In partnership with CBSSports.com
Online Now 2401
Online now 2627 Record: 18710 (2/25/2012)
We aren't just committed to college football; we're early enrolling in it.
Where the madness isn't just in March.
You have no favorite boards.
The most viewed topics.
The most replied to topics.
The most up-voted topics.
The most down-voted topics.
The most up-voted posters.
The most down-voted posters.
The most followed posters.
It looks like you guys changed the formula. After looking at one team, it appears to be a standard deviation of 9 (instead of 7) on the Gaussian curve, and you're using the player's grade minus 70 (base grade), instead of the full player grade. Just want to check.
Yes, both statements above are correct. Did we not update the info box? Maybe, that's waiting until April 15th.
Follow me at: http://twitter.com/NoahStanley247 or go behind the scenes of 247Sports at http://noahstanley.com/
What does that mean for people who can't math very well?
Basically we've widened the bell curve, but decreased the value (1 for a 71 instead of 71 or .1 and .71 with the composite) so that there is still great value in your core players, but no huge benefit for signing a very large class.
We did this as a bi-product of trying to automate Basketball Recruiting Team Rankings and Recruiting Coach Rankings... both of these will be launched on April 15th.
I'm not saying the formula may or may not be tweaked in the future. There may be some outliers that we didn't predict with our small samples and testing, but the consensus among the group is that these changes have helped improve on rankings.
Thanks, Noah. I didn't see it immediately in the info box, but I honestly kind of skimmed it. Mainly because it looked the same as it did before.
Just curious - are you guys using any historical data to determine what the best formula is for rankings? i.e. comparing rankings to team performance. Or maybe just eyeballing the rankings?
The base value of each recruit is zeroed at 70. So if you have a recruit that is rated 90, he's worth a base value of 20 points. Whereas before he was worth 90 points.
Then, each player's base value is multiplied by a factor between 1 and 0. The top recruit is multiplied by 1, but every recruit after him (based on where they are when you order them highest rating to lowest) is multiplied by a decreasing factor. Increasing the standard deviation to 9 from 7 simply means that the factors are all going to be a little higher for each recruit.
I am getting what appear to be small rounding errors. Do the rankings use composite scores taken out past 4 decimal places? If so that would explain the discrepancy since we only have the 4 decimal place scores.
No big deal, I just don't wanna keep looking for my error if it isn't something I can fix.
This post was edited by DST39471 12 months ago
BTW, I like the changes in the formula. I agree with the philosophy of limiting the advantage in scoring due to class size, but it was decreasing too rapidly before.
I got the same, and assumed that for the Composite, they use more than 4 decimal places.
If you check the 247 team rankings, you should get exact values.
Here is my spreadsheet if you wanna take a look. Doesn't have every team included, but most of the contenders.
There are 3 sections for adding speculative commits, but more can easily be added.
This post has been edited 3 times, most recently by DST39471 12 months ago
Dang, that's a lot of stuff to keep updated. Kudos, man.
When you hit the refresh data button it will automatically pull all of the Commit data from the web site.
Although for some reason i keep getting errors on Arkansas, South Carolina, Miss St, and Virginia Tech, and have to go into each page and click refresh individually.
Check your PM/Inbox.
Yes, the composite rating we display is rounded, but we do use the non-rounded number for the composite score.
Let me know what the errors may or may not be for those commit lists and I'll see what I can do.
247Sports In partnership with CBS Sports