Simply put, streaming in fantasy football is playing someone – typically off the waiver wire – based on matchup rather than true skill. Defenses are the perfect option to stream because their fantasy output directly relates to the opposition’s offensive play. That’s why you constantly see team defenses on and off the waiver wire throughout the fantasy season.

Another less obvious reason that streaming works for defenses is because a typical lineup starts just one of them. Anytime there’s a “onesie” position in fantasy football (typically quarterback, tight end, defense and kicker), there’s a small demand for the position. Those players aren’t needed as much as, say, a running back or receiver. And when that’s the case, said assets are easily attainable and accessible off the waiver wire; fantasy owners just don’t need them that much.

Onesie positions in fantasy are intrinsically devalued. And really, this standard structure of fantasy football is the sole reason the Late Round Quarterback strategy exists. If we started three quarterbacks in a standard lineup, then you better believe I’d draft a quarterback early; only 32 are starting in the NFL in a given week, and in a 12-team league, that’d be less than the number of hypothetical quarterback starters.

That’s not the case though. We start one passer on our usual squad, so quarterbacks, from a supply and demand perspective, aren’t all that worthwhile. (Trust me, there’s a lot more to the strategy in my new book.)

Yes, matchups and onesies are the two main reasons we should stream players. Why not play a dude who’s free (off the wire) and has a favorable matchup (against the Bills)? It’s an effective strategy. And it’s made even better when you can accurately predict the streamed position’s output.

For some detail, take a look at this snippet from The Late Round Quarterback: 2013 Edition:

At quarterback, the opposite is true. They have high floors. In other words, you can slot a guy like Mark Sanchez in your lineup against a weak defense, and you know he’ll score at least five to ten fantasy points. You know he’s going to be touching the ball on every snap. He’s a quarterback. His ceiling may be lower than Aaron Rodgers, but because he’s throwing the ball 30 times per game, there’s predictable opportunity. You don’t get that with a player like Ronnie Brown.

If you’re a Twitter person (if you’re not, get on it!), you may remember the hashtag #FreeSpiller that was used throughout the 2012 season. Chan Gailey, the coach of the Buffalo Bills at the time, kept Spiller’s carries per game relatively low for such a talented back, which frustrated fantasy owners. The hashtag continued to increase in popularity as more and more fans clamored for Gailey and the Bills to feed C.J. the ball.

Guess what? There’s no #FreeManning. There’s no #FreeRoethlisberger. Quarterbacks are already free. There were 21 of them who averaged over 30 attempts per game last season. That, in the end, gives them all the opportunity in the world to score points. 

As I noted, quarterbacks have predictable opportunity. We know, with a higher confidence than any other position in fantasy football, that a quarterback will produce closest to his expected output. Math, our trusty friend, can help show this.

If you recall from any statistics course you’ve ever taken, the standard deviation of a data set is “a quantity calculated to indicate the extent of deviation for a group as a whole.” Standard deviation shows us variance in data. It can help solve our constant struggle of predictability.

But the issue with looking at only raw standard deviation metrics from one group of data to the next is that the averages of each set could vary greatly. For instance, quarterbacks score more points in fantasy football than any other position, but they still have games where they score only a few points. Because of this, their averages are higher than any other positions, but so are their standard deviations.

In order to compare standard deviation metrics across multiple data sets, we should use the coefficient of variation. The COV (or CV) takes the standard deviation of a group and divides it by the mean, or average. The outputted number gives us a ratio that can be used for data group comparisons. This number (at least to you and me) is rather meaningless on its own, but the closer it is to zero, the more predictable or little variance a data set has.

In fantasy football, less variance isn’t always a good thing, and it’s important to remember that. Toby Gerhart’s weekly totals didn’t fluctuate much last year, but it was because he was consistently worthless in the fantasy football world. Unusable fantasy players typically will have a lower standard deviation in terms of their weekly performances throughout their careers because they’re not worthy to have a recognizable one; a zero is a zero is a zero is a zero.

You do want consistent production from your top players though. If a player is a usable one in fantasy football, then that player is producing at some relatively high level; he’s not sitting on the waiver wire the entire season. And because of that, his variance is important.

When you stream a quarterback, you’re typically doing so based on matchup. That passer, too, is usually one that ranks in the bottom half of usable quarterbacks. For example purposes, let’s pretend streaming quarterback options rank between 16 to 25 in standard leagues. Remember, your streaming option could be off the bench, too.

Using weekly performance numbers from 2012, the following chart reflects the average points scored per performance and coefficient of variation amongst these quarterbacks:

Average

Coefficient of Variance

QB16 – QB20

13.63

0.57

QB21 – QB25

11.55

0.57

Again, these raw numbers don’t mean much without comparison. All we know is that the smaller the COV (up until zero), the less variance.

But the only way we can really get something from these numbers is if we compare them to other positions:

Average

Coefficient of Variation

WR1 – WR10

11.91

0.61

WR11 – WR20

9.65

0.63

WR21 – WR30

7.85

0.83

RB1 – RB10

13.98

0.57

RB11 – RB20

9.8

0.69

RB21 – RB30

7.52

0.81

TE1 – TE5

8.23

0.78

TE6 – TE10

6.33

0.84

TE11 – TE15

5.43

0.91

Data consists of weekly performances for each player under each group during 2012.

First, I’d like to note that the data above does not include instances where a player did not play (e.g. due to injury).  Therefore, you can believe that the numbers reflect healthy players; ones that you would be thinking about starting in your fantasy league during a given week.

As you can see, the coefficient of variation of the “other” fantasy positions is higher than that of quarterbacks. To put another way, each tier or grouping of startable non-quarterbacks in fantasy football is actually less predictable towards their mean, or average, than lower-tiered, QB2s.

Now, the small COV for lower-valued quarterbacks isn’t necessarily a good thing. In fact, it does tell us that these types of signal callers produce closer to their average more than top-tiered receivers and running backs do in a given week. That’s not automatically favorable considering lower-option quarterbacks score less than their higher-producing counterparts. Do you really want to strive for that?

But what it does tell us is a couple of things. First, we can see which positions, in general groupings, are most predictable.  The wide receiver, running back and tight end chart makes sense; as you move down the list and get to your worse players, they tend to have a higher COV. In other words, they’re producing at a higher variation each week because they’re not as good as their elite peers. From the data, it’s clear as day that quarterbacks are the most consistent of all fantasy positions.

Stemming from this: Second, the instance of data shows us the importance of high-end running backs and receivers. I talked through this in my article on wide receiver volatility, but we can’t conclude a position is inconsistent without comparison. When you look at receivers and running backs above, you can see the quick drop off from top-tier to middle-tiered players, and what that does to the coefficient of variation. This is just more proof that these types of players are incredibly important in fantasy drafts.

Lastly, and most importantly to the article, this data shows us that we should be confident with our streaming quarterback choices.  Though they may skew and perform towards their “low” mean, we have to recognize that we’re not playing these quarterbacks in any matchup; we’re playing them in favorable matchups. Therefore, not only should we feel good about our quarterbacks returning average production, but we should also remember that their season averages, against a porous defense, could actually be their floor.

Pat Thorman wrote a great article on quarterback streaming and why it’s effective. He proved that middle- and bottom-tiered signal callers perform well against bad defenses. They perform so well to the point that a top-10 passer isn’t even much better than a middling one versus a bottom-half defense.

And now, because you understand the coefficient of variation, you can see why predicting those favorable occurances isn’t as hard as you’d think.