So it turns out I was very wrong about my final score prediction. Good thing, because the actual game ended up as the most exciting of the season. The Hoos clung to their bowl hopes with a tense and methodical 87 yard drive ending in a fantastic catch by Jake McGee. The previously maligned Mike Rocco calmly1 drove his team down the field, completing a critical fourth down pass, drawing a holding call on a different fourth down pass,2 and finally connecting on the clinching touchdown.
The myriad of big plays and momentum shifts, especially in the fourth quarter, made me wonder: how influential were the various plays in the game? And how close were the Hoos to losing? Rocco's game-winning drive also raised questions about the sanity of continuing to employ the rotating QB system. How significantly did Rocco out-perform Sims, and does he deserve a full-time starting role once again?
This post will attempt to answer these questions through the use of Win Probability ("WP") measurements. A graph and discussion of the overall WP is followed by an analysis of WP changes under the guide of Rocco and Sims.
Win Probability, as the name suggests, is an attempt to measure one team's likelihood of winning the game as a function of score, time remaining, field position, and down and distance.3 The formula is an algorithm combining historical data of various game situations and outcomes. There are a number of assumptions4 inherent in this analysis. Nonetheless, the formula should create an interesting result and give us a baseline for comparing various events within the game. The following chart shows the WP for various events over the course of the game.
The data in the chart is plotted in the following graph.
Somewhat surprisingly, the winning McGee catch at +40% was not the single biggest increase in WP. The fourth down holding call took the Hoos from their lowest chance of winning, at 4%, to 49% with one minute remaining. The previous fourth down conversion on the Dominique Terrell reception earned +29% WP. Also of note5 was the Maurice Canady forced fumble at the Hoos 2 yard line earlier in the fourth quarter. The fumble gave the Hoos a 35% chance of winning, while the WP would have been only 16% had Miami been down at the 2 yard line.
The phantom safety was not the crippling event that it seemed to be during the game.6 The call cost the Hoos -6% WP.7 As an interesting comparison, after recovering Canady's forced fumble at their own 2, Sims took over the offense. The failure to earn at least a first down from deep in their own territory cost -9% WP.
The Rocco-induced gigantic spike at the end of the game is something we've seen before.8 So the question is, should London/Lazor give him more chances for late-game9 heroics?
[Rocco and Sims Win Probability Differences]
Win Probability Added, popularized in baseball statistics, measures one player's contribution to his team's WP over the course of a game. I'll compare Sims and Rocco in the Miami game by slightly altering this concept. I traced the changes in WP and separated them based on the various offensive drives led by Rocco and Sims.10 Despite capturing some non-QB-controlled elements, I thought this metric would still give a sense of which rotating QB did a better job of "controlling the offense." The below chart depicts changes in WP by QB and drive number.
Although Sims played well on one drive, he was not even close to matching Rocco's overall performance. Rocco was the far superior quarterback regardless of how this data is analyzed. For instance, some might contend that counting the last drive seems a little unfair, since Rocco alone had an opportunity to earn nearly +80 WP. Even removing that WP gain, Rocco led Sims +53% to +1%. Further considering that Rocco commanded two more drives than Sims, per drive, Rocco led +9% to +0.2%. Just to continue proving the point, let's only compare drives 3-10, where each QB led one touchdown drive among three unsuccessful drives. Rocco led +15% to +1%.
We'll see what happens in Thursday's game against North Carolina, but this game alone makes a strong argument for eliminating the QB-by-rotation and starting Rocco. Last season, he responded to a rotation with David Watford, earned the sole starting job, and went on to excel. Maybe he just needed similar motivation this year.
1 or at least he looked relatively calm from the Hill
2 I guess we can give him the credit for that
3 for those who read my blog previously and have already seen this paragraph, I apologize
4 Some assumptions: 1. The home team starts the game with a WP of 53%. This is a generally accepted home-field advantage. 2. The model assumes that only score difference, and not score, matters. But imagine a game where Team B is at the opponent's 40 with 2 minutes remaining in a tied game. Would you think they are more likely to score if the game is 35-35 as opposed to 0-0? Is this just perception or reality?
5 besides the actual touchdowns, of course
6 as for the dubious call: I was about as far away from the play as I possibly could have been and could not see HooVision. It was definitely a tough call, but does anyone actually know what the rule is about spotting intentional grounding? I would imagine it's supposed to be where Rocco's feet are when the ball leaves his hand, but on a replay that I saw he appeared to be in mid-air. So can he jump backward and claim the forward spot? In safety applications, it has nothing to do with "breaking the plane," right? Because the ball did appear to break the plane as he was winding-up. Confusing.
7 here's why it shouldn't be a larger drop (imo): either way, the Hoos need a scoring drive before the end of the game. Without the safety, a touchdown would give the Hoos a win, while a field goal would give the Hoos a tie, or theoretically a 50% (or 53%) chance of winning the game. With the safety, a touchdown would still give the Hoos a win. Although the field goal option was eliminated, it was only one scoring option anyway, and still would have only resulted in a 50% chance of winning.
8 Penn State earlier this year, and at least twice last season
9 or preferably earlier-game
10 with this method, the QBs are getting credit even if they're not directly advancing the ball; as in, for example, an RB rushing