For me personally, I’m also going to continue adjusting slightly upward and downward based on the standard deviation at each positon as I look at the draft. I’ll also use change in standard deviation to put my tiers together too. Great stuff, Rudy. I look forward to chatting with you more on the message boards over the course of the year.

These were the results for the top 6 in each position:

(Total Rank Change,Total Point Shares Change)
C 14, 0.9
1B 4, 1.1
2B 6, 0.6
SS 6, 0.9
3B 9, 0.8
OF 0, 0.5

The biggest rank change was in Joe Mauer who shot up from #39 to #31 – with almost all of this driven by an increase in AVG contribution. No other hitter in the top 150 had their rank change by more than +/- 3.

Net-net, this would lead to an improvement in the data but it’s one that’s very subtle. It also leads to a number of other issues for producing the data – e.g., the query for calculating a different positional average for every player is more complicated and is problematic for players who won’t be drafted or are DH. So I’m still debating whether it’s really worth the trouble.

And don’t read into the formula being the driver for differences b/w the conventional wisdom. The more likely causes are projection differences (I’ve tied A-Rod to a bullish AB total by Fantistics that I’ll manually adjust in the March 1 version) and biases when it comes to valuing players.

When I calculated the deviation for all positions I found that 1st base actually has the second highest deviation, which also suprised me a bit. It’s somewhat counter intuitive because the lower rated 1st basemen are still relatively high value compared to other positions. It just speaks to how far superior the top 1st baseman are.

Here are the standard deviations for all positions from highest to lowest: SS 9.6, 1B 9.5, C 8.7, 3B 6.5, OF1 5.1, 2B 4.3, OF2 1.9, OF3 .83

I basically took two things away from looking at standard deviation for each positions and I’ll resummarize them here.

First, with regards to point shares, the further a player is from the mean at his position, the more he is either rewarded or penalized by being included in the mean and then benchmarked against it. In your example above you say your not sure how much of a difference it makes but it seems like an extra 8 runs and another .004 in average above the mean for the position might move the needle a little. It’s tough to tell just how much it would move it because you’d need to calculate all the point shares that way to see and as you said, that’s a tougher bit of work.

Of course every player who was not statistically identical to the average player would have an adjustment, but the point is, this adjustment becomes larger and larger the further a player is from the mean. The point share rankings are pretty tight so even a small adjustment to the outliers at the top and bottom of each category might potentially make a material impact on the rankings. I kind of think it would make more of a difference than you think. What makes me think it’s a possibility is that the positions where I see your rankings disagreeing more with conventional wisdom happen to be primarily at the positions where you have more deviation from the mean, c and 1B in particular.

The last thing is the reason I think you might want to adjust the 75/25 split, or at least tinker around with it, to account for deviation is that after you’ve normalized values by calculating point shares (getting to apples to apples) two players with the same dollar price at different positions aren’t necessarily of equal value. That’s because where dollar prices are clustered around the mean, you can’t separate yourself from the pack as much or be penalized as much relative to the players the other teams will be starting at those positions because the deviation is low. On the other hand, at high deviation positions, you get a lot more pleasure/pain at the extreme ends of the rankings.

Thanks for your detailed response, Rudy! Really appreciate the conversation.

Also, I know that there will be a few teams that choose to keep less than 5 players and draft in the 5th round.

Thanks for the help!

@Greytech Drafting: Good luck!

@Will: Do you know what pick you’ll be? Assuming keeper means you can keep for unlimited years, I’d definitely keep Cano, CarGo, and J-Up. I like the Dread Pirate and Choo but you’ve already got two OFs sewn up. B/w Zimmerman, Reyes, and Fielder, I think you can’t go wrong with any of these two. I’d probably let Reyes go as I think he’s the riskiest of the three.

@OaktownSteve: I still don’t know what you mean by STDEV. C, 1B, and SS have the most deviation in what? Some combined value? Average STDEV from the separate categories? I don’t see how 1B has that much deviation but makes some sense for C and SS.

You are right that using a positional mean that includes the player is suboptimal. But the calculation to remove each player – in effect, creating a different ‘average’ for each player, would be really complicated. Plus, it’s not clear the impact is that big.

Here are the differences for 12-team when you take Mauer and Pujols out of the average. Yes, it has some impact but there’s no sense it really hurts one position over another. (What’s worth noting is that only 12 catchers will be ‘rostered’ but a LOT more than 12 1Bs will be rostered once you consider CI and UTIL. So the 1B average isn’t as inflated as you think.)

Catcher With Mauer 70/18/77/4/.276
Catcher W/o Mauer 62/18/72/3/.272

1B with Pujols 88/28/99/4/.278
1B w/o Pujols 83/26/94/4/.277

I’ll do a separate post on some of the odder rankings. I agree Miggy is lower than I’d place him. One of the variables, though, is the underlying projections themselves (including # of ABs).

Choices: Cano, CarGo, Zimmerman, Fielder, J. Upton, Reyes, McCutchen, Choo. Any thoughts???

I wanted to start thinking of standard deviation instead of scarcity. I put the top 12 point shares at each position and checked the stdev at each and as I expected I found the highest deviation at catcher, shortstop and first base. The lowest deviation was at 2B, 3B and OF.

I think the reason why point shares tend to be slightly more wonky at positions with a high deviation is that in the calculation, each player is both contributing to and being measured against the positional mean. Mauer brings the mean way up and then gets a positional adjustment based on his relation to that average, essentially being punished for his good performance. Doumit, on the other hand, benefits from his own lowering of the mean by having a lower average to compare against. This is true across all positions, of course, but it’s particularly true where stdev is high. The further a player is from the mean, the more he benefits from or is penalized by this effect.

I would be really interested if you basically took the same methodology but compared each player to the positional average not including himself. I think this gets you a lot closer.

Finally, relevant to yesterday’s comment, I do still think that point shares, while a potentially excellent tool for comparing apples to apples value, still does not tell you enough about the relative advantages of owning a good player or punishment of owning a bad one where stdev is high (think of the matricies above). One thing you might consider toying with is instead of doing a flat 75/25 break out, weighting positional effects proportionally even higher where stdev of point shares at a position is high and reducing that proportion where it’s low.

Lastly, a word on why point shares as is are a good fit with Grey’s gut. I think almost everybody who plays fantasy misunderstands scarcity. We’re all equipped with a belief that scarcity = value. But you know there’s not a lot of elephant crap where I live, but that doesn’t make it valuable. But because drafters misunderstand scarcity, players at high stdev positions generally go way to early. So if point shares rank Mauer lower than he truly belongs, it doesn’t matter because you guys aren’t going to draft him anyway. Somebody else will take him too early. On the flip side, 1B is perceived as having low scarcity because the 12th ranked player is still high relative to the overall mean. But stdev is still high at that position because of the strength of the top tier. That’s also factored into Grey’s drafting strategy as he says he’s going to get a top rated 1st baseman no matter what, which is basically a good intuitive read on the position.

Thanks again for introducing me to point shares! Lot of hard work on your part and a good way to think about valuing players.

The big wrinkle this year is my auction team will be comamaged by my buddy. He calls me the “CEO” of the team but I know some of this razzball stuff doesn’t compute with him. And I can’t shoe him the source otherwise he’ll have my insider information at the snake draft where we’re not on the same team.

Cheers!

@Black Beard: that seems like a pretty awesome keeper list. a little OF heavy but those are great keepers.

Hamilton, Fielder, Heyward, Choo, Lincecum, and Felix.

@timmy riggins: Longoria, J-Up, CarGo, Reyes, Cano.

@Bill Lumbergh: I’d switch to 10th or 12th.

Should I swap for the guy picking 10th, the guy picking 12th, or stand pat at 16 and risk having my friend (@ #15) mess with my game plan?

Thanks for the perspective!

CI/MI/5OF

Thanks

Heyward and Hamilton for Hanley?

pitching i wouldn’t grab the first line ‘aces’ anyway so no problem there.

http://images2.memegenerator.net/ImageMacro/3050003/mind-blown.jpg?imageSize=Large&generatorName=jackie-chan

You do a good thing here Rudy. Thanks for the work. Its so bogus that ‘perts take “position scarcity” into account and arbitrarily bump players up and down spots without any rhyme or reason other than their (out of shape) guts.

I agree that what you have is a pretty good tool, especially in a draft where you have a lot of information, you have to think fast and sometimes rely on your intuition. However, I think there might be room for some refinement to you calculations. I do think that there is enough variance in stdev that your system might be slightly overvaluing or undervaluing players.

I’ve also been tinkering around with Mauer (the classic outlier). Unfortunately I have to run, but I’d like to continue the conversation. If you check back on this same comment thread tomorrow, I’ll put my thoughts up there.

Thanks again for your excellent thoughts.

I think Point Shares works better than most evaluation tools because it’s able to tie everything back to the one stat that matters – fantasy points.

While average might distort the data in extreme cases (when there are few data points in the series and/or there are dramatic differences like doing average Household Income for three HHs and one is Bill Gates’), those types of cases don’t exist in this data set. By comparing a player’s stats against the average of his position (and the average overall drafted hitter to a lesser extent), I’ve succeeded in changing the focus from his total stats to his marginal contribution – so if Mauer gets 90 RBIs but the average catcher hits 60, that could be worth more than a 1B with 110 RBIs where the average is 90 RBIs.

For sh*ts and giggles, I removed catchers #5 through #30 from my calculations to see what would happen with the top 4 catchers if the stats really dropped off a cliff. The results would be that Mauer would become #2 ($46), Buster Posey #4 ($38), Victor Martinez #5 ($37), and Brian McCann #21 ($29). As it is now, they are in the same order but #39 ($24), #50 ($22), #51 ($21), and #98 ($16).

Now what Point Shares can’t calculate directly is the concept of tiering. Let’s say you have a snake draft and deciding between a similarly valued SS and SP. But then you look and see that the next SS has much less value where there are a ton of similarly valued SPs. Common sense says take the SS. This can be calculated by looking at the projected value differences between the player and the next available player. This is something that can’t be valued ahead of time – it’s dynamic based on the drafting scenario. I know some draft software seeks to calculate this – I’ve found it’s easier just to look at draft sheets + the other rosters to make these decisions.

Let me know if I left any rock unturned….

I didn’t want to bog down the post with all the other assumptions/calculations that Point Shares is built upon. As is, I’ve left CT Old School with a sore brain.

So here it goes (and most of you can ignore):

I total up the projection sources for Point Shares (right now, Marcel, CAIRO, FanGraphs Fan Projections all weighted against Fantistics player projections) and rank players based on a prelim point total. I then make assumptions on how many players are drafted for each position including the CI/MI/UTIL slots. That’s how I get the averages for each position and the average drafted hitter.

I have modeled the league high and low totals for each category to determine how much is needed per category to gain/lose a point. Assume that the distribution between each team is the same – e.g., team 1 beats team 2 by the same difference as team 2 beats team 3, etc.) For average, I also factor in ABs.

Each player’s stat in each category is subtracted vs. the average for that position and against the average hitter. I then divide that difference into what’s needed to get a point in the category. Easier to show an example:

Jacoby Ellsbury projected at 53 SBs. The average OF in 12 team MLB has 18 SBs. The average hitter has 13 SBs. If I weight it 50/50, I get 37 (53-18) + 40 (53-13) / 2 = 38.5. I project it’s about 11.5 SBs per roto point. So his Point Shares for SB are 3.35. I add up all the 5 categories’ worth of Point Shares and have a formula to convert it into a $ value.

For the test, I just changed the positional vs. overall hitter weight across all batters and then looked at the average difference across drafted hitters (doesn’t matter what happens to hitters who aren’t drafted).

Rather than build more tests and calculations, I just used gut/experience to determine which weighting looked better but posted the results in case there’s anyone that feels the results of the 75/25 weighting didn’t look best.

Hope that makes sense…
Rudy