4th Quarter Comeback Analysis

[vince]

Was a scale established yet? (i.e. are there NFL norms and a range) This is too much number salad to sift through. But if the scale is from -100 (Grossman, Tagge) to +100 (Elway), then Rodgers and Brady are no different; it's all a wash.

The "best" score in this model would be 3.0 - if 100% of the QB's comebacks were successful and on the last possession of the game. The "worst" score would be -3.0 - if 100% of the QB's comeback attempts were unsuccessful, meaning the team never took a 4th Q lead in any situation in which it was down by one score or less in the 4th quarter.

Sharpe's comment about the impact of the weighting is true, which is the main reason I'm a bit skeptical about it. I'd tend to agree that, to the extent there's a value to "clutchness," the later in the game the comeback occurs the more "clutch" it is, but it incorporates a penalty for what some might argue is a preferred scenario. Just be good enough to go ahead earlier and stay ahead rather than letting the opponent hang around until it comes down to one possession.

That goes back to the simpler and arguably more meaningful 4th quarter "success vs. failure" measure. It just depends on whether - and how - you want to incorporate the idea of additional "clutchness" into what already filters out the first 3 quarters of play and non-close games entirely.

[ThunderDan]

Okay, so adjusting the formula to weigh Scenario 4 the same as Scenario 2 (2 points):

Rodgers career 4QCR = -0.2
Rodgers since mid-2010 = 0.95

Brady 4QCR = 0.60

So it looks like Rodgers' numbers are fantastic, but the Packers have suffered this type of loss in 6 out of 19 comeback opportunity games since mid-2010. Namely:

2010 MIA
2010 @ATL
2012 @IND
2012 @MIN
2013 SF playoff
2014 Game that shall not be named

In all of those games, the defense displayed total "unclutchness". In four of them, there was nothing the offense could've done to avoid what happened (scored the necessary TDs). In two of them, the offense had to settle for FGs.

Didn't Crosby clank the goal post on a game winning FG in regulation against MIA in 2010?

[ThunderDan]

Didn't Crosby clank the goal post on a game winning FG in regulation against MIA in 2010?

Nope, that was the Washington game.

[mraynrand]

The "best" score in this model would be 3.0 - if 100% of the QB's comebacks were successful and on the last possession of the game. The "worst" score would be -3.0 - if 100% of the QB's comeback attempts were unsuccessful, meaning the team never took a 4th Q lead in any situation in which it was down by one score or less in the 4th quarter.

Sharpe's comment about the impact of the weighting is true, which is the main reason I'm a bit skeptical about it. I'd tend to agree that, to the extent there's a value to "clutchness," the later in the game the comeback occurs the more "clutch" it is, but it incorporates a penalty for what some might argue is a preferred scenario. Just be good enough to go ahead earlier and stay ahead rather than letting the opponent hang around until it comes down to one possession.

That goes back to the simpler and arguably more meaningful 4th quarter "success vs. failure" measure. It just depends on whether - and how - you want to incorporate the idea of additional "clutchness" into what already filters out the first 3 quarters of play and non-close games entirely.

that helps. but it still needs a range, that is more data points - other QBs. Maybe TH87 should sell the App to footballoutsiders or footballfocus and earn tens of dollars.

[th87]

Interesting discussion. Correct me if I am wrong, but if you score early and often in the 4th quarter, you get lower rating under this metric. In the extreme case, a QB that scores 21 points in the 4th looks worse than one that struggles the entire quarter and throws one late TD.

I have no idea if that matters in the Rodgers vs Brady comparison, but I think the whole concept of "clutch" is suspect as a QB metric. This is still one of the best breakdowns.

Given that the metric measures one possession games in which the offense is behind, situations in which teams are purely trading scores all quarter, would be valued by what happens on the offense's last chance to take the lead (if it's there). If in this game, the offense scores (and wins) on the last possession, they are awarded 3 points. If it turns out that they needed one stop after their last score when time remaining is over 1:00, they get 2. And if on the team's last chance they don't score at all, in this situation, they'd get -1, because while they did get the lead at some point and lost it, they still had a chance to get that final score.

In the situation you describe, let's say the offense scores 21, thus going up by, say, 18. This would be dependent on defensive stops then, and while the offense did do an awesome job of scoring more, they are now out of the "come from behind" scenario right after their first score. They are awarded 1 point. The metric more values if-the-offense-doesn't-deliver-we're-screwed type situations, which means more pressure on the QB/offense, and thus, the more offensive "clutch" necessary. It's not perfect, certainly, though it still positively rewards the offense in that situation, but not as much as a "chips are down" situation, which is really what I attempted to measure.

[th87]

that helps. but it still needs a range, that is more data points - other QBs. Maybe TH87 should sell the App to footballoutsiders or footballfocus and earn tens of dollars.

If you're donating, I can make this my full time job.

(Working on Wilson now, but busy)

[MadScientist]

Here's what jumps out at me:
4. Losses where lead or tie was achieved, but the defense gave up the lead/tie with no chance to come back.
-Rodgers 7/35 (20%)
-Brady 3/68 (4%)

5. Losses where a lead was taken, but the lead was subsequently given up by the defense, but there was still enough time to win (> 1:00).
-Rodgers 4/35 (11%)
-Brady 3/68 (4%)

and the corollary
2/3 Wins where one or more defensive stop preserved the victory after a go-ahead score.
-Rodgers 7/35 (20%)
-Brady 22/68 (32%)


NE's defense doesn't choke like the Packers' defense too often does.

[sharpe1027]

Given that the metric measures one possession games in which the offense is behind, situations in which teams are purely trading scores all quarter, would be valued by what happens on the offense's last chance to take the lead (if it's there). If in this game, the offense scores (and wins) on the last possession, they are awarded 3 points. If it turns out that they needed one stop after their last score when time remaining is over 1:00, they get 2. And if on the team's last chance they don't score at all, in this situation, they'd get -1, because while they did get the lead at some point and lost it, they still had a chance to get that final score.

In the situation you describe, let's say the offense scores 21, thus going up by, say, 18. This would be dependent on defensive stops then, and while the offense did do an awesome job of scoring more, they are now out of the "come from behind" scenario right after their first score. They are awarded 1 point. The metric more values if-the-offense-doesn't-deliver-we're-screwed type situations, which means more pressure on the QB/offense, and thus, the more offensive "clutch" necessary. It's not perfect, certainly, though it still positively rewards the offense in that situation, but not as much as a "chips are down" situation, which is really what I attempted to measure.

I guess the main problem, more precisely, is that there is no difference in some of the same situations except what happens next. So, a QB can do precisely the same thing, in precisely the same pressure situation, but get different points depending on what the defense and opposing team do.

[th87]

I guess the main problem, more precisely, is that there is no difference in some of the same situations except what happens next. So, a QB can do precisely the same thing, in precisely the same pressure situation, but get different points depending on what the defense and opposing team do.

After equating Scenarios 2 (defense makes one stop after score) and 4 (defense allows score with no time left after score), there are no more situations (I don't believe) in which the QB is doing the same thing but getting different point values. If a go-ahead score requires more than one defensive stop to preserve, that means there was some higher amount of time left, which to me, means that it wasn't *as* high of a pressure situation as a last chance drive might be.

I'm open to modification suggestions though.

[sharpe1027]

After equating Scenarios 2 (defense makes one stop after score) and 4 (defense allows score with no time left after score), there are no more situations (I don't believe) in which the QB is doing the same thing but getting different point values. If a go-ahead score requires more than one defensive stop to preserve, that means there was some higher amount of time left, which to me, means that it wasn't *as* high of a pressure situation as a last chance drive might be.

I'm open to modification suggestions though.

Situation #1:
Score with 5 minutes left, defense gives up one 4:50 drive and no score. Offense kneels down to win. Result, 2 points.


Situation #2:
Score with 5 minutes left, defense holds with a 1:00 drive. Offense score again. Defense holds on next drive (or not). Result, 1 point (two defensive holds).

Both scores at 5:00 are identical. Situation #2 is at least as clutch as situation #1.
Maybe if the metric was per drive/opportunity rather than per game? Otherwise, the number of defensive stops can differ for identical go ahead situations.

[th87]

Situation #1:
Score with 5 minutes left, defense gives up one 4:50 drive and no score. Offense kneels down to win. Result, 2 points.


Situation #2:
Score with 5 minutes left, defense holds with a 1:00 drive. Offense score again. Defense holds on next drive (or not). Result, 1 point (two defensive holds).

Both scores at 5:00 are identical. Situation #2 is at least as clutch as situation #1.
Maybe if the metric was per drive/opportunity rather than per game? Otherwise, the number of defensive stops can differ for identical go ahead situations.

Maybe make it a 2 point value if the subsequent defensive stop results in another offensive score that now gives more than a one possession lead. This way, only one defensive stop was "needed".

[sharpe1027]

Maybe make it a 2 point value if the subsequent defensive stop results in another offensive score that now gives more than a one possession lead. This way, only one defensive stop was "needed".

That doesn't take into account differences due to the other team's drive time eating up more or less clock. How about weight the opportunities based on time left at the start of the drive? If you really wanted to get fancy, you could also weight based on starting field position using the expected score value from each starting position.

[Smidgeon]

The problem with advanced metrics is that there's always something else to consider. If we really wanted to go complex, each QBs "clutchness" would be weighed against the opposing QBs "clutchness" in each game. Is the defense trying to hold against Elway, Brady, or Manning? Or is the defense making a stand against Quinn, Gabbart, or Russell? That would affect things too.

Mind you, I'm not suggesting actually doing all that crazy work. But just highlighting that there's a point where you just have to summarize things.

Otherwise, very interesting statistics.

[pbmax]

There is also a problem of definition. Clutchness is such an broad excuse to dismiss and credit performance for reasons that may have mostly to do with outside factors that its easy to dismiss.

However, the "hot hand" concept, that streaks are not just statistical fluctuations, has been gaining ground after initially being dismissed by data types.

So you have to know what you are describing in order to be able to look at it analytically. Its a reason, as we have discovered in this thread, that 4QC is a problematic concept and gauge. Even the guy who started all this, Scott Kacsmar, admits we haven't narrowed down what exactly it is telling us about teams and the QB.

[th87]

That doesn't take into account differences due to the other team's drive time eating up more or less clock. How about weight the opportunities based on time left at the start of the drive? If you really wanted to get fancy, you could also weight based on starting field position using the expected score value from each starting position.

Yeah, that would be good, but more work than I can afford! I have to draw the line somewhere, and it would still be illustrative IMO, but with the understanding that it isn't perfect.