Semi-hookups

In this article, I talked about the hookup (starting a QB and a WR from the same NFL team) and the anti-hookup (starting a WR on the same NFL team as your opponent's QB). Several people emailed and asked me about what I've decided to call a semi-hookup: starting a QB/RB combination, an RB/WR combination, or even a WR/WR combination from the same NFL team.

A quick review of the relevant conclusions of the previous article:

  1. A hookup should increase the variance in your team's score. I.e. it should make your team less consistent.
  2. Increased variance is desirable for an underdog, but should be avoided by a favorite.
  3. Thus, hookups are a good idea if you're playing against stronger team and a bad idea if you're going against the league's doormat.

The punchline is point #3 above, which depends strongly on point #1. But it's far from clear that point #1 is still true with semi-hookups. To phrase that a little more clearly in terms of an example: if Peyton Manning has a good game, does that mean that Edgerrin James probably had a good game, too? Does Peyton's success open things up for Edgerrin, or does it "steal" some of his production? This is similar to the question posed in this article. And it turns out the answer will be similar -- or rather, the lack of an answer will be similar.

To measure this, I'll use the correlation coefficient. I gave a fairly detailed explanation of the correlation coefficient here, but I'll re-state the main idea in case you've forgotten. The correlation coefficient is a number between -1 and +1. -1 means a perfect negative correlation, +1 means a perfect positive correlation, and 0 means no correlation at all (technically, no linear correlation, but don't worry about that).

Using my usual definition of fantasy points ((RushYD + RecYD)/10 + (all TDs)*6 + (PassYD)/25 - (INT)*3), the correlation coefficient between Manning's fantasy points and James' was +.14. That's a positive correlation, which means that the two had a tendency to move up and down together. But it was a very weak tendency, as evidenced by the fact that +.14 is very close to zero. Let's take a look at some other premier QB/RB combos from 1998 and 1999.

Player1       Player2            Years       Games      Corr. Coeff
-------------------------------------------------------------------
P Manning     E James             1999         16          .14
K Warner      M Faulk             1999         16         -.34
B Favre       D Levens            1999         14          .24
B Johnson     S Davis             1999         14          .17
S McNair      E George            1999         11         -.16

S Young       G Hearst            1998         15          .02
C Chandler    J Anderson          1998         14         -.25
Interesting. Some pairs exhibited a positive correlation, and some a negative, but all were fairly weak. Note that the strongest correlation was Faulk/Warner, and it was negative. This shows up quite clearly in their week-by-week numbers:
   Week   Opp   Warner  Faulk
   --------------------------
      1   BAL      25     13
      3   ATL      36     23
      4   CIN      30      4
      5    SF      41      4
      6   ATL      11     27
      7   CLE      26     26
      8   TEN      33     24
      9   DET      27      9
     10   CAR      21     20
     11    SF      12     17
     12    NO      21     24
     13   CAR      26     20
     14    NO      23     33
     15   NYG      26     17
     16   CHI      28     32
     17   PHI      12     23
Every time Faulk had a bad game, Warner was huge, and Faulk came up big in most of Warner's off weeks. Given that Faulk is such a major part of the Rams' passing game, this is a little surprising.

Now let's look at the correlation over the last five years for some QB/RB duos that have been together for awhile:

Player1       Player2            Years       Games      Corr. Coeff
-------------------------------------------------------------------
T Aikman      E Smith        1995-1999         72         -.02
B Favre       D Levens       1995-1999         68          .09
J Elway       T Davis        1995-1999         57          .13
T Dilfer      M Alstott      1995-1999         57          .24
T Dilfer      W Dunn         1995-1999         42          .15
S McNair      E George       1995-1999         52         -.00
J Blake       C Dillon       1995-1999         28          .33
It looks as though the overall effect may be a positive correlation, but a small one. Just for comparison's sake, check out the correlations for some QB/WR pairs:
Player1       Player2            Years       Games      Corr. Coeff
-------------------------------------------------------------------
B Favre       A Freeman      1995-1999         62          .35
S Young       J Rice         1995-1999         43          .37
P Manning     M Harrison     1995-1999         28          .54
Always positive and much stronger than most of the QB/RB pairs.

What this says is that QB/WR pairs are likely to cause your team to be inconsistent, while QB/RB pairs will not necessarily do the same. Sure, sometimes McNair and Eddie George get shut down on the same week, but sometimes McNair and Stephen Davis get shut down in the same week, too. If the past is any guide, there is no reason to believe that one is more likely than the other.

Thus, there is no reason to suspect that a McNair/George team will be any more or less consistent than a McNair/S Davis team or a Gannon/George team. Draft McNair if you think he's better than Gannon and draft George if you think he's better than Davis. Don't seek the semi-hookup, and don't avoid it.

Let's take a look at some premier RB/WR pairs:

Player1       Player2            Years       Games      Corr. Coeff
-------------------------------------------------------------------
C Martin      K Johnson      1998-1999         31          -.01
E Smith       M Irvin        1995-1999         61           .01
R Smith       C Carter       1995-1999         57           .03
B Sanders     H Moore        1995-1999         63          -.15
Extremely weak correlations here.

What about WR/WR duos?

Player1       Player2            Years       Games      Corr. Coeff
-------------------------------------------------------------------
C Carter      R Moss         1998-1999         30           .07
R Smith       E McCaffrey    1995-1999         52          -.06
H Moore       J Morton       1995-1999         65          -.12
Again very weak correlations -- so weak that they are negligible. Sometimes a big game for Rod Smith has coincided with a big game for McCaffrey, and sometimes it has "stolen" some production from Ed.

Overall conclusions: except for the obvious QB/WR hookup, there seems to be no predictable relationship between the performance of teammates in a given game. Based on this, I would not shy away from something like Enis/Robinson/Engram or Holt/Bruce/Faulk. If those were the top guys on my list at the time I was picking, I wouldn't let the fact that they are teammates bother me at all.

I need to emphasize that this was not an exhaustive study, and its conclusions are therefore limited. It was just a quickie and I chose to look at just a few more-or-less random examples. I did include all the examples I looked at, but it is certainly possible that there is something hiding in the data that further study would reveal. This is not enough to prove conclusively that semi-hookups are irrelevant, but for me it's enough to shift the burden of proof. That is, if you want to convince that semi-hookups matter, you're going to have to show me some hard data that shows a trend my examples missed.