Tuesday - Jan 22, 2019

Home / Uncategorized / Welcome to the Stat Lab

Welcome to the Stat Lab

Fact: The average American is about 5 feet tall

This is what you get if you look at the Census data the right way.  If you want to average out Americans another way, take what makes about half the country boys, then take what makes about half the country girls, and divide by two.  However interesting a mental image this might make, this is not a helpful statistic (if you happen to be one of these kind of “Average Americans” please stop reading this article and call your nearest day time talk show host immediately).


Here in The Lab we’re going to take football data and commonly held ideas in fantasy football, tie them up, strap them down, and wring them into submission.  We’ll subject them to experiments until they confess to numerical lies like this one that help you lose, and reveal meaningful secrets that just might help you win.  More importantly, we’ll show how to figure out which is which – and this is how we begin!

What can statistics do for you?

In fantasy football stats mostly project performance, describing the probability of what might happen under certain circumstances.  We look at the past, look at the present, and if we find a meaningful relationship – at most! – we say that something will probably happen.  Projections are definitely wrong if you expect them to be exactly right.

For example, if I say: “Peyton Manning will throw 32 touchdowns in 2007”

…I’m really saying:  “Given what I know today, I’m 95% sure that the average of the most likely number of touchdowns Peyton Manning will throw is 32.”


Let’s break this down.  First, I’m describing a range, something like 30 to 35 TDs, and just giving you the number in the sweet spot.  Remember this when reading those long cheat sheets.  Second, I can only talk about what I know today.  Answers to certain football questions in April are meaningless in August.  Finally, you don’t know what I think I know or how I know it! In other words, you don’t know what I’m assuming or how meaningful my assumptions are.

About those cheat sheets…

When you’re looking at a bunch of different sources for guidance, there are few things to keep in mind.  To get the most value, forget the numbers and look at the rankings.  When you find large differences this means there may be a bias, someone has broken out the beer helmet too early, or one thought of something the other guy didn’t and you should try to figure out what they were thinking. 

Also be careful when you’re comparing exact numbers from different sources.  Big sets of projections are prone to inflation or deflation due to formula issues, author bias, threats from large flightless birds, and assumptions about the league as a whole.  If Manning’s TDs are lower on one list, have a look at Palmer, Tomlinson, and other top players on that list to get a sense if they’re just assuming less offense in the league altogether.

When you assume…

Big, honking, total league statistical projections start with assumptions.  A common assumption is that a good offense will play better against a bad defense than it will against a good defense.  The result you expect is somewhere between what the two teams usually do to other teams.  This is the kind of idea used in strength of schedule analyses.

This chart compares the Colts’ passing completion (light blue) to their opponents’ season average completion allowed (black).  It also shows the logically expected outcome: what’s between what the Colts usually did and what their opposition usually allowed (orange-ish).  It also shows the linear average for the Colts’ completion (dashed blue) because graphs look naked without trend lines.


As defense completion allowed goes up the Colts generally followed, and this is what you’d expect, but just look at those humps!  The Colts performed as expected only about 70% of the time, above or right around the orange line.  On the wrong side of that orange line, they average to about 2 more incomplete passes per game, but this understates the total flops you see on the left.  Actual performance isn’t very consistent either, as you can see in the big peaks and valleys — though 60% seems to be the breakout point for an efficient ride on Air Indy.  This is encouraging, but with a little mathemagic you learn that this relationship accounts for just about a third of the entire story.  Now even if you don’t know your coccyx from your correlation coefficient (or your tail bone from your elbow) these results shouldn’t thrill you.


We’ll go deeper in the future but look at what we have now:  Here’s a relationship that seems obvious to any football fan and should be near perfect when you look at one the best passing attacks in the league, but you find nothing overwhelming.  This should be as good as it gets, and we only got “not awful.”  Bummer.

End of article!  Condemn the lab!  Banish the geeks! Go with your gut!

Not exactly.  The completion example just illustrates the earlier point: Projections are definitely wrong if you expect them to be exactly right.

When you have an inconclusive result you can look deeper or you can say “well, that’s interesting” and throw it in the waste bin because it doesn’t prove anything.  You can also use this idea, but only if you treat it with the respect it deserves.  Here that respect comes from the honesty in saying what it has, what it doesn’t have (home/away, weather, injuries, etc.), that it’s only been right about 70% of the time, only tells a third of the story, and if you try to predict exact outcomes based on this they won’t be all that good.  If you get a claim without this sort of disclosure, proceed to the waste bin – or post it to the Article Discussion forum if it’s interesting, and we’ll add it to the hit list.

Now what?

So now that we know what to expect from stat analysis and from our experiments and a tidbit on when the laser sight goes on to the Colt Cannon, we can turn our tools to our victims so patiently waiting since the first paragraph, and begin to extract the information we want.  In the coming articles, we’ll look at  how to use projections, and take on a few common fantasy football ideas like “The Third Year WR Breakout,” “The Second Year QB Breakout,” and, my personal favorite, the “RB Touches/Carries Breakdown” that predicts doom the season following some large number of carries or touches.  We’ll also take a closer look at the Colts passing game, which is on the verge of a once-in-a-decade fantasy milestone this coming season.

Finally, I’d like to acknowledge the number of regulars in The Shark Tank who know all these rules of analysis and play by them.  If you don’t know them yet, stop into The Tank, and you’ll know them when you see them.  You’ll also know me when you see me once some other geeks get my picture up. 😉

About Fantasy Sharks

FantasySharks.com began in 2003, disseminating fantasy football content on the web for free. It is, or has been, home to some of the most talented and best known fantasy writers on the planet. Owned and operated by Tony Holm (5 time Fantasy Sports Writer Association Hall-of-Fame nominee,) Tony started writing fantasy content in 1993 for the only three fantasy football web sites in existence at the time.