Last week, I described a procedure for putting together a completely objective cheatsheet for RBs. Since then, I've gotten a ton of insightful feedback from the footballguys community (i.e., you guys). First, I want to apologize for not responding to personally to each of those emails. I read them all, and I appreciate the thoughts and the suggestions. I'll address some of them in this issue of the notebook, and then we'll get to the results for WRs.

A number of you suggested things which, if added to the model, might increase its usefulness. Among them:

1. How has the RB been playing over the past few weeks? That is, is there actually a tendency for "hot" players to stay hot? Would the folks down the hall agree that something like "last three week's performance" is worth considering?
2. How has the RB played in the past against this particular opponent (either in previous years, or earlier this year)?
   
   
3. Is the RB's team likely to win? Someone suggested that the model might be improved by adding in the Vegas line (or something similar) to take advantage of the tendency for winning teams to run more often than losing teams. If two RBs look identical statistically, but one of them is playing for a team that is a 10-point favorite, while the other plays for a team that is a 10-point dog on that particular week, is there legitimate reason to suspect that the 10-point favorite will be a better fantasy producer?
   
   
4. A few other people raised questions about the defensive ratings I was using. To refresh your memory, I rated every team's defense according to how many fantasy points they allowed to RBs (compared to league average). As some people pointed out, that figure would be skewed if a particular team has faced more than its share (or less than its share) of superstar RBs. That is absolutely correct. There are ways of dealing with this, and I will try to incorporate them at some point in the future. The same thing, incidentally, is true of the running backs. Instead of using the RB's actual yards per game average, I should be using some sort of adjusted average, which takes into account the strength of the defenses he's seen in the past. E.g. an RB who has averaged 100 yards per game against a very tough schedule should be viewed as being better than one who has averaged 105 yards per game against an easy schedule. Presumably, more accurate estimates of RB quality and defensive quality should produce more accurate predictions.
   
   
5. On the subject of defensive ratings, it's possible -- maybe even probable -- that I should be using yards allowed instead of fantasy points allowed to estimate the strength of defenses. After all, if using yards instead of points for the RBs gets better results, then using yards instead of points for defenses might get better results as well.

One serious weakness of the model is its inability to account for role changes. It does not (and cannot) know, for example, that Clinton Portis is now the main man in Denver. For that reason, the model as it stands will underrate Portis because it estimates his ability by looking at his overall season-to-date yards per game average. And since that average inculdes some games in which he got minimal action, it understates Portis ability to put up points in his current role. A couple of people made suggestions on how to deal with this:

1. you could just re-figure Portis' projection, but instead of using his overall yards per game average, use his average for the weeks he was a starter.
   
   
2. you could run these projections on a team basis instead of an individual basis, and then use your best judgement to divvy up the points the team would be projected to score. In other words, instead of using the model to predict that Portis will score 9 points this week, we could build a model that would attempt to predict how many points the Broncos RBs (all of them) will score. Then guess what percentage of that total Portis will get. Note that I wouldn't expect the previous model to work on the team level. It would have to be a new model, and I plan to give that a try when I get a chance.

Again, thanks to everyone for the feedback. There were many more intriguing thoughts than I have time to share here. As time permits, I'll do my best to get some of these ideas into the model. As always, I'll report back to you. Finally, with all the usual disclaimers and caveats, I'll present the week 7 RB list:

 Name                  OPP  FPPG   Y/G  DEF  PROJ
-------------------------------------------------
 Priest Holmes         den  27.0   160   -5  21.1
 LaDainian Tomlinson   oak  20.2   142   -1  19.3
 Ahman Green           was  16.3   139   -3  18.5
 Charlie Garner        sdg  21.0   138   -6  18.0
 Ricky Williams        buf  20.2   132   -1  17.9
 Marshall Faulk        sea  16.7   116   10  17.7
 Deuce McAllister      sfo  17.1   121    1  16.8
 Fred Taylor           bal  16.4   128   -6  16.5
 Jamal Lewis           jax  16.1   125   -4  16.5
 Edgerrin James        pit  12.6   114   -2  15.3
 Michael Bennett       nyj  10.9    85   17  14.8
 Shaun Alexander       ram  19.4   110   -3  14.8
 Travis Henry          mia  19.0   110   -3  14.7
 Stephen Davis         gnb  14.3    95    3  13.7
 Lamar Smith           atl  16.2   102   -2  13.6
 Anthony Thomas        det  10.8    84   10  13.3
 Moe Williams          nyj  11.7    69   17  12.6
 Clinton Portis        kan  10.3    83    1  11.8
 Michael Pittman       phi   9.3    93   -7  11.7
 Jamel White           hou   9.1    81   -0  11.2
 Garrison Hearst       nor  10.7    83   -2  11.2
 Duce Staley           tam  12.2    86   -8  10.6
 Thomas Jones          dal  10.0    85   -7  10.5
 James Stewart         chi   9.4    79   -4  10.2
 Emmitt Smith          ari   8.4    74   -3   9.8
 Curtis Martin         min   7.4    62    4   9.4
 Jerome Bettis         ind   9.6    60    5   9.1
 Kevan Barlow          nor   9.1    67   -2   9.0
 Olandis Gary          kan   7.2    57    1   8.2
 Mike Anderson         kan   8.1    57    1   8.2
 Warrick Dunn          car  10.3    55   -9   6.3
 Mike Alstott          phi   9.3    53   -7   6.3
 Stacey Mack           bal   9.0    30   -6   3.3
 
FPPG = the RB's fantasy points per game average
Y/G = the RB's yards per game average
DEF = the defense's rating (negative = good D, positive = bad D)
PROJ = projected fantasy points for week 6

Now, on to the WRs...

Same game plan as with the RBs. For completeness' sake, I'll quickly go through the details:

I looked at every game by a WR in weeks 5-17 of 2000 and 2001. I recorded the WR's yards per game and TDs per game averages going into the game and their opponent's ability to stop WRs (as measured by fantasy points allowed vs. WRs compared to league average) going into the game. I dismissed as irrelevant any WR who wasn't averaging at least 4 fantasy points per game. I threw all that into the pot and let the computer sift through it. Some interesting things came out:

1. As with RBs, TDs turned out to be unhelpful as a predictor of future fantasy points. Yardage predicts future fantasy points better than fantasy points do.
2. Unlike with RBs, strength of defense turned out to be totally and completely useless in predicting fantasy points (more on this below).
3. Predicting WR stats is tougher than predicting RB stats. The people down the hall are a lot less confident in their ability to predict weekly WR stats than weekly RB stats. Any longtime fantasy football player can sympathize.

The long and the short of it is this: the folks down the hall think that your weekly WR cheatsheet should simply have all the names listed in order of their year-to-date yards per game average.

At that point, I had to take my data back from the eggheads and fiddle around with it myself. Strength of defense doesn't matter at all? I had to take a closer look at this. I decided to break the WRs into two groups: the studs (averaging 9+ fantasy points per game going into the game) and the borderline starters (averaging 4-9 points per game). Something surprising emerged:

The studs were affected by strength of defense more than the borderline guys were. In fact, while the studs were somewhat hampered by tough defenses (not much, but a little), the borderline guys actually did slightly better against tougher defenses. Whether this is a fluke or a real effect is not clear to me, but either way, it's very very tough to conclude that you should systematically downgrade WRs, especially the borderline guys, because of tough matchups.

I said this last week with the RBs, but it bears repeating here: I believe that the defense probably does play a significant role in determining how well a WR will do in a given week. The problem is that it's very difficult to figure out ahead of time who the good defenses are. Handicapping defensive strength against WRs is even tougher than handicapping defensive strength against RBs.

So I don't know what to think here. Possibility #1 is that the folks down the hall simply aren't much help. That is, you have to use actual football knowledge -- the details of the specific matchups, gamplans, and tendencies, not just generic knowledge about the quality of the defense -- to have a prayer of picking the right WR. Possibility #2 is that predicting weekly WR production is a crapshoot, and it simply can't be done with any degree of accuracy. I'd be interested to see how a cheatsheet crafted solely from yards per game would fare against the expert cheatsheets. I'm not sure about this, but I suspect that most experts (by that I mean anyone who has been serious about fantasy football for a few years) are capable of putting together a cheatsheet that would beat the "down the hall" cheatsheet most of the time for WRs. I do believe, however, that most experts could improve their own cheatsheet by putting less emphasis on the numerical strength-of-defense estimates. That's what I'll be taking from this week's edition.

Next week, I'll look at QBs and TEs. I'm nearly certain that the TE conclusions will be similar to the WR conclusions (i.e. not much help). The QBs? I have no idea what we might find.

Unless otherwise noted, all stats come from football-reference.com and the disclaimer applies