The Standard Deviation as an Important FF Statistic?

[The_Dude]

Now, for debates sake, I will only focus on the WR's. I admit that 3 of the 4 WR's in this post had injuries or did not play, but it should help to prove my point assuming they would have played all 16 games as SSmith did.

Again, STDV should not be used as a tool to pick where people get drafted. It should be used as a tool to decide between 2 WR's that you feel are deserved of being picked at the same spot roughly. STDV is pretty much the same as saying I can pick (this is an example and not based on actual schedule) Dunn or Droughns as my 3 RB but Dunn has the same bye week as my number 1 RB, so I'll take Droughns. STDV is more like a tiebreaker if you will.

So looking at the WRs. Owens had a higher AVG score, lower STDV, and lower relative STDV when compared to SSmith. So given the choice, and assuming all else is equal, TO would be the more logical pick. Of course TO is in a different setting this year so his value could change drastically, but you see the point I am getting at.

On the same level, let's say you had the choice between SSmith and Holt. Yes SSmith had a higher AVG score, but Holt had a lower STDV and lower relative STDV. My argument would be that week in and week out Holt would be a more consistant WR. Looking at the hard numbers from Twisteds method, you can see that SSmith actually outscored Holt more weeks than Holt outscored SSmith, but there is not one week when Holt would have cost you a game, which unfortunately SSmith did a few times last year. Yes it is nice to see to 40+ weeks from a player, but we have all probably lost because our stud decided to put up a whooping 5 points or something to that effect.

So I hope the additional charts give a better understanding of what I use STDV for. And it should be said that this also goes along with the type of owner you are. Some owners are more of the gambling type and would prefer the boom or bust type player, whereas some would want the week in and week out performer. And depending on your stance, or stance at a particular pick in the draft, you can use STDV and relative STDV to help make that decision.

Great stuff here... no thunder stolen, whatsoever. I am still going to crunch the yardage st devs for the other positions (including TE), because I think that just conceptually, higher fantasy point standard deviations could result from touchdown variation, which I guess is important, but I always want 10 points out of my players (i.e. 100 total yards) and the touchdowns will come when they come.

I was also thinking that standard deviation may govern precisely which kicker I pick in the last round of my draft since consistency is all I want out of that position.

[Plindsey88]

I was also thinking that standard deviation may govern precisely which kicker I pick in the last round of my draft since consistency is all I want out of that position.

In that case, you should try for George Blanda... He's been nothing but consistent the past few years...

[The_Dude]

I was also thinking that standard deviation may govern precisely which kicker I pick in the last round of my draft since consistency is all I want out of that position.

In that case, you should try for George Blanda... He's been nothing but consistent the past few years...

Behind all the stats and sound fantasy advice, there's a comedian in plindsey yet, burning to come out. I think all this moderator/cafe starting lineup stuff is going to his head.

[mpw82]

You can not look at the standard deviation alone, as I think we have all agreed up. I think the best way to possibly use this is to construct confidence intervals with the stardard dev. Using something like a 90% (we dont need to be very confident, could possibly use something a little lower) to give the expected low and also a pretty good idea of what their highest performance might be.

I am not really sure how these work out, as I have not computed anything of yet, but these might be the way to go. It might turn out that everyone's CI is 0 to 20 and all this is irrelevant.

[The_Dude]

You can not look at the standard deviation alone, as I think we have all agreed up. I think the best way to possibly use this is to construct confidence intervals with the stardard dev. Using something like a 90% (we dont need to be very confident, could possibly use something a little lower) to give the expected low and also a pretty good idea of what their highest performance might be.

I am not really sure how these work out, as I have not computed anything of yet, but these might be the way to go. It might turn out that everyone's CI is 0 to 20 and all this is irrelevant.

The nice thing about the Standard deviation is that if you can assume that the data are normally distributed, then 67% of cases will lie within +- 1 standard deviation of the mean and 95% will lie within 2 standard deviations of the mean - so that gives you your interval right there (although I haven't checked the distributions yet to check on how realistic the normal dist. assumption is.