Posted by seble on 8/25/2010 8:54:00 AM (view original):
The # percentages are normalized, which means they're adjusted based on the league average for that season.  That helps some of the old guys, but it's not going to suddenly make a 40% shooter a 55% shooter.  For example, 51-52 Al McGuire gets a boost from 47.3% actual 2 point percentage to 49.7% normalized 2 point percentage.  The league average that season was 37.3%.

I don't understand the logic of this...
If a guy shot 125% better than the league average, how come you're only increasing him a couple of points?  If your normalization point is 45%, which again, it should be around that, he should be shooting 57%!

That's where the issue lies... you say you're "normalizing" the older players, but you aren't.  You're nerfing their rebounds but not boosting their fg% enough.  If someone shoots 20% better than the league average, then their fg%# should be 20% better than the normalization point (across all eras - you know, the reason for normalization?).  By the same token, someone like Charles Barkley, who's 133% fg+ should be against that same number.  I've used guys from both eras.  There's no way you're using the same number for 87-88 Charles Barkley and 59-60 Elgin Baylor. 

If you say you are, you're either lying or didn't do something right, because it's broken.

So, again - what is the normalization point?  My guess would be that you'd take the last five seasons (maybe ten) and take a league average over that span.

The league average for the last five years:

09-10 - .461
08-09 - .459
07-08 - .457
06-07 - .458
05-06 - .454

So, as you can see, your normalization fails.

The ultimate way would be to take the average fg% for the entire history of the nba and normalize to that... but since most people are more familiar with the modern game, I suggest the past five years.  You'll still see guys like Barkley & McHale's fg% go down, while guys like Baylor & Cousy's go up.

I understand that fully normalizing FG% basically would make the old players play much differently than they did in real life.  There is something to be said for maintaining their flaws as long as the salaries are made to adjust.  However, there are two other factors that come into play that could also be adjusted to compensate.  First of all, those lower FG% of old players were in part due to a different style of game based upon the rules in place.  This should probably be taken into account in the determination of Defensive Rating to favor these players.  No, they weren't necessarily more skilled defensive players, and much of what they did to disrupt an opponent would be a foul today (heck much of it was a foul back then, which is why their foul rates are much higher).  This also brings to mind that fouls should be used in determining Defensive Rating - if a player fouls a lot, he will be more difficult to score against because not every foul gets called (and the player is already punished for his high foul rate).  Plus, one reason players get a lot of fouls is when they are guarding the other team's better players because they are better defenders than their teammates.  The correlation is hardly perfect I realize, but it should be considered, especially when there are so few other stats to consider in setting Defensive Rating.  (Next step - make the Defensive Rating more effective in the game!)

The other factor would be assists.  Players today generally have higher FG% in part because they play in a much more assist heavy environment than existed previously.  Right now, those players with high assists give further bonuses in the sim to FG%.  But this doubles up on the "assist benefit".  For example, a Peja played on a team with above average assist environment, which is one reason he has such a high eFG%.  Now, play him with high assist players in the sim, and his eFG% is boosted even more.  To normalize assists, you would determine the actual assist environment of each player in the database (calculated in some fashion and likely excluding his own assists from the calculation) and compare it to the assist environment put in play with a particular lineup into the sim.  To the extent the sim environment is greater, eFG% should rise.  To the extent it is lesser, eFG% should decline.  Thus, assists should aid older players more than more recent players, helping to offset the lower FG% they had in real life.  Just some ideas to consider should an actual adjustment to the sim be in the works.

ETA:  It has obviously been awhile since I posted here.  My post count restarted. 

i totally agree with that ash.  we've got to get the old greats like cousy and russell from the NBA top 75 back into play in this SIM.  it doesnt do any good to normalize a player over only their season.  the style of play was different from one era to the next which contributes to the difference in FG averages.  Additionally, refs were calling games more tightly back then as well which led to higher numbers of fouls.  Do you think they should normalize fouls and turnovers as well?  or are they already and I'm not aware of it.

IIRC any stat in the database that has a "#" is normalized in some way.

Ash's comments about it not working right are actually an old discussion from years ago.  The normalization done in baseball (i think) is different than here.  Some of it is because many stats are estimated (blocks, steals, 3FGAs for older players).  I could be wrong though.  But I do know for a fact that we deeply discussed thd way rebounding was normalized and it was not "correct".  Its probably still the case.... but I say that not knowing how to fix it.  lol.

I guess I want to know if the baseball sim's normalization is done the same way as the hoop's sim.  It seems to work (normalization) in the baseball sim... at least much better than it does in the hoops sim.  Then again I do not troll the baseball forums so for all I know its not "right" there either. 

I've been critiqued at my job for using too much technical jargon when trying to explain things to customers.  This point will be relevant shortly.

If WIS truly intends to attract new users to the NBA sim, then they would want to keep the core statistics rather simple.  A first-time NBA sim user would be completely confused if all of the stats were normalized.  They look up 51-52 George Mikan's FG% on a historical stats database, and it is nowhere near what WIS shows as his FG%.  Meanwhile, the FG% displayed for Tyson Chandler is significantly lower than what his IRL FG% was.  If that were the case, a newcomer would be turned off pretty quickly.

We can't revamp the entire sim to utilize normalized stats, because we'll alienate any new users who would otherwise be interested.  The NBA sim has lost good players in the last year like Tiger Woods has lost fans, and making the game more complex won't get any newbies interested.

Back to my first point:  K.I.S.S.

I love how everyone does the work for seble and they still charge to play.  Its really crazy. 

Here's a hint to creating good salaries - try building a team once a century.  When you realize you play against the same 10 players you might figure out that they are too cheap. 

as for marketing... what is the question that's being asked when you ask "What if..."     Is it,,,  "What if any player in history could have played on the same team?"  or is it "What if the legends could play in today's game?"   Normalizing yesteryears player stats to today IMO is like asking what if Cousy could play today.  Do you really think he'd be one of the worst shooters in the game today?  I don't think so, cause the entire style of play is different and he would adapt.  And not just shooting either, you think he'd foul out more often than todays players or commit more turnovers than Stephon Marbury.  No, because they're not calling the games the same as back then.  You almost have to normalize the stats just to put them on the same floor.  I think if you market WIS as "What if the legends could play in today's game" then you could justify normalizing the stats with a blanket statement "we adjusted their salaries and stats to todays numbers to see how they'd play today"    By that same token, if you ask "What if Shaq played in the 60's" ... then you would need an completely new SIM formula.  The same SIM formula can't work in both ERAs.

Posted by benjihall on 8/26/2010 5:37:00 AM (view original):
I love how everyone does the work for seble and they still charge to play.  Its really crazy. 

Here's a hint to creating good salaries - try building a team once a century.  When you realize you play against the same 10 players you might figure out that they are too cheap. 

Benji

I really disagree with your first point, like the collaborative effort here.  It would be a lot worse if they came of with something that no one liked.  Having consumer input is very important and the only way to go.  But your second statment is spot on.

We use a log5 normalization technique that is the same as is commonly done in baseball to compare eras.  Here's a brief explanation that our stats guy (tinmanpb) gave in relation to baseball: 

"For example, if a .303 hitter from the National League in 2007 - when players hit .266 - faces a .303 OAV pitcher from the same league and year, the hitter will hit much better than .303 because the batter is far better than the league average and the pitcher is far worse than the league average. Whereas, if a .303 hitter from the National League in 1930 - when players hit .303 - faces a .303 OAV pitcher from the same league and year, the batter will hit .303."

The same methodology was used for NBA.  So, performance in the engine isn't directly equal to the FG%# values, but also depends on the defenders' as well and how they compare to the average for their season.

The only thing we normalize is shooting percentages, everything else is simply pace adjusted. 

I've said it before, and I'll say it again (I'm sure), but I thought the "simplier" SIM (without def, efg, etc) was a much better product.  I'd also say that since there weren't all these "fringe" things used, there was a lot less to "discuss" as far things that needed drastic ghanges.  As noted above by someone else K.I.S.S..

The changes that have been made over the years were in response to you guys wanting a better game.

Posted by seble on 8/26/2010 11:39:00 AM (view original):
We use a log5 normalization technique that is the same as is commonly done in baseball to compare eras.  Here's a brief explanation that our stats guy (tinmanpb) gave in relation to baseball: 

"For example, if a .303 hitter from the National League in 2007 - when players hit .266 - faces a .303 OAV pitcher from the same league and year, the hitter will hit much better than .303 because the batter is far better than the league average and the pitcher is far worse than the league average. Whereas, if a .303 hitter from the National League in 1930 - when players hit .303 - faces a .303 OAV pitcher from the same league and year, the batter will hit .303."

The same methodology was used for NBA.  So, performance in the engine isn't directly equal to the FG%# values, but also depends on the defenders' as well and how they compare to the average for their season.

The only thing we normalize is shooting percentages, everything else is simply pace adjusted. 

This doesn't make sense in a basketball frame where
1) It's not a one-on-one game
2) Defensive ratings don't really do much
3) Defensive ratings are inconsistent and at times ridiculous

one is true because:  straight from the Knowledge Database:

#130	Gameplay	Q.	How do defensive ratings affect the SimEngine results?
		A.	The defensive rating of a particular player impacts the FG% of the player he is guarding as well as a lesser effect on the FG% of the other 4 offensive players, which simulates help defense.

Two is true because of one.  Otherwise, anyone matched up on 96-97 Glen Rice (20 def) should absolutely go off.  But they don't.  Not even close.

Three is true because... you've heard the arguments.  All-time defenders being 90+ for most of their career, but sometimes a season in the middle will be a flat 50 (where did this come from?).  It's true because of counting Rebounds in the rating.  It's true because of 08-09 Troy Murphy, and 08-09 Zach Randolph, and almost any 90s Charles Barkley...


This methodology for fg% normalization flat out makes no sense.  Why not do it based on a universal league average, determined either by the last few seasons (as to not throw off new users who won't understand why Tyson Chandler won't shoot 60%) or by the entire span of the league (if you really wanted to be accurate?).  Once you have these numbers determined, then you let the defensive rating come into play.  Now that you have your 2pt fg%# of 57% guy shooting against a 90 defender, what does he really shoot like?  Against a 50 I'd say he'd shoot like normal.  What about against a 20 defender?

Churn out the number and then let the Defensive Rating do it's work.  You're putting too much into a very simple concept.


ALSO

People talk about fouls in the old era... they bring up multiple reasons as to why fouls were higher, but they leave out the simplest one:  THERE WERE FAR MORE POSSESSIONS because of pace.  Are fouls pace adjusted in this sim? Because I don't think they are...

oh and nice of nabo to show back up.  He always has such helpful input with his vast sim experience.

I really disagree with your first point, like the collaborative effort here.  It would be a lot worse if they came of with something that no one liked.

Collaboration is great, but WhatIf needs to produce some effort on there side first, or else all their work is done for them.  Seble has had years of feedback, if he actually reads it, to know what the current problems are.  We are then told he is working on sweeping changes and the end result is worse than the current?  This is crazy.  If they really want the power users to help fix it they should at least let them play for free.

Posted by seble on 8/26/2010 1:06:00 PM (view original):
The changes that have been made over the years were in response to you guys wanting a better game.

More complex, with more errors, and more arguing about which way to do things is not better.

Just look at number of users; that should tell you which version your customers though was better.

It's obvious that even the users themselves can not agree on how to reflect and value many of the features that were added, so why not just eliminate them?