Tannehill was more efficient and had two TD's. On the negative side, he was sacked once. So I'm not sure how a better completion percentage, two touchdowns and one sack makes him the lesser of the 2 QB's. You must lose like 30 cool points per sack or something.

That's why I never look at QB stats until after the fact- I just don't see value in them other than year-long comparisons.

 

Yea that's weird. Do they factor in 3rd down passing??

 

ESPN stats weighs heavy on where you turn the ball over and how well you do in the money downs(3rd down) and in the redzone. Tannehill threw a pick in the endzone. We didnt convert 3rd down or redzone too well.

Not saying I agree with ESPNs stat, but it simply has some parts of it I think are valuable.

 

ESPN's stat is in fact hot, smelly, rancid garbage, and always has been.

 

ESPN's QBR is a joke. They refuse to disclose the full formula for the stat and they factor in completely absurd things like how each quarterback's defense or special teams impacted the game.

Tannehill literally lost QBR after throwing the 75 yard bomb to Stills because the ensuing kickoff was returned for a touchdown.

 

Maybe it's worth pointing out what parts of QBR are valuable and what parts are.. as some have already said.. "garbage".

The part that's valuable is something called Expected Points (EP), which is defined as the average points scored, historically, for any NFL team in that same down, distance, game situation, etc.. Each play results in a change in EP, or EPA = Expected Points Added. EPA is how ESPN's QBR takes game context into account, and it's a totally sound approach statistically speaking.

The part that's garbage is what they do after calculating EPA, which is using a proprietary algorithm for dividing up EPA (dividing up credit) among the individual players. There's no statistical approach right now that can tell you how much player X contributed to an outcome, so they're incorporating hidden and subjective assumptions into that part. And it's proprietary because they know they could get ripped for specific assumptions they're making.

 

Well, in no world should a QB with 2 TD's in a win be ranked statistically worse than a QB with 0 TD's in a loss. That tells me that the formula doesn't take the most important factor into consideration- actual winning or scoring TD's. Because even if it determined that RT only had a 10% influence on both those scores, that's still 10% more than Mariota.

 

I guess while we're on the subject it's worth pointing out the main problem with EPA (or WPA). The basic idea is fine, except that the variances differ so much for different game conditions because of different sample size. For example, the variance for 1st and 10 in tiny because that occurs so often at so many different points on the field, but try 3rd and 12 or so and you get a mess with the data, meaning that there are so many different "valid" trend lines (or curves) you could fit that you really can't get a reliable estimate.

It's a great idea, but it's really hard to build on something where the reliability of the stat changes from game situation to game situation.

 

EPA (and thus QBR) takes into account whether you scored a TD or not (EPA is far more positive if you score). But what if the formula now says that the consequence of not scoring was a massive change in EPA (possibly over several plays) and that the QB was primarily responsible for that? Well.. now you can get weird results. So which outcomes they associate with a QB and how much influence they assume the QB had on those outcomes make all the difference in the world (all proprietary). But they do take scoring into account.

 

That's why I'm saying it's obviously flawed. RT had a higher completion percentage and two TD's. In no world should that be beaten by a lower completion percentage and zero TD's. I can see them hitting RT with a huge penalty for the interception in the end zone since that took 7 points off the board...but that should never weigh more than actually winning the game. That deep bomb to Stills was freaking beautiful.

If that's what happened, the formula is just broken.

I'm heading to bed...I'll argue with you more tomorrow. =) Have a great night!

 

Well.. that's the problem isn't it? No one knows precisely what they're doing.

The way I view QBR is that as a statistic it's trash, precisely because it's not a statistic. Stats are logical inferences from data and adding in all kinds of subjective assumptions means you're not creating a statistic.

But as a "model" I wouldn't diss QBR too much. It supposedly has a higher correlation to win% than passer rating (I say supposedly because I've never bothered to double check.. I abhor it too much to even put it in my database lol) so if all you're interested in is higher predictive power, regardless of assumptions, then it's fine I guess.

And from that perspective, let's remember that what matters is the correlation, not individual cases. I mean, you'll find cases where a QB performs great and have a low (traditional) passer rating, or vice versa. So what matters is the correlation.

 

Most importantly, no stat should ever, ever take into account wins and losses. If it does, or tries to predict that, then it's more than garbage, its offensive.

In the past when they've shown some aspects of the formula, ESPN showed that RT took huge hits to his QBr because his ol was awful and he was sacked a lot. useless.

 

I'd say precisely the opposite. If you can show near zero correlation between any stat and win percentage, then there's no reason whatsoever to care about that stat, because it doesn't affect wins and losses.

So it's imperative for every football stat used to either directly correlate well with win percentage, or to correlate better with win percentage than a competing stat (if you have to choose between two).

 

That's what I meant as well. It should never give "bonus points" for winning, but if the top 5 quarterbacks are all on losing teams then the stats aren't identifying the right things to be used as predictors. After all, the whole point of QBR is to identify the most successful quarterbacks.

 

Well, I will disagree with every fiber of my being until the day I die, lol. I strongly, strongly, believe that in a team sport, stats should show what the individual accomplished, and allow him to be easily compared to others at his position, regardless of how his team performed.

Traditional qb rating is obviously flawed in some ways, and is often now seen as a team stat. But all attempts to replace it seem worse to me.

In a perfect world, a stat would be able to show how a player did regardless of, or in spite of, his teammates. But that's easier said than done. And at the end of the day, most importantly, I never connect players and victories.

 

I think you have to step back a bit and ask yourself why you want to look at stats in the first place. Let's say you're talking to a an NFL coach and one person says to the coach that stat X suggests he should pick QB A instead of QB B in the draft, while another person says stat Y suggests he should pick QB B instead of QB A.

How do you decide which stat is more relevant? Well..suppose stat X is completely uncorrelated with winning while stat Y is highly correlated with winning. Obviously the coach should use stat Y as the basis of his decision, because the goal is to win!

This is exactly why moneyball had some success in baseball: some of their stats actually improved win probability. Point is.. if you're going to use stats to suggest how the team (or individual) make relevant improvements, you want to make sure the stat correlates well with win%.

I can see your side of the argument ONLY if you don't want to use the stat to suggest ways to improve the team's chances of winning.

 

This

But according to ESPN
Tannehill: 34.3
Mariota: 37.4

Is all you need to know that the stat is not logical and not weighted properly.

 

Why? Just because a team loses doesn't mean the QB is playing poorly. Actually, I think that's exactly why ESPN came up with this stat. They want to give a rating of the play independent of the win and loss. At least that's how's it seems to me. And it makes sense, in that respect, for instance, looking at the discrepancy between Mariotta and Tannehill yesterday.

 

I am simply a fan. Never a coach, a scout or an executive. I just want to see which players did well on a given day, or year, or career. And it interests me in the sense of history, it even novelty, when something is rare or special. And unlike some, I do still like counting stats, because they show that a player continued to be effective over a length of time. I like stats for fun.

 

News flash: ESPN is the TV version of clickbait. Their business model relies on creating controversy to drive viewership. As such, their content is pure garbage.

 

For one game, I agree with you. But these are annual statistics that we add to each week....that helps minimize the terrible/awesome games to give a true average of what a QB's made of.

For example, Rodgers QB rating last night was a 130.7, which in my opinion is WAAAY too low for the performance he put on while injured. In anyone else's book, he was better than the "perfect" 158.3. And GB almost lost that game- it wouldn't have changed his stat line one single bit. So it is important to have more than "just a number" and I appreciate ESPN trying to do that; I just don't see how RT with 2 TD's (one a 75 yard bomb) doesn't rank higher than Mariota who had two picks and no scores in 13 throws. There's something in that formula that's not correlating with what we saw on the field.

Like I said-

RT > Mariota in accuracy %
RT > Mariota in TD passes (2 to 0)
RT > Mariota in total completed passes
RT > Mariota in pass/int ratio

Mariota > RT in 3rd down conversions
Mariota > RT in sacks (0 to 1)

Unless I'm missing something critical here, t's very hard to make the case that Mariota outplayed Tannehill. Yet that's what ESPN is showing. I'm guessing a big part is ball placement and that should be a factor, but I have a feeling it's way too big of a factor given the outcome of the game.