What makes a stat impressive?

[Leftos]

For those of you that don't know me, first of all hi, and second of all, I'm developing a basketball stat analysis tool, named NBA Stats Tracker. This question will really help me develop it further, but it's a general basketball question.

So, for each game played, there's a box score. I want to create a sort of "Best Performers" screen to show in-a-glance the best 3 players of each team for that game (and then further it for multiple games).

I'm thinking of using the Game Score metric in order to pick 3 players from each team.

However, I don't want to show their full stat line in that screen. I want to pick three to five stats to display, depending on how impressive they are. I know I could just pick the stats per position, but what if a Center had an impressive game due to Points, Assists and Steals and not their rebounds? I know, it doesn't happen often, but it does happen.

So, any ideas on what stats I should pick to show for each player, or how to pick them based on how "impressive" they are? What makes a certain stat impressive? Is there a certain threshold above which each stat is impressive? (For example, if a player has over 7 assists, or over .500 FG% with over 5 FGs made, mark it as impressive and show it independently of their position).

[benji]

The last thing.

I remember the NHL series had a "three stars" thing and it just showed the "highest" stats for the players it picked. Even if you had something like a guy with nothing but four assists.

Thresholds seems like a good measure. No idea if you can set it per position because 7 assists from a center is more impressive than 7 from a guard. So on.

The game does a surprisingly good effort at picking the "vital" stats in the signature highlights or whatever the fuck it is called now that they took it away from me in MyPlayer to watch over and over as I masticate.

*orderlies come in...*

[Leftos]

I know the game does a pretty good job in that screen, but I have no access to its source code to know what thresholds and criteria it uses. That's why I'm asking. Thanks anyway.

[Nick]

Well, statistics are all relative to eachother. You can't have "good" without the "bad". It's a matter of relativity. So if you want tell your stats tracker thing to list the most impressive stats, then you first need to have a big ass sample size of stats to compare against, so that it knows what measures as "impressive" or not.

So i would suggest that this stats tracker thing would need to "know" the enter statistical NBA database. Atleast of the modern era. Then it knows that if x amount is the average APG (or whatever stat) for a point guard, and this player currently has y amount of assists, then it ranks as this significant on the impressive-stat-scale.

[benji]

You don't need any of that, b-r exists.

Just set a series of guidelines, since you're already calculating the game score you only have to check three players. If rebounds > 15 or if steals > 3 or if blocks > 4 so on and so forth. Set it as a % over the average after.

Or just display the five stats, points, rebounds, assists, blocks, steals.

[Leftos]

I might initially go with those 5 stats you mentioned, Benji. But keep the suggestions coming, I appreciate every idea and want to make something sophisticated, so that for example I could show FG% if it is really significant, and remove one of the 5 basic stats (the less impressive of the lot) to get FG% in its place.

Nick had a good idea, but Benji has a point as well. I don't need the entire database of NBA stats in the tool, I just need to compile the averages myself and get them in the tool as thresholds. If I can compile an average of each stat, any player that goes over it, should immediately have one of the basic 5 stats replaced with the impressive one.

Any idea how I would go about compiling that list, Nick?

[Nick]

Just set a series of guidelines, since you're already calculating the game score you only have to check three players. If rebounds > 15 or if steals > 3 or if blocks > 4 so on and so forth. Set it as a % over the average after.

That's a good method, but there still has to be some sort of ranking function though right? Because otherwise what do you do if multiple players meet the guidelines/filter? Or what happens if zero players reach the guidelines? Something in the tool has to rank stuff aswell, so that there is prioritized order in the stats.

Any idea how I would go about compiling that list, Nick?

Not really, but www.basketball-reference.com would be the first place i'd start.

[benji]

Play with these.
http://www.basketball-reference.com/pla ... finder.cgi
http://www.basketball-reference.com/pla ... finder.cgi

Trust me, it's not as absurd as it seems.

That's a good method, but there still has to be some sort of ranking function though right? Because otherwise what do you do if multiple players meet the guidelines/filter? Or what happens if zero players reach the guidelines? Something in the tool has to rank stuff aswell, so that there is prioritized order in the stats.

He's already picking the three players through game score, this is about which stats to display to justify the picking of the top three players.

Unless I'm reading Leftos wrong.

[Leftos]

Yeah, I won't be picking the players based on who has the most impressive stat, but I'm first using Game Score to pick 3 from each team.

So Phase 1 is going through both teams, calculating each player's Game Score, and then picking 3 from each.
Phase 2 is going through each of these 6 players statlines, and seeing which 3-5 of their stats out of their whole statline are the most impressive, so that I'll show those 5 instead of always showing PTS, REB, AST, STL, BLK.

[benji]

This has to be variable, but you can use the game finder with the player finder to narrow it down.

Which stats do you want to use, I can't imagine you want to use all of them including turnovers, fouls, etc.

[Nick]

He's already picking the three players through game score, this is about which stats to display to justify the picking of the top three players.

Unless I'm reading Leftos wrong.

You're right, but once those players are selected through gamescore then how do you decide which of the player's stats are most impressive relative to their own stats? That's where i suggest ranking against database. It's just a way of extracting the most impressive 3 stats of his personal line.

I.e. say a player has a line of 15 pts, 4 rebs, 1 ast, and 5 stls at half time... (first you put it through a "per minute", so you're only comparing per minute rates) you compare those numbers to the career averages of players and choose whatever ranks higher... 5 stls at half time is obviously 10 steals per 48 minutes... that would rank like #1 in history probably, so that gets selected... say 15 pts is like #10 compared to history pp48... say 4 rebs is like #4353 on the list... then 1 ast is like #9999... and viola, you have the 3 selected stats

But as you said, he could just display the 5 stats - PTS, REB, AST, STL, BLK. *shrug* Depends what he wants

[benji]

My point was that you don't need a database to do it, you just have to set a functional line. If you want it against the player's best for his career, that's different and difficult. If you want it against the "average" that's not really too difficult you just need to define your terms and measure for it.

[Leftos]

He's already picking the three players through game score, this is about which stats to display to justify the picking of the top three players.

Unless I'm reading Leftos wrong.

You're right, but once those players are selected through gamescore then how do you decide which of the player's stats are most impressive relative to their own stats? That's where i suggest ranking against database. It's just a way of extracting the most impressive 3 stats of his personal line.

I.e. say a player has a line of 15 pts, 4 rebs, 1 ast, and 5 stls at half time... (first you put it through a "per minute", so you're only comparing per minute rates) you compare those numbers to the career averages of players and choose whatever ranks higher... 5 stls at half time is obviously 10 steals per 48 minutes... that would rank like #1 in history probably, so that gets selected... say 15 pts is like #10 compared to history pp48... say 4 rebs is like #4353 on the list... then 1 ast is like #9999... and viola, you have the 3 selected stats

But as you said, he could just display the 5 stats - PTS, REB, AST, STL, BLK. *shrug* Depends what he wants

Here's another idea. What if I pick the top per-game or per-minute stats of the modern era, and use it as a normalization factor for the player's stats?

So, for example, all-time leader in the 3PT era of basketball (1979-today) in Assists per Game is Magic Johnson with 11.2 assists. Moses Malone leads in Offensive Rebounds per game with 4.8.

Let's say our player has 10 assists and 4 offensive rebounds. 10 assists normalized to the all-time best average is 10/11.2=0.89. 4 offensive rebounds normalized are 4/4.8=0.83. So obviously, his assist numbers are more impressive compared to his offensive rebounding numbers. Agreed? So I pick the top 5 normalized stats, and show those.

Obviously I won't focus on stats like turnovers or fouls, although I might want to consider showing Assists if the turnovers are low. I'd need to find the top Assist/Turnover player of the modern era, but it should work.

[Nick]

If you want it against the "average" that's not really too difficult you just need to define your terms and measure for it.

Here's another idea. What if I pick the top per-game or per-minute stats of the modern era, and use it as a normalization factor for the player's stats?

Ah yeh well that makes sense.

I would throw FGM/FGA and FTM/FTA as potential stats to display too.

[Leftos]

To improve on my normalization idea, I won't be using the top player's averages, but rather an average of the top 100 players in each category. That way, if a player was unnaturally high in a particular category, it won't make it that hard for others to reach that significant point.

Say, the top 3P% shooter had .45 but all other 99 top players were around .40. That would make an average of about .40, which is easier for a player to reach rather than that special .45.

Sound good so far? I'm also using the NBA rules for inclusion in League Leaders tables for each stat, to make sure that a player with a FG% of 1.0 on just 2 FGs doesn't make the cut.

UPDATE: Okay, so here's what I've got so far. Tell me what you think.
FG%: At least 4 FGM, normalizing factor FG% .548
3P%: At least 2 3PM, normalizing factor 3P% .398
FT%: At least 4 FTM, normalizing factor FT% .868
OREB: Normalizing factor 2.894
DREB: Normalizing factor 6.348
REB: Normalizing factor 9
AST: Normalizing factor 6.359
STL: Normalizing factor 1.673
BLK: Normalizing factor 1.731
PTS: Normalizing factor 20.104

[benji]

First glance? I'd lower the FT% and increase the FTM. Maybe ignore FT% and make it FTM/FGA.

[Leftos]

6 FTM and .83 FT%?

[benji]

To be honest, I'd strike them both and just do FTM/FGA. More important to know that a guy got to the line and hit free throws.

[NovU]

Interesting thread. Perhaps I will read into more details when I have more time.

To be honest, I'd strike them both and just do FTM/FGA. More important to know that a guy got to the line and hit free throws.

Agreed. I think Shaq's winshare was lower than David Robinson because his inability to knock down the freethrow hurt the team.

[Leftos]

To be honest, I'd strike them both and just do FTM/FGA. More important to know that a guy got to the line and hit free throws.

FTM/FGA? Now that's interesting. I'll need a baseline for that, a normalizing factor, but it shouldn't be too hard to calculate. Any ideas on how to do that? I don't think that BR supports such complex queries.

Here's a preview of what I've got so far:


EDIT: I'm over at BR using sorting by TS% which takes free throws into account as well, and here's what I've got. Tell me what you think.
FT/FGA: at least 3 FTM, normalizing factor .331

http://bkref.com/tiny/bplsC

[benji]

Only one season, and it's 2008-09, but this might help for now: http://dl.dropbox.com/u/820953/ftmfga.xlsx

[Leftos]

Yeah, pretty much the same. An average of .348, close to the .331 I calculated overall. Think I'll use the wider results and keep the .331.

[NovU]

Very interesting application you are developing. Since you don't have any access to the database, how does your application grab the data you need. Do you manually punch in the accumulated stats?

[Leftos]

The name is a bit misleading. Although it's built on the features and rules of the NBA, it works for any basketball league. Three ways you can use it, actually:
- Create a new league, add the teams and players, update via inputting accumulated stats or box scores which are added to the stats automatically.
- Import your NBA 2K12 stats.
- Download the real NBA stats, which now only works for regular season team stats; it's in my plans to have it automatically download player stats, playoff stats as well as box scores, so that you can do your own analysis on them. For the real NBA, I use Basketball-Reference.com to download the stats from.

[NovU]

Sounds like a cool project.

For the real NBA, I use Basketball-Reference.com to download the stats from.

I am not really sure what you mean by "Download" here because you actually don't get any stats in a downloadable format but the stats are only displayed in the tables in HTML.