8/17/09
This article is a continuation of last week’s investigation
of historical QB performance in which we dug into the distribution
of each QB’s historical fantasy points and isolated some specific
players that stood out as being more or less consistent than their
peers. This week, we apply a similar approach to WR’s.
Approach
We focus on only the last four seasons. Our intention is to consider
only games in which a fantasy football manager would have considered
starting the player. With QB’s this was a little easier –
we basically counted all the games the QB started, and we didn’t
count the games they didn’t start. With WR’s it’s
a little trickier. For example, although a manager might consider
starting Arizona’s #3 WR in some games last year, a manager
probably wouldn’t ever consider starting Atlanta’s #3
WR last year.
If a player has been a clear fantasy starter for all four years
(e.g. Reggie Wayne), we include all his games. This includes games
in which he played but scored zero points, but it doesn’t
include games where he did not play (e.g. because he was injured).
If a player became a viable fantasy starter more recently, we make
a subjective judgment on when a fantasy manager might have considered
starting him. We do this individually for each player. Then once
a player is considered a viable starter we include all his games
subject to the same constraints as the veterans above.
The table below shows the sample size we are left with for each
player.
Note that since we look at 4 years of games in Weeks 1 through 16
the maximum possible total is 60 (4 years of 15 non-bye weeks).
Donald Driver and Reggie Wayne are the only receivers with a sample
size of the full 60 games.
Scoring
The value of wide receivers can vary greatly based on whether or
not points are awarded for receptions (i.e. PPR). This analysis
focuses on a PPR scoring system. A future article will include an
expansion of the analysis to a non-PPR league. That article will
also include an extensive investigation of WR’s whose value
is highly dependent on the scoring system. The scoring system we
assume for this article follows below.
PPR Scoring |
PPR |
Yds |
TDs |
Fumles Lost |
0.5 |
0.1 |
6 |
-2 |
|
Note we give no credit for special teams, passing, or rushing performance.
Totals
Let’s start by looking at the average fantasy points scored
per game under this set of assumptions. We look at the points per
game rather than the total points, since we don’t want to
imply a player has been more valuable just because they have been
around for more seasons.
Terrell Owens is on top, along with other names that are probably
not surprising. Calvin Johnson’s stats are drawn down by his
2007-2008 season, which is assigned equal weight as last season.
Wes Welker’s stats are drawn down by his 2006-2007 season,
which is assigned equal weight as his last two seasons. Note we
don’t include Welker’s 2005-2006 season as we assume
he was not a viable fantasy starter at that point.
This gives us some perspective on the average fantasy points per
game the top WR’s have been scoring over the past few years.
We continue our analysis by next investigating the distribution
of their scores per game.
Volatility (and Coefficient of Variation)
Volatility is a measure which quantifies how widely a data set varies
from its mean. If a player scores about the same amount of points
almost every game, his scores will tend to have low volatility.
If a player is just as likely to score 40 points in a week as 0,
then his scores will tend to exhibit higher volatility. Let’s
look at the volatility of the WR scores.
What stands out here is that the lowest scoring WR’s tend
to be at the bottom of this list, and the highest scoring WR’s
tend to be at the top of this list (one notable exception is Larry
Fitzgerald). Part of this might be due to the higher scoring WR’s
being more volatile, but another piece is due to the fact that just
looking at the volatility without scaling it for their average scores
will tend to effectively overstate the volatility for the high players
and understate it for the lower players.
Let’s scale their volatility by their average score. In other
words, we’ll look at their coefficient of variation (CV).
This will allow us to compare WR’s who average low scores
against those who average high scores a little bit better. (Thanks
to reader Patrick for the useful suggestion on looking into this!).
If I’m choosing a wide receiver with one of my first couple
picks – I want him to score a lot of points and do it consistently.
Larry Fitzgerald, Reggie Wayne, and Andre Johnson historically share
these characteristics.
When choosing a wide receiver in the later rounds of a draft, the
available options probably will not be scoring a lot of points historically
on average. If they were, then they probably would tend to have
been picked earlier. When choosing a WR like that, a high coefficient
of variation tends to be more attractive than a low one. In other
words, if a WR doesn’t score many points on average, then
it’s preferable for him to be inconsistent than consistently
bad. In some ways when people talk about a player with a low average
score with upside, they are indirectly referring to a high coefficient
of variation.
Distribution Of Scores
Next we investigate the distribution of scoring per game of each
wide receiver. The following table shows the maximum, minimum, and
percentiles of scores for each player. 90th percentile indicates
90% of the time the player scores less than that score. Median indicates
50% of the time he scores more, 50% of the time he scores less.
The table is sorted by the median score.
So 90% of the time Larry Fitzgerald scores less than 23.1 fantasy
points, and 10% of the time he scores more.
The following table shows the same ideas expressed as a ranking
rather than raw scores (again sorted by the median).
The table is sorted by the ranking of their median performance from
best to worst. The 10 for Larry Fitzgerald under 0.9 indicates that
the 90th percentile score of Larry Fitzgerald is the 10th highest
90th percentile score of all the players considered.
The 1 for Terrell Owens under Max indicates that his highest score
is the highest high score of all the players considered. The 6 under
Min indicates that his lowest score is higher than all but 5 wide
receivers.
Let’s step back and see if anything stands out.
- Larry Fitzgerald, Reggie Wayne, Terrell Owens and T.J. Houshmandzadeh
are ranked pretty high at the high percentiles. This indicates
their good days are better than a lot of other players’
good days. They also stand out as being ranked high at the low
percentiles. This indicates their bad days are better than a
lot of other players’ bad days. These are favorable characteristics
to have.
- Dwayne Bowe has an interesting profile scoring well at the
median and low percentiles. This indicates that his bad days
are better than those of most other players. But he doesn’t
score nearly as well at the higher percentiles. This indicates
that his historical upside has been lower than that of many
other players.
- Randy Moss scores well at the high end indicating his good
days are better than those of most. But that is coupled with
mediocre (but not bad) scores on the bottom end. Note this is
consistent with his relatively high coefficient of variation.
- Lee Evans has a good maximum score, but he performs badly
at most other levels. This indicates he has had a couple great
games but has had relatively mediocre performance outside of
those.
- Donald Driver is in the top 25 at all levels, and that stands
out favorably against the peers around him.
- Steve
Smith scores well at the top and poorly at the bottom. This
indicates his good days are better than many others, but his
bad days are worse than the bad days of many others. Some might
find T.J. Houshmandzadeh’s historical profile more attractive
despite the lower historical upside. Many would find Reggie
Wayne’s profile more attractive, despite the lower historical
upside.
OK That’s Nice – But What Can
I Do With This?
This provides some additional perspective on the distribution of
historical wide receiver performance. In practice it a couple ways
it can be used include:
- Let’s say you are thinking of taking a wide receiver in
the first round of your draft. You might be able to decrease your
risk a bit if you constrain yourself to only wide receivers that
both score a lot of points historically and do it with a low coefficient
of variation. If those aren’t available, that might be one
good indication to push more towards a choosing another RB.
- Let’s say you are in one of the later rounds in your draft
and you are thinking of taking a wide receiver. No one left scores
a lot of points historically. You might have a better chance of
your receiver overperforming if you choose one that has a high historical
coefficient of variation.
Conclusion
We highlighted some wide receivers that have score well against
their peers anyway you look at it (e.g. Larry Fitzgerald, Reggie
Wayne, and Andre Johnson). Some wide receivers that have performed
among the best overall, but have involved more risk historically
include Randy Moss and Steve Smith.
Your #1 and #2 picks are extremely valuable. Aim for a lot of
points and consistency with whatever players you choose here.
My previous article includes
a more detailed discussion of the costs of taking on higher risk
in early round draft picks.
Sometimes having good perspective about the performance of available
players historically is useful. This article aimed to help build
some of that perspective, and I hope you find it useful.
Next Steps
Next week we will put together a similar analysis for running
backs. And the week after that we will circle back to wide receivers.
In that article we’ll expand this analysis to non-PPR leagues
and we will highlight wide receivers whose performance is particularly
sensitive to the PPR format, making some relevant predictions
along the way. As always, feel free to contact
me with any questions or suggestions. And I hope to see you
back next week.
|