That other misleading graph

Last night ESPN ran the story that Brett Favre was staying retired.

On the 11 PM EST edition of Sports Center, they ran a graphic showing that over the last two years Favre had thrown 13 interceptions when his team was trailing by 7 or fewer points in the fourth quarter, Ben Roethlisberger (boo!) had thrown 7, and no other quarterback had thrown more than 6. (Sorry, I can’t find a video or screen cap of this, but don’t worry – it’s not the graph I’m blogging about).

I’m not a Favre fan, but I felt that was an obnoxiously misleading stat. Favre threw 13 interceptions when his team was trailing by 7 points or less in the fourth quarter and Roethlisberger threw 7? Well, how many times was each in that situation? Maybe Favre’s team were in that situation 20 times over two years, and Roethlisberger’s teams only 9. Maybe a quarterback in the 6 or fewer group only was in that situation a handful of times and threw multiple interceptions? Who has the ball? For how long? And what abou those desperation, last second, “Hail Mary” bombs thrown as time expires – should they really be counted?

As I said, I’m not a Favre apologist, but that was clearly a statistic lacking complete context.

So, I thought I might write about it, and in the process of searching for that graphic, I found this one.

Chances of winning an NFL game

Now, I found it via Google’s Image Search, and didn’t really look at the article that sparked it. I (foolishly… I’m smart enough to know better) didn’t read the article at all, actually, I just went by the key and the heading and kind of glanced at the graph.

Almost immediately, I felt like something wasn’t right. I noticed that the +3 and +7 score differential lines (indicating that the team was ahead by 3 or 7 points at that instant) spiked toward 100% chance of winning as time expired in the game. Makes sense… and even the +1 line went toward 100% as time expired. The -7 score differential line headed down toward a 0% chance of winning. However, the -3 line behaved unexpectedly -it didn’t drop toward zero percent as time expired, it went to about 20% with a local minimum at about the 6 minute mark. Strange.

Upon closer inspection, I started to question why the lines weren’t mirror images (with the line of symmetry at 50%). You would think that if the team leading by 7 points with 30 minutes left in the game will win the game 75% of the time, then the team losing by 7 points at that instant only wins 25% of the time. But that’s not what this graph indicated.

I was puzzled and somewhat irked as I consider myself kind of an expert at mathematics and statistics. I contemplated it for a bit, but couldn’t really rationalize it. In fact, part of me wondered why they even showed the -1, -3, and -7 lines… they shouldn’t be necessary as each of their values should have been the value which adds to the +1, +3, or +7 value to make 100.

Ah, but let’s get back to the thing I didn’t do that I really should have… read the actual article.

The graph comes from the Advanced NFL Stats blog, and it is clearly (?) indicated within the article that the graph “plots the WP [win probability] of the team with possession of the ball.”

I had erroneously assumed that +7 was the team winning by 7, and -7 represented their opponent. But the graph doesn’t show data for both teams in the game. It only shows the data for the team who is playing offense at that particular moment of the game. Yes, I know! (read in Craig Ferguson voice)

So, that changes everything, and I feel foolish.

I still feel like the graph’s a bit wonky and subject to misinterpretation, but I’m excited to explore the Advanced NFL Stats site (as I do LOVE that sort of thing) and I’m proud to haveĀ  figured out my mistake (go me!).

Contributed by: Scott Copperman


Leave a Reply