game remaining on the schedule at this time.
Other teams' performances notwithstanding, Tech has shown it's a serious threat to win the Coastal by pouring on points and not letting the other team score. Have you met my friend Jerod??
Hooray Football! Hooray Offense! Hooray Confidence-that-we-can-win-any-game!
Week 4* and Bye in Week 5*
Thanks for posting these! I love how we went from multiple games under 30% to all games above 40%.
You're right - updated.
Week 2 just never actually happened.
At least we get to play Miami at Lane for the first time since 2012!
Those conference expansions will get after ya!
I guess we're just erasing 2014 from memory?
Didn't we have Miami in Lane, in 2014? It's a game that I'd love to forget, but IIRC Duke Johnson ran wild that night.
Brian KellyIt's amazing what a win over CMU can do for a team. Now, instead of predicted to beat UVA 90 out of hundred times, VT is predicted to beat UVA 89 times.
Damn. That's lower than the series win percentage in the 21st century.
This latest version has us winning 5 more games? I would be ok with that, since it's a meaningful improvement over last year, but more than a bit dissatisfied.
I choose to look at it as we will win at least 40% of the games that are at 40% or greater. So a solid chance for 6 more wins.
Add the win probabilities to get the 'expected' total.
Bye Week Update:
No massive changes after Virginia Tech was unable to score any points this week. The big game to note is the upcoming UNC game. Coming off of a
huge upsetwin over an overrated FSU team at Tallahassee (how bout that kick, eh?) there's may be a little bit of Coastal hype for UNC. Tech's first real conference challenge may be a divisional semi-final game with Miami looming.The Dame game is looking more and more winnable at 49.9%. Because of that 0.1%, the FPI is holding steady at 7-5 for VT. The hype is real for the Hooville Broncos as they've gained a 2% chance of winning in Lane after rattling off two straight victories. Just kidding - they're still gonna get their asses waxed.
New LOLUVA football uniforms going to show off dat derriere?
Now I understand this one:
Musky, go home. You drunk.
Here comes everyone's favorite weekly update! Another Tech win and more increased probabilities of winning games. Virginia Tech is now favored in all but two remaining games (Pitt and Miami) and would finish 9-3 if the percentages were actual indicators. As we saw last week with the systematic dismantling of UNC at a perceived 35% win rate, the numbers don't mean much more than interesting off-day discussion points. However, it's fun to see how programs' perceptions have changed as the season progresses. Not much change this week, I guess it looks like the computers have begun to find each team's identity (weird how pre- and early-season assessments aren't totally accurate).
The tough stretch will be finished in a few weeks and I expect both Thursday night games to be Hokie victories since Lane will force Miami to collectively soil its trousers. Pitt can't seem to put anyone away late in the game, and Fuente hasn't lost in a football stadium all year. ND is finally penciled in as a statistical victory, but it doesn't matter all that much since it's a preseason game.
Happy for fun football and looking forward to Saturdays again! Buckle up, Syracuse. You're next.
Interesting to the see the GT line go down despite their loss
If you're a fan of Bill Connelly's numbers, here's VT's stat profile.
http://www.footballstudyhall.com/pages/2016-virginia-tech-advanced-stati...
Their most likely outcome is 9-3, with VT solid favorites against Syracuse, Duke, GT and UVA and coin flips against Miami, Pitt and ND.
Overall Defensive Havoc Rank: 3
I don't even know what that measures or how you measure havoc, but that sounds good.
http://www.footballoutsiders.com/varsity-numbers/2014/varsity-numbers-cr...
It's a simple measurement of impact plays by the defense. I like it because I like simple stats I can recreate.
What site/program do you use for your charts?
Pretty remarkable how poor that performance was that we had against Tennessee... 4% in percentile performance
Thinking back to the preseason predictions feature, I remember #s like 5-7, 6-6, 7-5, maybe 8-4
and 15-0were common predictions. We've basically crossed UNC off the likely loss column. But given entire body of work and that of our future opponents, 11-1 or better is entirely conceivable (from a prognosticatory standpoint).If we were to make our preseason predictions today, how many of us could realistically pick (with confidence) 3 or more losses?
Here was Bill C.'s season preview: http://www.sbnation.com/college-football/2016/6/10/11895330/virginia-tec...
They've gone from a 7-win team to a 9 win team.
Sudden outbreak of common sense based on numbers?!?!? WTF? Oh, wait... I concur :)
I like the logic here!
Interesting to see the GT line go down after a loss.Interesting to see the GT line go down after a loss.
----------------------------------------------------Harambe used to be interesting
I have a bad feeling about that game in South Bend. It seems everyone is quick to pencil in that game as a win, and I think we are a better team for sure, but they have a ton of talent and a legitimate NFL QB. I would hate to ruin our playoff shot with a late season loss, especially if we win everything leading up to it.
Playoff shot? Winning everything?
Remember, all we have to do is beat Syracuse to go 1-0
I feel like this is the new cake vs pie / burrito joints argument these days.
Burritos vs joints huh? That's tough, seem like they go hand in hand.
This is why you have two hands.
I thought you were supposed to have an airsoft pistol in the other hand or is that just Cody Journell?
Oh if I had the option of sacrificing a pie in order to get an equivalent amount of burritos you'd be hard pressed to find a pie at my house.
Totally hypothetical, I'm worried about all the rest of our games. I just mentioned that one in particular.
Well then. After dreams of an 11-1 run to the ACCCG, the Hokies got a reality check in New York. Now, I'm still hopeful that Tech can pull out a Coastal title, but they'll need to be perfect through the rest of the ACC schedule. They still control their future, but I don't see UNC losing to anyone else.
Rightfully so, the ESPN computers lost some faith in the Hokies but it's a good thing computers don't play football. Time to settle it on the first of two #ACCThursday games and validate the fact that Da U is not bak.
With six games remaining, VT is favored (FPI-wise) in half of them. I was hopeful after seeing how bad Pitt's defense has played recently, but now I'm not sure what to think. It does seem like the computers are much more confident in the Hokies now than at the beginning of the year, and I think we can all agree with that. Or maybe it's the cannibalistic nature of the Coastal Chaos that's lowering the surrounding expectations. This team has potential to win, they just need to prove that they can do that consistently.
Hokie Hype Train got a fresh load of coal this week! Feel those fossil fuels burnin' hot! Now that we're into late October the nation can finally agree on one thing: Miami is unequivocally, undeniably, absolutely, fantastically NOT back. FPI numbers returned to the pre-Syracuse state. Really, it looks like that game never happened. Looking at the remaining schedule, one big thing jumps out at me. There are no more regular season games against teams color-coded in orange (10-2 confirmed). The computers still have Pitt with the edge on Thursday, and I'll say I'm terrified of some of the potential outcomes of that game. The numbers suggest a 9-3 season and the
"official"ESPN projections suggest 8.4 wins and 4.1 losses. Interesting. Not that these numbers actually mean anything. Only 5 more updates to go.The fact that ESPN's FPI W/L numbers sum to 12.5 for every team, make them highly suspect. If we want to use math to predict this week's outcome vs Pitt, I suggest that we look at VT's ACC game score margins:
49 (BC), 31 (UNC), -17 (Cuse), 31 (daU).
Mathematical symmetry predicts that VT will win by 49.
We beat the U by 21, not 31.
Dang, my maths are bad....
OK, so adjust 49, to 39.
Don't understand this. FPI has to be taking into account our historical troubles at Heinz Field. Vegas opened us at 4.5 point faves. I understand Vegas is trying to make money, but it's still weird to see.
Not according to the documentation that describes FPI. I do think that it credits home vs road vs neutral and does so in a way that exceeds what Phil Steele's models predict, for some teams. As such, it may basically say Pitt has an average 7 pt home field advantage and VT has an average 7 point road disadvantage, to net out that VT has a higher than 50:50 chance of losing, even though it would give VT comparable odds of winning at a neutral site.
Here we are two thirds of the way through the regular season. With four games remaining, the FPI projects the Hokies to be favored in each one, including the road game in South Bend. Wouldn't it be nice to put the axe to ND's bowl hopes in that game? This was the only game that took a significant dip when compared to last week. It dropped about five percentage points after their win over Miami. The other three against the basement of the coastal continue to look positive. But as we all know, rankings and projections mean jack-diddly so anything can happen.
The good news is that 10-2 still looks very possible with a trip to Orlando impending. As of Friday/Saturday, ESPN's Andrea Adelson said that Tech has a 72.2 percent chance to win the coastal and a 12.8 percent chance to win the ACC.
Well another week down and another change in perceptions, albeit small. After an underwhelming victory in Durham that could be considered, at the very least, a neutral site game (road venue + home crowd = null) the Hokies have three games remaining to prove their belonging in Orlando against Clemson – still a 12 percent chance to take the ACCCG. The two conference games remain strongly in Tech's favor, as they're at home against the rest of the bottom of the Coastal. There was a slight dip in the LOLUVA projections due to their tough road loss against bowl-eligible Wake Forest. Georgia Tech will pose familiar problems, but I know Bud's ready. I wish I could watch next weekend, but I'll be at a wedding. Maybe I can avoid the sure family drama and watch the game with my Jacket uncles instead...
Rounding out the trio of contests is the final preseason game against Notre Dame. Following their tornadic demise early, the game has held steady at coin-flip odds. The outcome doesn't matter for the conference, but if there is any shred of CFP hope the Hokies need this win. All games are winnable, and favorable. I'm going with the computers to finish the season at 10-2. See y'all in Orlando.
I know these projections haven't quite garnered a ton enthusiasm and thought-provoking discussion as things have leveled out over the second half of the season, but I'd like to point out one item that draws my attention. Early in the season, ESPN's CPUs thought the Hokies would have a mediocre rebuilding year, which most of us agreed with (6-8 wins, a bowl, and a positive future outlook). Tech has shattered those expectations in Year 1 of the Fuente reign and the computers acknowledge that. So far this year, there are 47 positive-slope lines against 23 negative-slope lines (with 6 of each of them changing by < 1 percent). This tells me that the future is bright and I'm very
arousedexcited.I know there hasn't been a ton of discussion, but I have greatly appreciated the running updates as I don't have time to keep on top of all this otherwise.
I wish there was a way to generate theoretical matchups so that I could plot the ACCCG projections. Or to see how past games stack up at current ratings. I'd love to see today's matchup against Tennessee.
Anyone know anybody with a job at ESPN that wants a quick project to work on?
Thanks for doing this. I check in every week to see the graph update.
I really don't get how the ND game is a coin flip. We have had our less than stellar moments, but they have been awful.
Thanks for the post!!!!
They beat Miami and we seem to struggle away from Lane. I'm kind of surprised that we are picked to win.
GT is next GT is next but I really want the boys to stuff it up nd's butt.
I was really hoping to see our chances at Notre Dame jump up a little more with another W on the record. I would love to walk out of their stadium with a smile on my face.
Here we go. I didn't get to watch the game due to a #FallWedding, but my uncle (a GT alum) said we looked like garbage. If he was correct, then I'm glad I didn't watch. So onto the final two weeks, both games took dips in their percentages, with ND taking the larger hit. Down to 43 percent, this feels more accurate going on the road to South Bend and not knowing which Tech team shows up. Will it be the one that demolished BC, Miami, and UNC or the one that looked like microwaved cat poop against GT and Cuse? UVA game fell to 87 percent. I think Fuente has the guys geared up for this one to finally put away the Coastal.
Side note, ESPN says that a VT/Clemson matchup is an 83 percent chance of occuring, with Clemson having an 86 percent chance of winning. I'll take those odds, and I'll be there if that's the matchup. Virginia Tech has approx. a 90 percent chance of winning the Coastal (win against UVA or loss by UNC to NCSU). Lucky number 13 sounds good.
Still almost twice as many positive-slopes vs. negative slopes. That shows that this team is leaps and bounds ahead of schedule and with additional talent and coaches finding their ways, we're looking at perennial ACCCG contenders.
I genuinely look forward to these every week, thank you for taking the time to do whatever it is you do to get the info in front of my hairy face.
From my hairy face to yours, you are very welcome.
Also, I forgot to mention:
Welcome to Week 12. Or as it's called in the SEC: Cupcake Wars
The penultimate update for the 2016 regular season goes as follows:
I was very concerned for about 2 hours yesterday afternoon. I didn't want to have to write another negative post in here and I knew I couldn't possibly skip it since so many of you count on these funky graphs. However, we have positivity! Tech pulled out one of the greatest second halves I've witnessed and finished the preseason 3-1. With one ACC game remaining against Charlottesville High School (fitting that spell-check wants to correct Charlottesville to Chancellorsville), FPI gives VT a slight up-tick to 88 percent chance of victory.
On the other side of the division, NCSU has a 19.4 percent chance of pulling the upset and helping the Hokies win the Coastal. Next
Saturdayweekend we're all Hokie fans first and Wolfpack fans second.Current chances of beating Clemson: 13.9 percent. I'll take it.
Correction: We're NCSU fans on Friday; their game with UNCheat kicks off at noon on Black Friday.
Here I stand, corrected.
Wow, that makes things very interesting. So, we will know for certain at gametime Sat exactly what we need to do to get to Orlando. In a strange way, I almost hope we have to win Sat to get there. Okay, maybe not. But, considering how the team played vs GT (when they knew they had some "cushion"), I feel like we'll only get the team's 100% if they know they have to win on Sat.
Obviously, we want a win on Sat for obvious, LOLUVA reasons.
I don't think this takes into account the two rivalry drubbings from today, but I feel like this is a ltitle short. Clemson is beatable. Update coming tomorrow once numbers are crunched again. See y'all in Orlando, FAM.
So 85% chance of winning means Clem -14 or so?
Umm...so you actually agree with it? The metric obviously thinks Clemson is beatable, otherwise they would have given Clemson 100%.
Well UVA had a 12% chance yesterday.
Here it is, folks. The final FPI update for 2016. The graph looks like a little bit of a jumbled mess but I'm happy with the way it turned out, both from a football success standpoint and a complete painted picture of the Virginia Tech football season along with its opponents. In the initial post, I mentioned that the FPI was 78 percent accurate for FBS-FBS games in 2015. The FPI was 5-6 (45 percent) for VT games this year with a 4 game stretch of incorrectness (UNC, Syracuse, Miami, Pitt). I wish I could be 45 percent correct at my job and still claim to be an expert. The Hokies went 4-1 in games expected to lose and 5-2 in games expected to win. This is why we let the players decide the games.
Part of this experiment also suggests that the preseason numbers mean nothing. Look at where teams like VT, Miami, ND, and Tennessee were projected early on and see how after 4-5 weeks the numbers finally leveled out. It's really tough to have a robust predictive model when the teams take a while to figure out what they're doing.
Last night I looked at the VT/Clemson matchup for next week and saw that VT was expected to win this game 151 times out of 1000. Now that the numbers have refreshed following the respective dismantlings of intrastate rivals, the Hokies are expected to win the game ten more times out of 1000, up to a 16.1 percent chance. So that basically guarantees a victory. The Hokies also have the 2nd toughest remaining strength of schedule while Clemson's is 14th.
Side Note: the Game Score for the UVA game was only 76. What in the world did the computers want Tech to do differently yesterday? Score 131? Block every punt? Make Sam Rogers rush without an OL?
EDIT for proofreading errors.
I wish I could be 45 percent correct at my job and still claim to be an expert.
It's all relative to how the best are doing. Pretty sure FPI is one of the more accurate models.
Couple of typos, but good analysis overall.
Hungover math is hard, man.
I like those line thingies that all shoot to the top of the graph. Makes me feel all warm and victorious inside.
Also, when you say that we went 4-1 in games we were expected to lose, is that games we were expected to lose based off of preseason predictions or gametime predictions? I would be interested in seeing a study of these types of models to see if there is an obvious inability to predict teams with new coaching staffs (my gut says situations like ours wreck havoc on predictive systems)
Those were based off of gametime numbers. I'd also like to be able to generate matchup predictions on my own. It would've been cool to see how our chances against Tennessee continued to evolve as the season went on.
Would you say the fpi had a better beat on our team as the season wore on or was it that unpredictable throughout?
It seemed like after Week 5 or so, the projections leveled out a bit. Excluding the wild upsets, there weren't many surprises. A lot of the "unfavorable" games for Tech were in the 40-48 percent range (at gametime), so a win really isn't that unlikely. I do think that this team played above its early preseason projections and once the computers saw the numbers that were coming from all of the talent, it adjusted accordingly. I would imagine that for most teams, the first 4-6 games would seem unpredictable since they're trying to figure out their identity and adjust to whatever unmeasurable changes have taken place in the offseason.
Also, what are everyone's thoughts on this data for the 2017 season? It's not terribly labor-intesnsive but I'm worried about the viability of the data and whether it paints any sort of reasonable picture worth discussing. If folks want to see it again next year, I will and please tell me any additions or changes you'd like to see. If people don't want to see this again next year, I'll either save my ten minutes each week or you can save your mouse clicks when it appears in the sidebar.
I admit I loved seeing the data and projections on the week by week basis in one thread.
I loved getting to see this every week. I actually really look forward to seeing, hope you decide to keep it going next year. Thanks!
Yes please!
Have to admit, I loved seeing the lines shoot to the top this season.
I liked it and looked forward to seeing how their model changed week-to-week.
This was an excellent picture of our matchups and I was looking forward to the updates every week.
I always read them and enjoyed them.
More please.
This was great. I enjoyed it.
Next year will be interesting for the fact that VT will have a some preseason hype (like UNC this year). FPI, especially for 1st half of the season, seemed to rely too heavily on those preseason projections than how the games actually panned out. Even against Notre Dame, it seemed that ND was favored more for their name than actual play on the field (and to be fair, VT almost shit the bed in that one too). So, I would like to see what the FPI will look like when expectations begin high.
Most advanced stats phase out preseason predictions over the first half of the season. This might seem wonky, but it's actually pretty reasonable when you consider the alternative of putting out rankings based on 1 or 2 games of play. And for what it's worth, several other advanced stats are fairly kind to Notre Dame and I promise the computer models have no concept of name brand. FEI has them at 39th and S&P+ has them at 30th with an expected win total of 7.
Will be interesting to see based on who ends up declaring early for the draft, and if we pick up any major recruits.
love 'em, keep it up.
Maybe would be interesting to see how FPI stacks up against other systems like FEI or S&P+, but not sure how feasible that would be
Any index that incorporates subjective recruiting rankings as part of its objective formula is inherantly suspect.