FOF 7 - Initial Draft testing

[w24olfpack]

Started playing around with the combines and draft tonight.

My working assumption of FOF 2007 was that the combines for a player were built on a possible distribution for each rating column and rating.
SO a QB with a 625 Accuracy and Timing in a draft file could come up with a position drill with a median of x and a std deviation of y. So you could have a QB with a 625 in those and a PD between say 95 and 80.

THis is one of the things that made drafting more diffcult and where Malcpow's initial work helped break some of that out.

I decided to test with the new COMBINE accuracy.

Hypothesis
If the combine accuracy is set to a smaller number the mean of the combine for that particular skill will remain the same but the standard deviation will increase.

Essentially you will get a larger spread from high to low in the draft file for that combine attribute.


Here are the intial settings for the test.
Built files of 880 players - All ratings 625.
55 each position (DT had 60 - ooops).
Ran the import each time for Combine accuracy settings of 0, 25, 50, 75 and 100 (i'm guessing the most commom settings going forward).
TCY Loaded and licensed.
All HT / WT at average for position in FOF 2007
Alll went to Clemson
All from Philly
All have volatility and intel of 50
Combine skippers excluded in analysis

[w24olfpack]

Looks like the hypothesis holds up on the QB.s
Average SOL for a 625 rated QB is around 39. It's a little lower on the 100 COMBINE ACCURACY setting.
The spread of values (standard deviation goes up considerably)... from 3 score points at 100 to 7 at 0.
The range at 100 is 32 to 45 so 13 points. At 0 its 28 to 53 for 25 points.

So there is a big difference depending on the setting. Looks like 100 and 75 are pretty tight and true on the combines. 50 starts letting the error creep in and 25 and 0 I would say becomes a crapshoot.



http://www.fof-tfl.net/images/625_combine_chart_QB.jpg

[w24olfpack]

Data for the WR with a 625 rating.


http://www.fof-tfl.net/images/625_combine_chart_WR.jpg

[w24olfpack]

Something a little more visual
Distribution of each Combine for QB at 625

http://www.fof-tfl.net/images/QB625.jpg

[w24olfpack]

Distributions for WR at 625


http://www.fof-tfl.net/images/WR625.jpg

[zbuckley]

this is very cool to see wolf and good stuff. I'm wondering if you need more players? I only say that because i'm guessing (without any evidence) that he didn't change the player creation process but just added the x factor. So my thought is that it's strange that your 100% accuracy #'s don't totally match Malcpow's? For example 7.8 agility can produce a 625 score in 2k7 but you didn't get that in FOF7. Of course it's possible there's new breakpoints but I wouldn't think 55 players is enough to make that assumption. Let's get some statistic majors in this thread.

[w24olfpack]

55 is actually a pretty good sample size when it comes to statistics.

THis however was put here for more discussion than actual samples.

The biggest issue I have with using Max and Min (as were in Malc's original analysis) is you deal with outliers. A 7.8 MIGHT give you a 625 QB but it may be a 1 in 1,000 event. I would play more to the mean and standard deviations. It can give you a good 97.5% or 99% certainty while taking out the outliers.

Also, look how the spread increases with smaller combine accuracy numbers. We can ONLY guess that the old game was on 50.


Will be interesting when I can get the rest of this run.

[zbuckley]

55 is actually a pretty good sample size when it comes to statistics.

THis however was put here for more discussion than actual samples.

The biggest issue I have with using Max and Min (as were in Malc's original analysis) is you deal with outliers. A 7.8 MIGHT give you a 625 QB but it may be a 1 in 1,000 event. I would play more to the mean and standard deviations. It can give you a good 97.5% or 99% certainty while taking out the outliers.

Also, look how the spread increases with smaller combine accuracy numbers. We can ONLY guess that the old game was on 50.


Will be interesting when I can get the rest of this run.

I had that thought as well. If 50 was what 2k7 was on that's very interesting. No doubt it's safer to pay attention to the mean and standard deviations but that's not hard to do. It's finding those outliers that i'm interested in doing and those are the guys that fall into the later rounds......

[zbuckley]

It's funny we've had this conversation many times. I say drafting is an art and you try to put it into #s

[w24olfpack]

I want the outliers as well

If you can say that a number is 95pct confident to be an outlier that is close enough. I'll take my chances with the rest

The old A AA and AAA charts are golden but it's also good to know if there is a bottom to a ratings range (like bench in QBs or broad jump )

[aston217]

1) the 7.80 tackle score only showed the potential to produce around a 70 (550) rating in 2k7.

2) why is 55 players a good sample size? Just curious.

[w24olfpack]

55 is going to give you a good representation of the central tendenancy (Mean) and the spread (Standard Deviation).

As with anything, more data more better but you get the bang for the buck with 55.

When I roll this out, I'll probably go with about 325. That will be 5 differents runs with 990 players so that will make for a nice dataset.

IF the 95% percentiles for Bench is 16 I don't need to know if 17 or 18 is the biggest outlier. I just need to know that 17 is a stud,

[zbuckley]

Yeah the upside #'s are not as important it's the bottom of the range #s that is important. Knowing that if you draft a WR with agility above 7.20 you're getting a WR with GD below 30 helps eliminate so many turds in the draft.

I'm curious how whole combine accuracy thing will play out.

2k7 WR agility 7.21 = 20ish

say with combine accuracy at 0 could you have a WR with 60 GD in FOF7 if he had the same agility score?

[Ben E Lou]

I posted this elsewhere, but it's probably worth adding to this particular thread: even at very low combine accuracy scores, the basic constraints are still applicable. It's just that you'll find that there are more good players with combines at the bottom of their constraints and bad players with combines near the tops of theirs. I'm finding it more fun in SP (and in the WOOF, which uses 25) to use a lower setting than 50. And yes, I tend to agree that 50 is about what we were used to in FOF2K7.