Valuing categories in roto leagues

Great discussion y'all. To weigh back in on this one here (stupid personal life getting in the way of fantasy hockey -- WTF!), we've nailed down z-scores pretty well. They are a great way to describe a player's stats with respect to whatever population we're comparing with (all skaters, D, LW, whatever). BUT we're still left with the original question raised in the start of the thread: how do people weight/value categories against each other?

I suspect we all use experience-based factors (you know, goals are harder to get than assists, etc, etc) but I would like to look at it mathematically to see if we can draw any interesting conclusions.

For reference, I'm using the following weight factors:

G: 1.0
A: 0.7
P: 0.8 (2 assists per goal, so (0.7*2 + 1)/3 = 0.8
+/- = 0.5
PIM = 0.75
PPG/PPA/PPP = same as G,A,P
SHG/SHA/SHP = 0.2 due to wild swing in value when 1 SHP can be around 20% of the players' category!
SOG = 0.8

Obviously these factors will vary based on the league setup, and I'll try and quantify that. I'm going to root around and look to see what stat techniques could be applied to give us some answers. Any input is obviously appreciated.

And just for interest (everyone loves results, LOL) here are the top 30, using z-scores to evaluate each category and scaling as indicated in the post above (is that even valid??) in a 4C 4LW 4RW 6D 2BN league that scores G, A, P, +/-, PIM, PPP, SHP, SOG, all using the above-the-bar technique:

1. Ovechkin
2. Phaneuf
3. Green
4. Gonchar
5. Crosby
6. Malkin
7. Iginla
8. Rafalski
9. Heatley
10. Lidstrom
11. Zetterberg
12. Markov
13. Chara
14. Pronger
15. Lecavalier
16. Corvo
17. Boyle
18. Hossa
19. Gaborik
20. Kovalchuk
21. Kovalev
22. Burns
23. D Sedin
24. Getzlaf
25. Nash
26. Niedermayer
27. Doan
28. Thornton
29. Hemsky
30. Vanek

These results (and pretty much all the ones that I have come up with using the above-the-bar technique) show that stud D are hugely valuable in roto leagues, especially roto leagues that have a high number of defensemen owned.

I'm obviously way late too this thread but here is what I did to try and assign numerical values to each player in my roto league.

First off I added up the winning totals in each of the 4 skater stats. Then I divided each individual stat by that summed total to determine the weight of each category relative to the total amount of production.

After that I divided the highest production category, PIM by each individual category G,A,PPP and that total became what I called my "multiplyer."

The categories with lower totals were deemed to be more valuable and got a higher multiplyer.

The results from this phase:

Category Winning Total Weight Multiplier
G 273 0.14 3.4
PPP 277 0.14 3.3
PIM 915 0.47 1.0
A 463 0.24 2.0
1928 1.00

Basically what this was telling me is that Goals are worth 3.4x as much as PIMs, PPP are worth 3.3x as much and A worth about 2x as much.

Still with me?

So now that I found out how valuable each stat was I wanted to use that to assign each player a total score based on their production. I used pro-rated totals from last season to generate my list.

The formula was essentially this:

SUM=((G/GP*82*3.4)+(PPP/GP*82*3.3)+(A/GP*82*2.0)+(PIM/GP/*82)


Anyways, I found out that PIM guys were scoring way too high as Carcillo was significantly ahead of Ovechkin in this so what I did was divide PIM's in half to get what I felt was a more representative score.

Here are the top 10 scores for each position in terms of my formula for ROTO production based on last years pro-rated totals.

Center
Sidney Crosby 472.5
Evgeni Malkin 472.4
Jason Spezza 428.0
Ryan Getzlaf 411.7
Joe Thornton 411.0
V. Lecavalier 406.0
Pavel Datsyuk 405.8
Mike Richards 391.2
Mats Sundin 391.1

Left Wing
Alex Ovechkin 475.9
H. Zetterberg 436.9
I. Kovalchuk 400.5
D. Carcillo 372.6
Ray Whitney 354.6
B. Morrow 354.5
Daniel Sedin 337.2
Rick Nash 331.8
T. Plekanec 301.8
Thomas Vanek 300.8

Right Wing
Jarome Iginla 435.6
A. Kovalev 426.7
Dany Heatley 413.8
D.Alfredsson 413.6
M. Gaborik 374.8
T. Selanne 349.4
Marian Hossa 342.9
Shane Doan 342.1
Alex Hemsky 339.5
M. St Louis 332.9

Defence
Sergei Gonchar 370.9
Dion Phaneuf 360.9
N. Lidstrom 334.7
Chris Pronger 316.8
Dan Boyle 307.4
Zdeno Chara 297.0
B. Rafalski 296.9
Andrei Markov 292.3
Mark Streit 291.6
B. Campbell 272.4


A couple important notes:
- The minimum games played to be in consideration was 20
- This is based on pro-rated production for last season and the relative value of each stat to the highest scoring stat, PIM's
- Dan Boyle's PIM's are obviously way inflated which gives him a higher score


I don't know if I explained this all that well but I was very pleased with the resulting rankings despite their probably being some very transparent flaws.

- Mikey

Very interesting results though I must admit I am struggling with the D vs F comparisons still, and wondering if we are missing something there. It's just that I can't put my finger on exactly why it seems off.

I need to think on it some more.

On a secondary note, can anyone give me a bit of an idea how to copy and paste a set of last year's stats from say Yahoo into my current Excel Sheet (which only lists Goals, Assists, Points and PIM) and be able to match up the player names with the current list I have there? I just want to be able to put things like PIM, +/-, PPP, SHP etc into the Excel sheet but not have to do all 500 players by hand.

doulos wrote:
Very interesting results though I must admit I am struggling with the D vs F comparisons still, and wondering if we are missing something there. It's just that I can't put my finger on exactly why it seems off.

I need to think on it some more.

On a secondary note, can anyone give me a bit of an idea how to copy and paste a set of last year's stats from say Yahoo into my current Excel Sheet (which only lists Goals, Assists, Points and PIM) and be able to match up the player names with the current list I have there? I just want to be able to put things like PIM, +/-, PPP, SHP etc into the Excel sheet but not have to do all 500 players by hand.

I would copy and paste all the stats in a seperate worksheet. Then make new columns in your old sheet to incorporate the new stats.

Sort everything alphabetically by player name in both sheets and then you can just cut the new stats from the new sheet and paste them in the old sheet all at once if all the same players are used in both sheets and sorted alphabetically.

I hope you can understand that. If you have any specific questions about excel I can probably help you but it's pretty tough to explain in words.

Hah okay yeah that actually was a very easy work around, I feel sort of silly now. Thanks!

Oops nevermind, it was very close to working except that Dobber's Guide lists about 500ish players and Yahoo has over 800, so I have 300 extra players in there.

doulos wrote:
Oops nevermind, it was very close to working except that Dobber's Guide lists about 500ish players and Yahoo has over 800, so I have 300 extra players in there.

Just have each worksheet open in a seperate window side by side and then delete the rows for all the players that were in the yahoo list but not the dobber. Alot easier than entering all the dobber stuff by hand IMO.

Good insight, Mikey, thanks. So essentially your weighting scheme normalizes each category to have the same numerical total value, right? This essentially accomplishes the same thing as standard scores (z-scores) without comparing the different positions relative to each other.

pooladdict wrote:
Great discussion y'all. To weigh back in on this one here (stupid personal life getting in the way of fantasy hockey -- WTF!), we've nailed down z-scores pretty well. They are a great way to describe a player's stats with respect to whatever population we're comparing with (all skaters, D, LW, whatever). BUT we're still left with the original question raised in the start of the thread: how do people weight/value categories against each other?


Hey pooladdict, the way I was thinking about it, there should be no need to weight the z-scores, since making every stat category into a z-score normalizes the data across all categories. (a z-score of 3 in Goals should be just as valuable as a z-score of 3 in assists, no?)

So 30 goals would get a player a much higher z-score than 30 assists. So it may be easier to get assists but that means it is tougher to get a high z-score right? I guess lowering SHP value is a good idea, since it can give s small number of players a large boost (in my ranking system Alfredsson is tops, mostly because of a z-score of 5.66 in SHP).