they are promising us tempo again in 2018. gase's schemes are murder already in a non tempo based speed of play they'll be lethal in a fast tempo not allow the opposition to match personnel or force them into basic coverage looks. I'm not playing with that tempo with osweiler or fales cause they would screw it up pre to post snap. I know osweiler would. but with tanny provided everyone else knows their assignments absolutely. we didn't come out of it a couple years ago because of the qb despite what anyone might tell you. we came out of it cause ryan was spending more than half the time at the los getting guys set up correctly instead of verifying protections and making what the offense calls for pre snap reads. ideally we go to the line everyone's lined up there's 25 seconds left on the play clock tanny can verify the protections force the opposition to declare with dummy snap counts make the pre snap reads or audible into a better play vs the coverage or front played. also you can't run this speed offense with true ceiling unless you trust the qb to not have to to rely on motion based coverage id's...
This times 1000. A few of us on here have been saying this stuff for years, but people can't seem to grasp it.
Or they were trying to manually calculate the ceiling by solving the function over and over again with different values. On paper. We did it in my graduate stats class as an exercise in humility. A roll of newspaper print and markers. Teams of 4. Yuck.
We used to swear that Kenny Rodgers wrote the song "Ruby" about my grandmother (who's name was Ruby)...she would get angry and take off walking. Even in her late 90's, she'd walk for miles through Miami to get something simple from the store...just to get out and see the world. One time she was vacationing in Jacksonville at one of her daughter's houses, got mad because they wouldn't go buy her a beer, and she announced that she was walking back to Miami. That was her thing; she had one beer every night before going to bed and swore that she couldn't sleep without it. To older generations it was considered like an elixir for good general health, so she drank one (and only one) every day of her adult life about 30 minutes before bed. So off she went.....they found her hours later like 30 miles south on I-95 making her way through St. Augustine (1/10th of the way to Miami...). My grandmother lost her mind in her mid-90's but that didn't slow her down one bit- she'd have no idea where she was going but off she went! Search parties became almost a daily occurrence if you refused to give her that beer...if necessary, she would wander off every single night in search of a cold one. Only, she wouldn't even consider entering a bar because that's where low-life's went; she was definitely born of a different era. I smile every time I hear that song verse- Oh Ruby, don't take your love to town. Oh Ruby, for God's sake, turn around.
Hmm.. we can estimate the number of separate calculations needed here. If they looked at all games over a ~10 year period or so in the 60's and early 70's that easily puts you at 1000+ games. When doing linear regression (calculating expected win%) you'll have 4 separate columns in a matrix, one for each component of passer rating, meaning you have to do 4x1000 calculations right off the bat just to create that matrix. Add one more column of just 1's in that matrix (5 total columns) and you have a 1000x5 matrix. Now you need to do matrix multiplication of the transpose of that matrix with itself which is 25,000 (5x5x1000) separate multiplications at minimum. That's just on the passer rating side. Then.. you'll have to do 1000 divisions to calculate win%, multiply the transpose of the 1000x5 matrix mentioned earlier with this 1000x1 matrix of win%, which is 5k calculations. There's some other stuff, but that's the absolute minimum. So added together you'd have to do ~35k separate calculations by hand for 1000 games. And you can't make big mistakes either lol. I think in this case you really need a computer!
I picked the middle option, but it wouldn't surprise me if he either did 1 above or 1 below. I don't see the 2 extremes as likely.
I'm not a math guy and never will be when it comes to this but what are the expected wins at a 90 and 100 qbr clip?
8.56 expected wins for 90 rating and 10.34 expected wins for 100 rating. You just need to plug PR = 90 and PR = 100 into 0.1785*PR - 7.51 to get those numbers. Here's a graph that might be easier to work with. Keep in mind these are all 2017-adjusted ratings, so they're comparable to ratings in the year 2017.
You're probably right, because there were only five qualifying QBs last season who had a rating over 100 - Brady, Brees, Smith, Goff and Wentz. Those guys led some of the very best offenses in the league. What surprised me more though, is that there were only four more QBs who started at least half of their team's games and had a rating over 95 - Rivers, Stafford, Wilson and Keenum. That's it. So in 2018, anyone at or above was a top 9 QB, as far as rating is concerned. Roethlisberger, Cousins, Ryan, Winston, and McCown round out the other qualifiers with a rating over 90.
I’ll tell you from experience that our manufacturing input cost optimization curve with only four inputs needed a computer also. So I definitely agree with you.
I should also point out that Tannehill's 93.5 rating in 2016 was #12 that year, and would have been #12 this past season also. Right around what Kirk Cousins, Ben Roethlisberger did, and better than Matt Ryan. Frankly, if Tannehill was playing for a team other than the Fins and was going to be a free agent, I think that a lot of the same Miami fans who bash him would be drooling all over the idea of signing him. They only knock him because the team's record while he's been here.
In my more optimistic moments? Dependable to start the season, as Tanny gets used to playing speed and is a bit cautious overall. The second half of the season is when the offense gets in stride and Tanny gets close to elite status. Top 8 QB play, which is enough for a winning record and a playoff win.
Generally pretty true. Last season, all seven QBs who had a rating of 98+ had a winning record as a starter. Only the Lions and Matt Stafford missed the playoffs.
I voted 95 and if I’m right we win 9-10 games....if he is a 100 qbr guy the dolphins are the eagles of 2018
And matt Stafford has eliminated his penchant for the big mistake in the clutch. Something Kirk cousins has yet to do. Anybody want to take a guess what a jay cutler 2017 80 qbr gets you? A lock lottery pick
I hate to sound like a wussy, but the Dolphins tweeted Tannehill running out on the field and I kind of got nervous seeing him run. I probably wouldn't normally feel that way, but it's not like they didn't clear him ready to go back in 2016. He will probably be ok. I just think I am going to need a while before I am totally confident he is back. Just saying. Not that it matters what I think.
He will get the protection and the offense weapons he never had in his career before - so he will be surprise any expert and improve as he did in 2016. He has WR targets, TE targets and TEs able to block, RBs who can get the rushing game going AND block, too. He also has an O-Line worth to be mentioned. No more injury-proned Pouncey, no more Bushrod, two reliable Guards, a worthy backup for Ja´Wuan to stay healthy - the Phins even would heavily improve if they had JC starting. If the "menthoring" by Sitton (for Tunsil), Gore (for Drake) and Amendola (for the WRs) goes the right direction, the Phins offense will become at least average - at least!
I'm curious- your formula only accounts for offensive production based on a number ((95 QBR, 100 QBR or whatever). We do know that the defense also wins some games without the QB hitting a top tier threshold and since you didn't use that in a formula, it automatically assumes we're an average, 16th ranked defense with average stops, turnovers, etc. (because QBR is essentially ranking a QB's performance against opponents...the math is assumed even without being there at all). So here's my question- could you give us the same comparison solely based on defensive performance? In other words, a defense would have to rank top <whatever> for it to correlate with a <x> win season. The reason I'm asking is because I'm wondering how much a top <x> defense even matters if you have a top QB torching up the field.
An important stat, regarding how many of the team's losses have not really been Tannehill's fault: The team has been 3-19 in games where he's been forced to throw 39 or more passes. The average score allowed by the Fins defense in those losses was 30.47 points. (23.66 in the three wins) In games where he's thrown 34-38 passes, the team is 12-12. The average points allowed by the Fins defense in those games was 19.53. There are a few blowouts in both directions. In games where he's thrown 19-33 passes, the team is 23-8. The average points allowed by the Fins defense in those games was 18.65. So, despite all of the other things that RT has had to deal with between the coaching changes, terrible offensive lines, variable receiving corps, and inconsistent run games, when the defense can hold the opponent to a league average point total, Tannehill has done enough to help the team win at a .636 percentage. (better than ten wins over sixteen games). When the defense performs like one of the worst ones in league history, the team tends to lose (gasp!). And note that the Fins defense has done just that in about a quarter of Tannehill's starts.
Let me give you two different answers. First, to do this right we need to do all calculations using z-scores (standard deviations above or below the mean). That allows you to directly compare across era without adjusting anything and I can use stats from all of NFL history to do this (well.. I'm going to do it from the 1970 merger). First question is what is the relationship is between the z-score for points allowed and win% across NFL history. The best-fitting line to that data is W% = 14.36*PA + 50. That means that if you are 1 standard deviation above the mean in points allowed, then expected win% = 64.36%. All you need to do now is to find the mean and standard deviation of points allowed for any given year to use that equation. For 2017, the mean was 347.5 points allowed per season and the standard deviation was 44.7. https://www.pro-football-reference.com/years/2017/opp.htm That means that giving up 347.5 - 44.7 = 302.8 points in 2017 should on average give you a 64.36% win%, or 10.3 wins in a 16 game season. Closest team to that in 2017 was the Ravens with 303 points allowed. They won 9 games, meaning their offense underperformed by 1 win last season. The second answer I'll give you is the one you really want. Here's the relation between win% and BOTH points allowed and passer rating from 1970: That equation in purple is what you want: W% = 8.97*PR + 11.63*PA + 50. That tells you how z-scores for passer rating (PR) and z-score for points allowed (PA) are related to win% through NFL history. All you need now is the mean and standard deviation for passer rating in any given year to use this. For 2017 the mean passer rating was 85.1 and the standard deviation (this is team passer rating, not individual QB btw..) was 9.884. So for example.. if Tannehill had a 95 passer rating in 2017, that's z-score = 1 for passer rating. With an average defense, which by definition is a z-score = 0, you get 8.97*1 + 11.63*0 + 50 = 58.97% win%, or 9.44 wins in a 16 game season. You get the idea.. and now you can play around with that to see what expected win% is for different combinations. Oh.. one more thing to note: the ratio 11.63/8.97 = 1.2965 tells you how much more important defense is than passer rating for winning: it's 29.65% more important. Contrast that with the best-fitting equation between win% and both points allowed and points scored. That equation is W% = 11.29*PS + 10.76*PA + 50, so overall.. offense is 11.29/10.76 = 1.0493 or 5% more important for winning, but defense is 30% more important than whatever passer rating captures.
Ha!.. yeah Club members might get that reference but those not in Club won't so let me summarize. That equation W% = 11.29*PS + 10.76*PA + 50 relating win% to z-scores of points scored and points allowed shows that over the regular season offense is more important than defense by about 5%, and it's pretty stable across NFL history. So the question was raised as to whether that was only true for regular season, and whether it changed for wild card, division, conference championships or SB. So.. with what I think were a bunch of nice figures I showed the following: 1) For Wild Card winners, mean z-score on offense is 0.6383 while for defense it's 0.5655. 2) For Division winners, mean z-score on offense is 0.7831 while for defense it's 0.4887. 3) For Conference Champions, mean z-score on offense is 0.6431 while for defense it's 0.7026. 4) For Super Bowl winners, mean z-score on offense is 0.9389 while for defense it's 0.4072. 5) For all playoff games the winning offense z-score averaged 0.7294 while for defense it was 0.5464. What that shows is that: 1) you tend to be well above average on both sides of the ball if you're a playoff team, 2) that even for playoff teams the offense is on average more important than defense, and 3) that for some reason the one exception historically is for winner of the conference championship where defense has been slightly more important than offense. Anyway.. that's the summary for those not in Club. Good reference though!
Here’s a thread with some relavent stats I started a year back (please note cbrad takes the statistical analysis about 20 levels higher than I did) https://thephins.com/threads/building-a-winning-team.91138/ The correlation of offensive pr to win% is 0.67 The correlation of defensive pr to win% is -0.53 The run game appears to be a Nash equilibrium about getting defenses to commit resources away from defending the pass. So while the run game does not correlate to winning or losing, teams that run too little face defenses that are geared to stopping the pass and suffer a decrease in passing efficiency. Teams that run too much do have improved passing efficiency, but since the run is less efficient at gaining yards and points they will suffer on the scorebaord. I haven’t done a deep dive into it but my initial thought is that the Nash equilibrium for the optimum rush% changes team by team, or even game by game, depending on talent/scheme matchups. The short answer is that if you are going to use one number to predict a team’s record, offensive paser rating is the best one.
I followed you on everything but finding the standard deviation for points allowed, or specifically what levels would constitute increasing/decreasing by 1. Or is it simply 0 is average, 1 is better than average and -1 is worse than average? That's the only part that lost me yet it seems to be the key to the whole formula. You also answered where I was indirectly going anyway- the D has up to 29.6% more possible impact than the QB. Which makes sense; you have 11 defenders as a group vs 1 QB picking and choosing his targets. Let's hop to Pauly for a sec though and then get back to this since it directly corresponds where I was going. I was going to ask the same thing about the run game but that's probably too misleading- we know that the run game is a sign of being in the lead in the first place, so I think that's going around in circles to some extent. Running teams win more, but they only do so because they are already winning when they ran those extra downs. If they fall behind, they lean on the pas just like everyone else. So let's pretend (for now) the run game has little overall factor in wins (even though we know it certainly is a factor, especially if established early or it slows the pass rush....it just doesn't fit as nicely in this exercise YET). Can you show a chart only with the deviation of wins vs points allowed? From there I think I'll have all the ammunition I'll need to put something together. My thoughts here is that we've assembled a super-fast, super-aggressive D built for playing with the lead....AKA, not stuffing the run while playing from behind but smothering receivers and getting after the QB hard. That's how we beat NE last season and it was absolutely beautiful to watch...but it was one of the few times we've ever had a solid lead and could turn the dogs loose like that (which equates to the run game analogy earlier.....you only do "x" when "y" condition is already met). Where I'm going with this is that if RT puts up his 95 QBR and earns us an average of 9 wins on his own- that doesn't mean we'll have a 9-win season......that's just how far he likely carries us. To me, the bigger formula is if he comes out hot and scores early, that means we're "releasing the hounds" on defense and playing on top-5 D levels...maybe even higher with the group we're putting together. First quarter dominance also means we can add the run game to the conversation and that dial just keeps pushing upwards towards that 1.0 win percentage. But for the run and the D to deliver like they did in NE, we need those first quarter leads to bring Gase's full vision to life. I know I'm talking in circles here but I see a potential for dominance if just one more domino will fall where we want it to.
I’ll let cbrad answer the stuff about SDs because he’s much better at explaining it than I am. As for the run game I will quote myself from this thread https://thephins.com/threads/how-much-does-a-qb-contribute-to-winning.89352/ This leads to another question. If team passer rating and team passer rating allowed are so important and rushing is statistically unimportant, why rush at all? This has been something that annoys me about some stat sites, Football Outsiders is one culprit but by no means the only one. Often they will say something like “play action passing is more effective than passing without playaction”. However they compare all play action passes to all other passes without controlling for down and distance, or other game situations. Firstly comparing yards/carry, passer rating. And rush/pass split NFL AVERAGE: 4.1 ypc; 88.4; 40.9/59.1% 1st and 10: 4.2 ypc; 85.6 50.2/49.8% 1st and 10+: 4.4 ypc; 84.7 36.2/63.8% 2nd and 1-4: 3.5 ypc; 112.3 61.6/38.4% 2nd and 10+: 4.4 ypc ; 89.4 33.3/66.7% 3rd and 1-4: 3.5ypc; 98.0 43.3/56.7% 3rd and 10+: 5.4yps; 77.4 14.4/85.6% Basically what you see is that in situations where a rush is expected ypc goes down and passer rating goes up and vice versa. The only exceptions is 2nd and long. If we look at splits by snap type. Shotgun: 4.9 ypc; 85.8 23.6/76.4% Under Center: 3.7 ypc, 95.5 Again where the defense predicts run there is lower ypc and higher passer rating and in passing situation there is a higher ypc and a lower rushing average Looking at splits by game win probability 0-19%: 4.0 ypc; 65.1 (4.1% interception rate) 27.7/72.3% 20-39%: 4.0 ypc; 76.4 (2.8% interception rate) 39.0/61.0% 40-59%: 4.0 ypc; 87.1 (1.7% interception rate) 40.7/59.3% 60-79%; 4.3 ypc; 101.2 (1.5% interception rate) 42.6%/57.4% 80-99%: 4.2 ypc; 120.0 (1.0% interception rate) 53.8/46.2% In this case rushing efficiency doesn’t change significantly. But as passing becomes more predictable defenses are able to shot down opposing passers. It is very interesting to me that nearly half of the total NFL interceptions are recorded (225 of 457) occurred when the win probability was below 20% anyway. Which makes me think that interceptions are more likely to be a symptom of losing, rather than a cause of losing. Looking at splits by score differential Leading 3.9ypc; 92.1 50.3/49.7% Tied 4.1ypc; 90.0 43.2/56.8% Trailing 4.3ypc/84.7 32.9/67.1% Again we see the same pattern, as passes become more predictable, defenses get better at defending passes. Conclusion The rushing game is only important in how it opens up the passing game. On the offense your goal is to get the defender to commit to the run sufficiently so that it can open up the passing lanes, but there is no significant benefit to being great at rushing. Abandoning the run will allow opposing defenses to set their most efficient pass defenses and will lower your passing efficiency. I strongly believe this is something we saw with Ryan Tannehill last year. For defense you only want to commit enough resources to slow down the run to the point where passing becomes predictable. We saw in Vance Johnson’s over the last 2 seasons enough instances where teams forced us to play a safety in the box and then took the top off the defense.
OK.. so if you have a set of numbers, let's say all points allowed in 2017 (all numbers in the column "PF" under the "Team Defense" table): https://www.pro-football-reference.com/years/2017/opp.htm you can calculate the mean and the standard deviation of those numbers. For 2017 mean is 347.5 and the standard deviation is 44.7. When converting to z-scores, the mean of 347.5 = z-score of 0. Each increment or decrement of 44.7 is 1 standard deviation above or below the mean, so it's +1 or -1 z-score. For example.. let's say a team allowed 420 points. How many z-scores is that? Calculate 420-347.5 = 72.5 and then calculate 72.5/44.7 = 1.6219. So 420 is 1.6219 z-scores below the mean (because this is defense). So when I said "average" defense that's 347.5 points allowed, and 347.5 - 347.5 = 0, so that's z-score of 0. And when I said 95 passer rating, note that 95 is 9.9 passer rating points above the mean of 85.1, and the standard deviation for passer rating is 9.884, and 9.9/9.884 = 1.0016, so 95 passer rating in 2017 is 1.0016 z-scores above the mean. That's why I used 1 and 0 for PR and PA in that equation. Hope that clears it up. These equations are really useful so you should learn to use them (and please use them!!). It makes the debate more rigorous. Yeah.. keep in mind these are points allowed over a season, not game by game.
That's the part that I have problems with because it's generalized and not team specific. The Eagles ran early and often last year busting open huge runs...it had nothing to do with the passing game if the other team couldn't slow them down. For years Miami has taken the same approach with a run-first philosophy, and even though it is to set up the play action in some cases, you just can't discount Drake going 80+ yards for a TD on a one-off play. Add to that the number of screens we threw to Landry the past few years (which are by definition run plays if he caught the ball at the LOS), and we'd probably be around 70% run/30% down-field passes. The run was just setting up a different variation of the run, so we could run again until we saw a shot at Stills downfield. To me it was a broken philosophy but it's the same path the Super Bowl champs took.
The other thing to remember about SD is the 68-95-99.7 rule Given a normal distribution: 68% of results are in the +/- 1 SD range 95% of results are in the +/- 2 SDs range 99.7% of results are in the +/- 3 SDs range. Roughly speaking 1/6 or 5 teams a season should be more than 1 SD above and 1/6 or 5 teams a seqson should be more than 1 SD below the average in a year. About 1 team in 40 should be more than 2 SDs above the average or about 4 teams in 3 years, and the same for below To be more than 3 SDs from the average should happen about once in 10 years, so in a 20 year span you can expect 1 team to be +3 SDs over and 1 team to be -3 SDs under. The other thing to remember is that random distribution is not even.distrubution. I.e. some years you may have a lot of variance and in others there may be little variance. You are not ‘owed’ a result.
The interesting question here is whether essentially all the effects you're seeing are due to score differential or win probability in the 4th quarter: And if that's the case, what is the effect of the running game on the passing game and/or vice versa in quarters 1-3? It's a real hard question to answer.. good observations though!
Okay, cool...you probably had the deviation in there all along and I just flat out missed it (or didn't understand it). I was trying to figure out if there was an additional formula to get the 44.7. I followed the math...I just didn't know how you got there. So for a 15 win season off defense alone, a team would have to give up an average of 13.3375 points per game (using the 34.75 mean from 2017 and jumping up three deviations, then dividing it all by 16). We can't calculate for a perfect season, of course, because it's only happened a few times and it wasn't SOLELY from defense. It pains me to say it that way though because there's only been 1 perfect season ever in my book. Likewise, holding opponents to 16.13 points per game would average 12.8 wins per season. An average of 18.9 PPG would earn around 9.6 wins....so a difference of less than a TD per week is what separates a potential Wild Card team from a Super Bowl contender. Again, that's all based on 2017 scoring numbers and comparing it historically.....but it's a good place to start. If Ryan Tannehill manages to put up his 95 qbr (expected 9 wins) and our defense can hold opponents to under 19 on average (expected 9 separate wins), that could be one heck of a season. I'm combining apples and oranges because I don't know the formulas for the advanced math like you and I realize that most of those would overlap one another.....but 12-14 is certainly possible if Tannehill gets hot early and we get to release the Kracken often.
I agree that while it is generally true for the league, it will vary team by team as to what is the right mix for that team. My main issue with the screens, going back to Lazor’s day has been that they became so predictable that opposing Ds shut them down too easily, so any benefit they had in getting Stills open was devalued by how inefficient the screen plays were. I actually saw some film of Paul Brown’s Browns running screen olays back in the 50s and 60s and thinking those type of screens looked more efficient thanthe ones we were running.
For me the important thing is that as rush% goes down pass defense efficiency goes up. The other interesting thing is that interception % is heavily influenced win probability, teams that are in the 0-19% range throw about half of all interceptions and teams in the 80-99% range throw very few., with the 21% to 79% band being roughly level. Thaat’s why I thought it important to loom at stats like down and distance as well as win probability and scoreboard posiition.
Isn't this intuitive? When you play from way behind, first you need to pass more and second you have to take more risks, which should lead to a higher interception percentage. Conversely, if you're likely to win the game, you can play more conservative which should lead to a lower interception percentage.
LOL, that's the old Tony Sparano philosophy- if we were up three with 14 minutes left in the first quarter, we were guaranteed two things- 1) There was a big old fist pump coming 2) We weren't throwing the ball until we lost the lead
The common wisdom is that turnovers cause losing, what I said is that sometimes it is the opposite. Similar things happen with the fumble rate too.
