Standard Deviation to Determine Player Consistancy (1 Viewer)

[vitaflo]

Something I've been thinking of for a while is the way people slot players based on yearly totals and points scored for their league without much thought into how "streaky" a player may be. Now, I have seen some of this, people looking at past stats to get a quick feel of how consistant a player is, but I'm curious if anyone has actually went all out and done a Standard Deviation calculation on all players to determine how consistant they really are?

For example, I took stats from two players last year, Santana Moss and Larry Fitzgerald. They both scored about the same but the difference in standard deviation is notable.

Santana Moss:

REC YDS TD PTS4 96 0 95 159 2 296 87 0 88 116 0 1110 173 2 315 112 1 174 34 0 17 79 0 74 79 0 74 53 0 56 65 1 123 58 0 57 56 0 52 73 0 75 160 3 384 83 0 8Total Points: 200Mean Avg: 12.5

Std. Dev.: 10.4

Larry Fitzgerald:

REC YDS TD PTS13 155 1 214 70 0 93 41 0 47 102 1 169 136 1 194 71 1 134 36 0 38 102 0 109 141 1 209 104 1 164 41 1 108 129 0 124 23 0 26 85 1 145 93 1 156 80 1 14Total Points: 198Mean Avg: 12.375

Std. Dev.: 5.6

They both scored around the same number of points last year, but it's obvious that Fitzgerald is way more consistant than Moss was. You can somewhat see this by just looking at the raw numbers, but a 10 pt deviation is quite a lot and can win or lose a game for you easily (key word there being *lose*).

I'm wondering if this info is actually usefull and if anyone uses this kind of data when looking at players to draft. Is there any place out there that has this info already tallied? Should we be just as worried about a players consistancy when drafting and not just their raw point totals?

 

[rascal]

I started to look into this until somebody posted an article by Doug that pretty much stated you can't expect somebody to be consitent one year and be consistent the next year.

Check his site though in the blog section and you should find it.

 

[radballs]

There's been a few threads about this as recently as a couple months ago. I used to do something similar, but I've since learned that players have not been found to have any statistically significant measure that shows consistency in year X and year X+1. In other words, just because a player showed week to week consistency one year does not mean that they will show consistency in the next year.

 

[JayMan]

There's been a few threads about this as recently as a couple months ago. I used to do something similar, but I've since learned that players have not been found to have any statistically significant measure that shows consistency in year X and year X+1. In other words, just because a player showed week to week consistency one year does not mean that they will show consistency in the next year.

As usual from radballs...This is a major issue - and the other issue with using Standard Deviation (StDev) to find if a player is consistent or not is an incorrect way to look at this.

As you might know, StDev shows the dispersion to the mean from a dataset... but the influant variable here is the "mean"... For example, one player can score 2 points every week of the year - thus is StDev will be 0 - he's the most consistent FF player there is - but he's constantly poor!

On the other hand - it is a well known fact that we are looking for consistent players - i.e. guys that can score a good amount of point week-in week-out to help our teams win H2H games... and it is also proven that a group of "consistent good players" will beat a group of "inconsistent good players" over 70% of the time... (but as radballs mentioned - it is very difficult to predict the consistent players just by looking at historical data)...

What should be used instead to figure out if a player is consistent or not - is the percentage of games that he scores at least the number of points the "last starter" from his position scored on every week...

You can even cut into those percentages and state that if a guy was in the best 25% of the starter at his position for a given week - he had an "elite week"

(i.e. if you start 12QBs... then the best 3 QB performances on week1 gives an "elite" week - while the top12 gets a "starter" week)... gathering that information, by position, for every week of the year...

Depending on your scoring system - that % can vary obvisouly, but for example in '05 - the most consistent RBs were:

name----- elite --- starterAlexander - 12 ------ 14Johnson --- 10 ------ 12James ------ 8 ------ 12Barber ----- 9 ------ 11Tomlinson -- 8 ------ 11With that in mind - it can help you identity guys that can consistently score enough points to warrant a starting spot on your FF roster...My 2 cents