We'd all like to know how our fantasy players are going to do the rest of the year. In this article I'll look specifically at running backs using two measures: fantasy points per game (FP/G) from Weeks 7 to 17 and the change in FP/G between Weeks 1 to 6 and Weeks 7-17.

If you've read my articles for any length of time, you may think they can be too complicated with stats and math and stuff. And you may be right. I hope this piece is a little simpler. But there will be some math.

I went back to 2002 for the data (since the NFL has had 32 teams). I'm used PPR scoring per game: 1 fantasy point (FP) for every 10 yards; 6 FP per TD, and 1 FP per reception. I threw out RBs who played fewer than 3 games in the first 6 weeks of the season and those who played fewer than 5 games in the last 11. And then I limited myself to looking at just RBs who ranked in the Top 50 in FP/G in Weeks 1-6.

First, I calculated the correlation between several stats from the first six weeks with FP/G from Weeks 7 to 17:

Relationships Between RB Stats in Weeks 1-6 and Their FP/G in Weeks 7-17

Weeks 1-6 Stats



Yards Per Carry



Rushes Per Game



Touches Per Game



Opportunities Per Game






By "correlation" I technically mean correlation coefficient, which can be a number between 0 and 1, and be positive or negative. If the coefficient is positive, it means as one number (the Weeks 1-6 stats) goes up, the other number goes up (FP/G in Weeks 7-17). The closer the coefficient is to "1" the stronger the relationship between the two stats.

Touches are rushes plus receptions; opportunities are rushes plus targets.

Let's look at Rushes Per Game in Weeks 1-6. How often an RB carries the ball each game to start the season has a 0.53 correlation coefficient with his FP/G in Weeks 7 through 17: the more rushes he has in his early season games, the more points he's likely to score in the rest of his games (if the coefficient were a negative number, more carries in Weeks 1-6 would mean fewer points in Weeks 7-17). The strongest correlation to FP/G in Weeks 7-17 is with FP/G in Weeks 1-6.

R-squared can be a bit more complicated stat to calculate than correlation coefficient, but in this case it's just the square of the numbers in the table. What it tells us is (approximately) how much of Weeks 7-17 FP/G is explained by the stat in the first column: Rushes Per Game in Weeks 1-6 explains about a quarter of the Weeks 7-17 FP/G. In a way, this isn't too surprising, since football is a complicated game and no simple stat can be expected to explain a lot.

FP/G isn't an exactly simple stat: it's the result of not only opportunities or touches, but also production or performance driven not just by a player's own ability but by that of the players and coaches around him, and opposing him.

Because football is complicated – and while FP/G may not be as complicated as the game itself, it is the outcome of complex inputs – sometimes numbers guys (like me) can make it too complicated, with fancy stats.

I tried a little of that in this analysis, using the values of each type of rush or target discussed here to weight the values of different types of opportunities in Weeks 1-6 and then use that to try to better predict Week 7-17 FP/G. But the correlations and r-squared numbers weren't as good as the FP/G for that period[i].

The bottom line is that after six weeks, the best predictor of an RB's FP/G the rest of the way is his FP/G so far. Simple and obvious.

But RB FP/G for Weeks 7-17 is only one of the things I want to know. I also want to know who is going to get better or worse. The sad fact is that of the Top 50 RBs in Weeks 1-6 from 2002-2015, this far in the season, about 63% scored fewer FP/G the rest of the year:

RB FP/G Rank in Weeks 1-6 and Change in FP/G the Rest of the Season

FP/G Rank




% Improving

FP/G Change































Of the 700 RBs in the Top 50 in FP/G each year between 2002 and 2015, 84 did not play even 5 more games, that's why the table above only has 616 RBs in it. Think about that: 1 in 8 of the Top 50 so far will not play 5 more games. (A lot of the Top 50 backs for Weeks 7-17 aren't even in the opening weeks' Top 50, I'm not analyzing those players).

And the RBs who see their FP/G dip the most on average are the RBs who have scored the most so far. The overall change of -0.8 FP/G by the Top 50 RBs explains why the correlations in the next table are all negative:

Relationships Between RB Stats in Weeks 1-6 and Change in FP/G in Weeks 7-17

Weeks 1-6 Stats



Yards Per Carry















Yards per carry doesn't look as bad in this table as in the previous one but notice that only FP/G for the first 6 weeks explains more than 1% of the change in FP/G. And 8% is not much.

There is a stat that explains about half of the change, and it's a pretty simple one. The change in touches per game from Weeks 1-6 to Weeks 7-17 has an R-squared of 0.49[ii]. The thing we want to figure out is which Top 50 backs will get a heavier workload and which backs will see their touches drop. Obvious right?

That's the thing. I'm not sure that stats through Week 6 can tell us a lot about that. And changes in future efficiency may not be adequate either: for example, the change in yards per carry between the two periods only has an R-squared of 0.08. Not great, and anyhow, how can we tell whose yards per carry is going to drop? Or whose workload is going up?

The answer, I think, lies mostly in subjective analysis – injuries (and not just to the RB but also to his line, his QB, and even the receivers and defensive players), coaching changes, evolving roles in a backfield that aren't captured purely in the stats to this point in the season, etc. I do think strength of schedule is also a factor and I'll look at quantifying that next week. But it's important to admit that stats alone can't predict the future – or at least my stats can't.

[i] I didn't how those numbers because I didn't want to spend a lot of time explaining something irrelevant. Someone may come up with a stat that better explains Weeks 7-17 RB FP/G than what I looked at.

[ii] Of course it's not fair to compare the value of a future stat that we don't know yet to a past stat. That's the point.