Inside CH2K8: Research and Experiments

[OSUPiper]

I have been tinkering with what sliders and settings really do and noticing nuances within how the game manages itself. This thread will be a place where I will post what I find as I work through things. Hopefully my discoveries are useful to everyone. Originally, I intended to post this in my slider thread, but I feel like it may be more useful in a general sense.

I look forward to everyone else's contributions as well.

EDIT:
I am adding a Table of Contents. Hopefully, this helps organize the posts a little more. I am grouping loosely by topic because I feel like that how people will be looking for information, but please let me know if there is a better way to catalog this stuff.


TOPICS

SLIDERS
Offensive Awareness - Page 1 - Post #3


ROSTERS AND PLAYER GENERATION
Player Ratings and Letter Grades - Page 2 - Post #12
Roster Comparisons by Position (CH2K8 VS. NCAA) - Page 1 - Post #2
Roster Comparisons by Player Height (CH2K8 VS. NCAA) - Pages 1,2 - Posts #2,8-11

[OSUPiper]

PLAYERS AND ROSTERS IN CH2K8 AND COMPARISONS TO REAL LIFE

BACKGROUND
After 30+ seasons in a CPUvCPU legacy (with house rules and edited recruits), I finally had a roster that just dominated everyone through 23 games. This roster was legitimately the best team that season, but the domination of other top tier teams did not seem right.

I had not changed the sliders for several seasons, so I knew that was probably not the issue. I started looking at my roster edits, but aside from one center, all edits were in line with computer generated players.

What I noticed, however, was how rosters were "built" by the game (i.e., CPU recruiting and roster management) versus how I recruited. Humans always have a recruiting advantage because we consider how a roster will look 2-3 season down the road.

Personally, I was struggling in the NCAA tournament because I could not recruit effective big men, so I had started targeting the best big man I could get each year regardless of skill set. Then, two years in a row, I got lucky and landed HSAAs. The next season, my team could not be stopped at all.

(I had a set of sliders which actually seemed to correct this issue by making it harder to feed the post. I liked the sliders, but they were not going to work long term for other issues.)


ROSTERS IN CH2K8 VS. NCAA
I discovered that CPU coaches will not stockpile big men, even though they are available and interested. The position and height distribution of CH2K8 rosters are very similar to real life NCAA rosters in this respect, but in real life, recruiting is not nearly as simplified. We have ratings to look at and compare and, over time, we basically know what we are getting from year to year in CH2K8 recruiting. Missing on a recruit is not as devastating to our team, if it happens to us at all.
CPU coaches appear to be programmed to mimic NCAA rosters. Also, it looks like some teams do not see the writing on the wall and fail to cut bait on a recruit they will not get, while human coaches in the late signing period tend to grab the best available player in that situation.

In short, my roster looked nothing like a CPU roster. So, I decided to see how it compared to NCAA real-life rosters. I am posting those results.


Men’s NCAA Basketball Rosters
Here is the distribution by position of 13-man NCAA rosters at small schools and power schools.

Small schools
G = 8.07
F = 4.44
C = 0.49

Power schools
G = 7.13
F = 4.95
C = 0.92

These numbers are generated by compiling rosters by position and normalizing for 13 players per roster. What these numbers mean is a small school will have about 8 guards and 5 bigs, and there is a 50/50 chance that the last big guy is a center. In contrast, power schools have 7 guards, about 5 forwards, and usually 1 center.

This makes sense to us, right? If every coach is looking for quality players, and quality bigs get swallowed up by bigger schools, then small schools are taking the best players available, which are shorter guards and forwards. Obviously, there are more of these players to go around as well. Small schools and mid-majors often have guards that could play and succeed in power conferences.

To be sure, CH2K8 does not do a great job recreating this. There are not enough real "busts" in the game, and it is rare for power schools to be fighting with small schools, even for low-tier players. If you rank players by overall rating, you are going to see an almost linear correlation between (1) conference type (power to small) and (2) player rating.

Graphically, this is what my sample looked like when distributed by player height:

Distribution by height
Power school Gs
5'8 ***
5'9
5'10 ******
5'11 ********
6'0 ****************
6'1 *******************
6'2 *********************************
6'3 ***************************************
6'4 ******************************
6'5 *********************************************
6'6 ****************
6'7 *******
6'8 ***

Distribution by height
Power school Fs
6'4 *
6'5 *******
6'6 *****************
6'7 ************************
6'8 ***********************************
6'9 *****************************
6'10 **************************
6'11 ***********
7'0 *****
7'1 *

Distribution by height
Power school Cs
6'9 **
6'10 ******
6'11 *******
7'0 *********
7'1 ***
7'2 *
7'3
7'4 *
　
Distribution by height
Small school Gs
5'5 *
5'6
5'7
5'8 **
5'9 ***
5'10 ***********
5'11 ************
6'0 *************************
6'1 ************************************
6'2 *******************************************
6'3 **************************************
6'4 *******************************
6'5 ************************
6'6 **************
6'7 ****
6'8 ***

Distribution by height
Small school Fs
6'3 ***
6'4 ***
6'5 ******
6'6 ************************
6'7 *************************************
6'8 ****************************************
6'9 **************
6'10 ******
6'11 **
7'0 *

Distribution by height
Small school Cs
6'9 *****
6'10 ******
6'11 **
7'0 *
7'1 *

As an initial point, schools decide who is a guard, forward, or center. Many players don’t fit in one spot, and many players would play a different position at a different school. So, take this for what it is worth by position, but overall, the data is very helpful in comparing NCAA rosters with CH2K8.

Note that power school guard distribution has two mode heights (that is, most frequently occurring) – we would expect more of a curve, which exists in every other position at power schools, and every position at small schools. My theory is that you have more 2/3 swingmen at power schools, so taller guards and small forwards essentially give us two separate curves mashed together in this case, explaining the slight dip at 6'4". This also explains why there are more guards at 6'5" than at 6'3", which is not otherwise intuitive.

It appears that power conf. guards are usually 6'0" to 6'5". G/F swing players, generally perimeter guys, will range from 6'3" to 6'7" mostly. Most forwards are 6'7" to 6'10", and could be stretch guys but are most likely frontcourt players.
Small school guards run about an inch shorter, but a 2/3 swing at a power school is probably playing straight 3 at a small school just out of necessity. So you can argue that guard height is pretty consistent across the board.

The real difference is at F and C. Small school forwards are mostly 6'6" to 6'8". Notice that centers at small schools are always at least 6'9. Again, my opinion is that big guys at small schools are probably playing positions based on height as much as skill set, while power schools have more bigs overall, so they can position by skill more effectively.
　
CH2K8 Roster Distribution
Because we can, I broke down CH2K8 rosters across all five positions:

Power schools
PG 3.7
SG 3.6
SF 2.7
PF 2.0
C 0.9

Position comparison - Power schools
NCAA 2K
G 7.13 7.3
F 4.95 4.7
C 0.92 0.9

Assuming we can combine all guards as Guards, and all forwards as Forwards, this looks exceptionally close to real life.

Small schools
PG 2.8
SG 3.1
SF 2.8
PF 2.3
C 2.0

Position comparison - Small schools
NCAA 2K
G 8.07 5.9
F 4.44 5.1
C 0.49 2.0

With the same assumptions present, small school roster generation in CH2K8 is way off. This should not surprise anyone, as the game generates enough big men, but most are just awful so they slide to low tier schools. This is so prevalent that small schools AVERAGE 2 centers per team. (One team had 3 total guards and 5 centers.)

You would need to edit the average small school roster by turning 1.5 centers and 0.5 forwards into guards; however, this is a four-year roster and you would likely just be editing incoming freshmen each season if you were fixing this problem. I would use frequencies to do this.

For example, look at centers. An average of 0.49 centers on NCAA rosters is about one center per two teams. This means, on average, a small school lands a true center once every eight years in real life!

HOW DO WE USE THE DATA?
This stuff may be helpful to some of you, but it is interesting at the very least. Personally, I will develop some scheme to edit small school rosters in-game to achieve more realistic rosters. This will translate to more realistic seasons as well. Additionally, I will implement recruiting rules that follow these distributions more closely.

At small schools, I may look at not targeting any center when one is on my roster. Then, since your target list is essentially the same as your roster size, I might double the NCAA average for forwards (4.44 x 2 = 8.88, or about 9), then add 1 target (9 + 1 = 10), and that would equal my total forwards on my roster PLUS my targets. In other words, If I have 4 forwards, I could have 6 more targeted as recruits. I’m obviously still working this part out.

As always, I love to hear input from the forum.

[OSUPiper]

SLIDERS - OFFENSIVE AWARENESS


This slider appears to affect:
- Free throw %
- FG%
- Floor spacing
- pace of ball movement in the halfcourt offense
- attempts to feed the post
- frequency of high-level offensive plays and animations
- dunk attempts


Notes: While I really like the look and feel of this slider at higher numbers, the game gets too arcade-ish. The best part of the low values of OA is deliberate pace of play, better shot distribution, and more realistic (but not perfect) FT%.

[tril]

Quote: Originally Posted by OSUPiper

PLAYERS AND ROSTERS IN CH2K8 AND COMPARISONS TO REAL LIFE BACKGROUND After 30+ seasons in a CPUvCPU legacy (with house rules and edited recruits), I finally had a roster that just dominated everyone through 23 games. This roster was legitimately the best team that season, but the domination of other top tier teams did not seem right. I had not changed the sliders for several seasons, so I knew that was probably not the issue. I started looking at my roster edits, but aside from one center, all edits were in line with computer generated players. What I noticed, however, was how rosters were "built" by the game (i.e., CPU recruiting and roster management) versus how I recruited. Humans always have a recruiting advantage because we consider how a roster will look 2-3 season down the road. Personally, I was struggling in the NCAA tournament because I could not recruit effective big men, so I had started targeting the best big man I could get each year regardless of skill set. Then, two years in a row, I got lucky and landed HSAAs. The next season, my team could not be stopped at all. (I had a set of sliders which actually seemed to correct this issue by making it harder to feed the post. I liked the sliders, but they were not going to work long term for other issues.) ROSTERS IN CH2K8 VS. NCAA I discovered that CPU coaches will not stockpile big men, even though they are available and interested. The position and height distribution of CH2K8 rosters are very similar to real life NCAA rosters in this respect, but in real life, recruiting is not nearly as simplified. We have ratings to look at and compare and, over time, we basically know what we are getting from year to year in CH2K8 recruiting. Missing on a recruit is not as devastating to our team, if it happens to us at all. CPU coaches appear to be programmed to mimic NCAA rosters. Also, it looks like some teams do not see the writing on the wall and fail to cut bait on a recruit they will not get, while human coaches in the late signing period tend to grab the best available player in that situation. In short, my roster looked nothing like a CPU roster. So, I decided to see how it compared to NCAA real-life rosters. I am posting those results. Men’s NCAA Basketball Rosters Here is the distribution by position of 13-man NCAA rosters at small schools and power schools. Small schools G = 8.07 F = 4.44 C = 0.49 Power schools G = 7.13 F = 4.95 C = 0.92 These numbers are generated by compiling rosters by position and normalizing for 13 players per roster. What these numbers mean is a small school will have about 8 guards and 5 bigs, and there is a 50/50 chance that the last big guy is a center. In contrast, power schools have 7 guards, about 5 forwards, and usually 1 center. This makes sense to us, right? If every coach is looking for quality players, and quality bigs get swallowed up by bigger schools, then small schools are taking the best players available, which are shorter guards and forwards. Obviously, there are more of these players to go around as well. Small schools and mid-majors often have guards that could play and succeed in power conferences. To be sure, CH2K8 does not do a great job recreating this. There are not enough real "busts" in the game, and it is rare for power schools to be fighting with small schools, even for low-tier players. If you rank players by overall rating, you are going to see an almost linear correlation between (1) conference type (power to small) and (2) player rating. Graphically, this is what my sample looked like when distributed by player height: Distribution by height Power school Gs 5'8 *** 5'9 5'10 ****** 5'11 ******** 6'0 **************** 6'1 ******************* 6'2 ********************************* 6'3 *************************************** 6'4 ****************************** 6'5 ********************************************* 6'6 **************** 6'7 ******* 6'8 *** Distribution by height Power school Fs 6'4 * 6'5 ******* 6'6 ***************** 6'7 ************************ 6'8 *********************************** 6'9 ***************************** 6'10 ************************** 6'11 *********** 7'0 ***** 7'1 * Distribution by height Power school Cs 6'9 ** 6'10 ****** 6'11 ******* 7'0 ********* 7'1 *** 7'2 * 7'3 7'4 * 　 Distribution by height Small school Gs 5'5 * 5'6 5'7 5'8 ** 5'9 *** 5'10 *********** 5'11 ************ 6'0 ************************* 6'1 ************************************ 6'2 ******************************************* 6'3 ************************************** 6'4 ******************************* 6'5 ************************ 6'6 ************** 6'7 **** 6'8 *** Distribution by height Small school Fs 6'3 *** 6'4 *** 6'5 ****** 6'6 ************************ 6'7 ************************************* 6'8 **************************************** 6'9 ************** 6'10 ****** 6'11 ** 7'0 * Distribution by height Small school Cs 6'9 ***** 6'10 ****** 6'11 ** 7'0 * 7'1 * As an initial point, schools decide who is a guard, forward, or center. Many players don’t fit in one spot, and many players would play a different position at a different school. So, take this for what it is worth by position, but overall, the data is very helpful in comparing NCAA rosters with CH2K8. Note that power school guard distribution has two mode heights (that is, most frequently occurring) – we would expect more of a curve, which exists in every other position at power schools, and every position at small schools. My theory is that you have more 2/3 swingmen at power schools, so taller guards and small forwards essentially give us two separate curves mashed together in this case, explaining the slight dip at 6'4". This also explains why there are more guards at 6'5" than at 6'3", which is not otherwise intuitive. It appears that power conf. guards are usually 6'0" to 6'5". G/F swing players, generally perimeter guys, will range from 6'3" to 6'7" mostly. Most forwards are 6'7" to 6'10", and could be stretch guys but are most likely frontcourt players. Small school guards run about an inch shorter, but a 2/3 swing at a power school is probably playing straight 3 at a small school just out of necessity. So you can argue that guard height is pretty consistent across the board. The real difference is at F and C. Small school forwards are mostly 6'6" to 6'8". Notice that centers at small schools are always at least 6'9. Again, my opinion is that big guys at small schools are probably playing positions based on height as much as skill set, while power schools have more bigs overall, so they can position by skill more effectively. 　 CH2K8 Roster Distribution Because we can, I broke down CH2K8 rosters across all five positions: Power schools PG 3.7 SG 3.6 SF 2.7 PF 2.0 C 0.9 Position comparison - Power schools NCAA 2K G 7.13 7.3 F 4.95 4.7 C 0.92 0.9 Assuming we can combine all guards as Guards, and all forwards as Forwards, this looks exceptionally close to real life. Small schools PG 2.8 SG 3.1 SF 2.8 PF 2.3 C 2.0 Position comparison - Small schools NCAA 2K G 8.07 5.9 F 4.44 5.1 C 0.49 2.0 With the same assumptions present, small school roster generation in CH2K8 is way off. This should not surprise anyone, as the game generates enough big men, but most are just awful so they slide to low tier schools. This is so prevalent that small schools AVERAGE 2 centers per team. (One team had 3 total guards and 5 centers.) You would need to edit the average small school roster by turning 1.5 centers and 0.5 forwards into guards; however, this is a four-year roster and you would likely just be editing incoming freshmen each season if you were fixing this problem. I would use frequencies to do this. For example, look at centers. An average of 0.49 centers on NCAA rosters is about one center per two teams. This means, on average, a small school lands a true center once every eight years in real life! HOW DO WE USE THE DATA? This stuff may be helpful to some of you, but it is interesting at the very least. Personally, I will develop some scheme to edit small school rosters in-game to achieve more realistic rosters. This will translate to more realistic seasons as well. Additionally, I will implement recruiting rules that follow these distributions more closely. At small schools, I may look at not targeting any center when one is on my roster. Then, since your target list is essentially the same as your roster size, I might double the NCAA average for forwards (4.44 x 2 = 8.88, or about 9), then add 1 target (9 + 1 = 10), and that would equal my total forwards on my roster PLUS my targets. In other words, If I have 4 forwards, I could have 6 more targeted as recruits. I’m obviously still working this part out. As always, I love to hear input from the forum.

great information. this is great analysis. In My Opinion this is thesis worthy.
I wonder if developers do in depth analysis like this. this info could be applied to the programming behavior of ai controlled teams in all sports games

great stuff man. you get multiple thumbs up from me.

[VandyRedskins21]

Another aspect that affects all of this seems to be the "Recruiting Priorities" of the teams. I have seen some teams that have had up to 6 or 7 centers solely because size was a priority. I wonder if someone put every single team's recruiting priorities to "None" if that balances out the teams' position needs.

Also, I wonder if that actually worked, if that was an oversimplification, and if there would be a better solution. Has anyone tinkered with this before?

[OSUPiper]

Thanks, tril.


I get more out of these games because of everyone sharing on the forums, so I am happy to contribute what I can.

[OSUPiper]

Quote: Originally Posted by VandyRedskins21

Another aspect that affects all of this seems to be the "Recruiting Priorities" of the teams. I have seen some teams that have had up to 6 or 7 centers solely because size was a priority. I wonder if someone put every single team's recruiting priorities to "None" if that balances out the teams' position needs. Also, I wonder if that actually worked, if that was an oversimplification, and if there would be a better solution. Has anyone tinkered with this before?

I thought someone had a set of rosters with coaches edited to not use height as a recruiting priority. In any case, it is an issue for sure. I never thought of eliminating priorities, but I like it.

[OSUPiper]

Here is a comparison side by side of distribution by height for teams in a power conference. The sample sizes are slightly different, since I am really just doing this as a demonstration, but I have normalized the results... so these graphs are informative even if not completely accurate.


* denotes NCAA
x denotes CH2k8


Distribution by height
Power school Gs
5'8

***
5'9
xxxxxx

5'10
xxxxxx
******
5'11
xxxxxx
********
6'0
xxxxxxxxxxxxxxxxxxxxxxxx
****************
6'1
xxxxxxxxxxxxxxxxx
*******************
6'2
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
*********************************
6'3
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
***************************************
6'4
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
******************************
6'5
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
*********************************************
6'6
xxxxxxxxxxx
****************
6'7
xxxx
*******
6'8

***

Distribution by height
Power school Fs
6'4
x
*
6'5
xxxxxxxxxxxxxxxxxxxxxxxx
*******
6'6
xxxxxxxxxxxxxxxxxxxxxxxxx
*****************
6'7
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
************************
6'8
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
***********************************
6'9
xxxxxxxxxxxxxxxxxx
*****************************
6'10
xxxxxxxxxxxxxxxxx
**************************
6'11
x
***********
7'0

*****
7'1

*

Distribution by height
Power school Cs

6'8
x

6'9
xxx
**
6'10
xxxxx
******
6'11
xxxxxxxx
*******
7'0
xxxxxx
********
7'1
xxxxxx
***
7'2

*
7'3


7'4

*

Looking at the distribution by position, Guards seem to be in a good place. From this data, you could argue that Forwards should be an inch taller (but see analysis of height over the entire roster in next post). Center height also appears to be about right -- small sample size with a tight distribution makes this a tough call.