Writers of Pro Football Prospectus 2008

06 Aug 2005

FO in the NYTimes: Third Down is the Charm for NFL Turnarounds

The New York Times gives me space to expound upon FO's new favorite theory, third down as a future indicator. If you've been asking yourself, "What would the San Diego chapter of Pro Football Prospectus 2005 look like if it were only 800 words long," well, here's your answer! (Free registration required)

Posted by: Aaron Schatz on 06 Aug 2005

14 comments, Last at 14 Aug 2005, 4:25pm by Randy


by Moses (not verified) :: Sat, 08/06/2005 - 10:17pm

Aaron, don't mean to throw a spanner in your works, but what about the yang to the 3rd down yin. i.e. what is the frequency of the 3rd down yds/play being the correct indicator vis 1st & 2nd down being the correct indicator?

Did you measure both sides, or did you find something interesting on the other and run with it...?

by Ted (not verified) :: Sun, 08/07/2005 - 7:12am

This thread has got nothing to do with what I want to say but I don't know where else to put this. I see the FO crew will be on Cold Pizza soon. You will have a fan for life if you can mention, on the air, "The road from Bristol" (For those who haven't found this site, my HTML sucks but try googling that) and tell Woody Paige that he is more loathed than even Joe Theissman. If you want to console Paige you can tell him that at least he has nothing on Skip Bayless.

by Aaron (not verified) :: Sun, 08/07/2005 - 11:04am

Yes, that's seriously off-topic -- I remind people if you want to send a personal message to me or any other FO writer, esp. an off-topic method, we have the new CONTACT US choice on the top bar menu. (Not that Road to Bristol isn't hilarious, I just don't want it discussed in this thread.)

by mistamaxwell (not verified) :: Mon, 08/08/2005 - 8:53am

Why, then, would third down production be so volatile?

by Starshatterer (not verified) :: Mon, 08/08/2005 - 9:41am

Why, then, would third down production be so volatile?
One reason is sample size: there are more first down plays than second down plays, and more second down plays than third. So small variations affect third down performance much more than all plays as a whole.

by SJM (not verified) :: Tue, 08/09/2005 - 3:57pm

Where can I get stats for all 32 teams to compare yards per play on 1st and 2nd downs to yards per play on 3rd down?

by Jeff F (not verified) :: Wed, 08/10/2005 - 1:08pm

SJM - I know you can use the stats on nfl.com or espn.com (One or the other) to look at detailed splits, so you can see how an individual QB does on 1st down, second down, when ahead by 7 points, down by 14, or when their son has an epilleptic fit, however, they don't have similar stats for the team - and, those stats are just yadage numbers/completion percentages, which often don't tell the whole store, as we all know.

by someone (not verified) :: Sat, 08/13/2005 - 4:03am

This is an interesting theory but needs more explanation. What is the evidence to suggest that over time teams tend to play as well on third down as they do on first and second down? What exactly does that statement mean anyway, yds per play, relative ranking to the rest of the teams in the league, or something else?

Pulling out a few examples of teams who drastically improved is fine, but are there any examples of teams who are good on first and second down but miserable on third down, and who don't improve year after year? You need to split out what proportion of teams who were bad on third down actually improved versus those that don't. And also see if there is a consistent set of factors - go-to TE like Gates coming through in San Diego, reliable slot WR signed etc - causing this improvement. Otherwise it's just a cute way to push a team like the Seahawks who you like anyway this year, while sounding scientific.

by big_adventure (not verified) :: Sat, 08/13/2005 - 12:46pm


If you spend a little more time searching on this site, you'll find more evidence pointing to what you are asking for. The Times gave Aaron 800 words, not 80K. :) He mentions his research conclusions in other places and threads, and in more detail. The gist is that teams typically bounce back towards their "expected" level of production.

Of course, if you don't believe Aaron did that research, and that he made it up to support a FO Seahawk love-fest, I recommend providing YOUR sources for that.

The point he makes in other places is that third-down performance fluctuates far more wildly than performance on first and second downs. As #5 mentions, sample-size likely has a lot to do with this.

The Dolphins "went for it" on 1022 plays last year (passes, runs, sacks), of which 232 were 3rd down and 16 were 4th down. This leaves 774 on 1st and 2nd down. Assuming an exactly even breakdown (very unlikely, of course, even the 'fins got a few 1st downs on 1st), that leaves 387 1sts and 387 2nds. You would expect to see greater consistency, test for test, in any measurement you took 72% more times. Thus, if the 'fins averaged (you have to consider situation, of course, which Aaron typically does...) 5 per play on 1st and 5.1 on second and 6.3 on 3rd, which one seems like the outlier? Which one is likely to fall back to earth the next season? Once you compare every team-season since 1998, you get about 210 team seasons. It's still not a truly massive sample-size, but it's pretty good and, if you can find an r-sq of greater than .5 in that time, I think you've shown pretty convincing correlation.

It's possible that EVERY team that improved added a "go-to guy" in that period, and that every team that got worse lost one, but the only people that are going to believe that still believe that there are clutch hitters in baseball. ;)

I'm not saying it's impossible, and I certainly haven't carefully reviewed the work in question, but I'm not going to believe that team X is consistently, unexpectedly good or bad on 3rd down just because 8 games of stats or John Madden tell me that they are.

The Chiefs offense went 4-for-14 on 4th down last year. The Bucs went 4-for-6. Do you REALLY think that this is an accurate view of the relative capabilities of these two offenses? Is there ANYTHING ELSE TB did on offense as well as the Chiefs in 2004? Or do you think this is just a skew caused by insufficient sample size? If your life was on the line, no pun intended (well, maybe mildly), and you had one shot to move a football 1 yard, or 3 yards, or 8 yards or 25 yards to avoid DEATH, would you choose the Chiefs' offense or the Bucs'?

My money, and my life, would be on Green/Gonzalez/Holmes - without a doubt. No matter what one small-sample-size stat said.


by big_adventure (not verified) :: Sat, 08/13/2005 - 12:53pm

Sorry for the double-post -

I should point out that the play # data for the 'fins are real to the best of my abilities to calculate: they were calculated by adding up all of the pass/run/sack plays.

However, I made up the yardage/play numbers to make a point. It still makes the point quite well.


by someone (not verified) :: Sat, 08/13/2005 - 10:29pm

"If you spend a little more time searching on this site, you’ll find more evidence pointing to what you are asking for."

Oh ok, I should spend time sifting around various articles on this site to find justification for this theory, as opposed to a place where it is propounded as FO's favourite theory of the moment. Scuse me.

I'm aware a smaller sample size tends to lead to greater variance. As the sample size for third downs remains smaller every season however, why should there be a reversion towards the performance on first and second down? - isn't it as likely that another outlying stat is recorded for third down? As i said, i would like to see evidence to suggest that this reversion to expectation actually occurs. So 210 seasons have been analysed? Tell more - how many of those seasons exhibited this tendency for worse third down performance, and what proportion of those exhibited a bounceback, and to what degree.

"It’s possible that EVERY team that improved added a “go-to guy� in that period, and that every team that got worse lost one, but the only people that are going to believe that still believe that there are clutch hitters in baseball."

The way this type of analysis actually adds value is if it is first substantiated with hard evidence (i.e. not just plucking two or three exmaples from the last ten years and expecting anyone else to fully accept an 800 word article as gospel) and then the factors causing this can be to a certain extent explained. Unless all teams exhibiting this tendency bounce back uniformly and the theory can be utterly relied upon, examination of the factors causing the bounceback in particular situations and not in others is highly relevant.

"Of course, if you don’t believe Aaron did that research, and that he made it up to support a FO Seahawk love-fest, I recommend providing YOUR sources for that."

I'm not here to take peoples' word on this stuff. If he has performed the analysis, great lets see it, even as an appendix to an article, I would love to have an accurate predictor for bounceback years. I'm not necessarily questioning the validity of it - I don't know - I'm requesting further information along the lines of how any reasonable scientific study would be assessed.

by Aaron (not verified) :: Sat, 08/13/2005 - 10:52pm

There is more explanation in the book, a long extended essay in the San Diego chapter. There is only 800 words of explanation in the New York Times because the KEEPING SCORE column is 800 words, give or take a few. When you write for the New York Times, you follow the rules of the New York Times.

by someone (not verified) :: Sun, 08/14/2005 - 12:16am

Fair enough

by Randy (not verified) :: Sun, 08/14/2005 - 4:25pm


Could you please answer post #6. It would be a useful tool to track what team is under/over performing.