What makes a stat impressive?

[Leftos]

Yeah, I'm actually parsing the HTML files, grabbing the data from the tables manually. It's not much work, but I'm sure I could find a .NET library that already does that, which will make my life much easier when I get around to implementing "downloading" player stats and box scores.

[NovU]

That is nuts. Making your program dissect html source code and making it grab the data you want. C# in .net programming I guess.

[Leftos]

Pretty much.

[SkyLine03LT]

For one thing I think you need to increase the made shots required for each category, 4/4 free throw shooting isn't impressive in my mind, unless it's Shaq, of course.

Another thing I'd correct is rebounding statistics displayed- as you can see on your preview there, you wasted at least 1 stat on rebounds by displaying offensive, defensive and total, when you would have been better off just displaying offensive and defensive rebounds (most people these days can add 8 and 4, I think) and had another stat there, like Points scored, for example (actually I think points scored should always be included in the stat line).

The third thing is creating different guidelines for each position- you have set your rebounding factor at 9, but if a point guard gets as much as 5 rebounds that counts as a good number (although Rajon Rondo would debate me on this one ).

By the way, I like how you made the program to display both made/attempted shots AND shooting percentage. But I would change the minus (-) sign with a slash (/), just so it looks better and that's how those are displayed on television and everywhere else.

[Leftos]

Thanks for the suggestions. I might go back to BR and get some better normalizing factors per position.

About rebounding, I was thinking of removing the Defensive Rebounds, instead of total. Most outlets advertise total and offensive, I don't think many focus on defensive, which are a given.

What would you say should be the lower limit on FGM, 3PM and FTM for inclusion?

[Nick]

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

What happens if no players reach these numbers?

[Patr1ck]

Is the normalizing factor your scaling system so a player with 5 minutes ,10 points, 3 assists, and 2 steals doesn't end up the player of the game after putting up an amazing game score?

[NovU]

He's planning to use the Game Score metric to choose the best players for each game. I don't think 5 minutes player would become the player of the game that often.

I am also interested to know what metric will be used for the multiple games.

[Patr1ck]

He's planning to use the Game Score metric to choose the best players for each game. I don't think 5 minutes player would become the player of the game that often.

Ok, I was thinking more of along the lines of PER, as I have never seen the equation for game score until now.

What happens if no players reach these numbers?

I don't think the normalizing factor is a baseline minimum. Or maybe the normalizing factor is for choosing the different statistics to display.
So with -
STL: Normalizing factor 1.673
BLK: Normalizing factor 1.731
and the top performer of the game has 2 blocks and 2 steals but the program can only choose one of those categories, I guess the steals would show up because the normalizing factor is lower.

[Leftos]

He's planning to use the Game Score metric to choose the best players for each game. I don't think 5 minutes player would become the player of the game that often.

Ok, I was thinking more of along the lines of PER, as I have never seen the equation for game score until now.

PER is a very complex function, that requires far too many parameters that aren't available in a box score, and takes into account the league pace, team pace, which are things that are season-long averages.

I based my analysis on BasketballReference's box score page (here's a sample), which as you can see, has no efficiency metrics. A good metric, however, that can work for just a game is the Game Score metric.

Game Score; the formula is PTS + 0.4 * FG - 0.7 * FGA - 0.4*(FTA - FT) + 0.7 * ORB + 0.3 * DRB + STL + 0.7 * AST + 0.7 * BLK - 0.4 * PF - TOV. Game Score was created by John Hollinger to give a rough measure of a player's productivity for a single game. The scale is similar to that of points scored, i.e., 40 is an outstanding performance, 10 is an average performance, etc.

If Hollinger says it's alright, I trust him. Sure, it doesn't take into account minutes, but if a player has enough of a performance to make it to the top of the Game Score metric, he's done well enough to deserve the Player of the Game, statistically. Then again, I don't think Player of the Game is a purely statistical result. There are things that don't show up in the box score, and maybe PotG is a subjective pick of the player that did the most during the game for his team to win.

What happens if no players reach these numbers?

I don't think the normalizing factor is a baseline minimum. Or maybe the normalizing factor is for choosing the different statistics to display.
So with -
STL: Normalizing factor 1.673
BLK: Normalizing factor 1.731
and the top performer of the game has 2 blocks and 2 steals but the program can only choose one of those categories, I guess the steals would show up because the normalizing factor is lower.

As Pdub guessed and as I mentioned earlier, the normalizing factor is not a minimum threshold.

The normalizing factor is calculated from the average of the top 100 players in each per-game average stat since 1979. Each player's stat is divided by that factor, to find out which stats are more significant compared to the average of the league's modern era.

Example: let's say a player has 15 points, 4 assists, 8 rebounds and 2 steals. But we only want to show the two most significant stats.

PTS normalized: 15 / 20.104 = 0.746
AST: 4 / 6.359 = 0.63
REB: 8 / 9 = 0.888
STL: 2 / 1.673: 1.195

So, out of his whole stat-line, the most impressive stats are his 8 rebounds and 2 steals. Granted, the current procedure doesn't take position into account, but I'm going to fix that and get new normalizing factor per position. 6 assists from a center are definitely more significant than the same amount from a point guard, as Skyline said.

Is the normalizing factor your scaling system so a player with 5 minutes ,10 points, 3 assists, and 2 steals doesn't end up the player of the game after putting up an amazing game score?

The player of the game as well as the other 3 top players of each team are decided by game score. The normalizing factors only decide which stats to show of each player. They're not taken into account as to pick the best player. Even if a player has 15 assists and the highest normalized stat overall of the game, if he made 20 turnovers and didn't get a single field goal in, he'll have a really low Game Score. That's why I judge them as players by game score, and then I judge each player's stats in-between themselves by normalizing them.

I am also interested to know what metric will be used for the multiple games.

I was thinking of allowing the user to pick it themselves. It will be PER by default, but mind you, PER needs the whole league's stats, so if the user doesn't keep track of all teams and players correctly and fully, PER will surely be off. Also, PER isn't calculated when the user specifies a timeframe other than season (e.g. 5 days, last month, last 2 years), as the program accumulates those time-framed stats by adding box scores, and calculating PER would mean that the program would have to add the whole league's box scores. The users will have to rely on less thorough metrics, such as Game Score, if they want to get the best performers of a specific timeframe. They'll still get PER however, if they want the season's best performers, which only uses already accumulated stats rather than adding them up from box scores at that moment.


UPDATE: Okay, so I went back to BR to get per-position normalizing factors. Again, 100 top players per position per stat per game, accumulated over their careers.

	if made at least	Guards	Forwards	Centers
FG%	4	0.487	0.528	0.529
3P%	2	0.393	0.380	0.230
FT%	4	0.862	0.826	0.753
OREB		1.242	2.671	2.328
REB		4.153	8.145	7.431
AST		6.324	3.037	1.688
STL		1.619	1.209	0.680
BLK		0.424	1.240	1.536
PTS		17.160	17.731	11.616

UPDATE 2: Yeah, seems much more focused now.



Kobe's rebounds have surpassed the measly 3-10 3pt shooting, and Bynum now has space for his points as well now that defensive rebounds are gone. Plus, those points are really good for a center.

[NovU]

But I think PER is a flawed metric for this particular use as it doesn't take pace into an account. And people often mis-interpret meanings of PER. So wouldn't WinShare be more fitting in this case?(ask benji on this) Giving users a choice is a good idea though. Or creating your own point system giving best players 3 stars or 2 stars each night and accumulating the stars to decide who has been the best in multiple games would be someting unseen.

My recommendation would be something like WAR. It's a new metric that sounds very interesting and probably B-R will never implement this to their site. But it gives a whole new perspective to the measurement and your program will offer something new to the users, something outside of things that can be found from B-R, 82games, or any other sites. Unique that perhaps even experts will find your program useful.

I was thinking of making something similar to your program as a hobby project but I thought access to the database posed the biggest problem. Great job on your part going distance to get the datas from B-R. It was not an easy task to do so I believe. Anyways, I think more unique you make your program, more demand there will be. I would try to include all kinds of metric if it was my program and make it unique. It was my idea initially when I thought of making this sorta application.

[Leftos]

From the Calculating PER page over at BR, it's pretty obvious that it takes both league and team pace into account. Are you sure it doesn't?

WAR does sound interesting, but it does need a whole lot of calculations. I may get into it later, once I finish the "Get Real Stats" feature.

[NovU]

Pace as in significance/insignificance of the plays or defensive role or plus minus ratings. I don't think it's a bad metric but there are some flaws there as opposed to John Hollinger's claim.

[benji]

Pace as in significance/insignificance of the plays or defensive role or plus minus ratings.

That's not pace. Why would you use an established word "pace" to mean whatever gibberish you said.

I don't think it's a bad metric but there are some flaws there as opposed to John Hollinger's claim.

Except Hollinger has never said it's flawless and acknowledges its flaws quite openly.

My recommendation would be something like WAR. It's a new metric

lol no it's not.

Great job on your part going distance to get the datas from B-R. It was not an easy task to do so I believe.

Can you copy-paste? Because that's all you need to get data off B-R.

Parsing is faster but if you don't know how to do it, copy-paste is faster. 82games.com is the hard one to get data off of for me because of that.

I based my analysis on BasketballReference's box score page (here's a sample), which as you can see, has no efficiency metrics.

TS%, ORtg, etc. are all efficiency based. What you're talking about is production. PER and Game Score are more measures of production than efficiency.

so if the user doesn't keep track of all teams and players correctly and fully, PER will surely be off.

You don't need the players, you only need the teams.

[Leftos]

Well, someone's in a bad mood.

Downloading the stats off of B-R to get it into the tool isn't a simple as copy-paste. It does require parsing and handling of some sort.

As for mixing up efficiency with production, you're right. I got the terms confused in my head because I was used to the "Efficiency" word of the PER acronym, although I know the difference of the words.

And about needing only the teams for the PER calculation, how about this part of the Calculating PER guide?

The final step is to standardize aPER. First, calculate league average aPER (lg_aPER) using player minutes played as the weights.

Is there an easier way to do this?

EDIT: BTW, Benji, which Metric would you suggest to determine the best performers of a team for a multitude of games?

[benji]

I'm not in a bad mood, NovU's just a moron who needs to be insulted regularly.

Well, I don't know what database you're using, but copy-pasting into Excel is simple enough.

And about needing only the teams for the PER calculation, how about this part of the Calculating PER guide?
Is there an easier way to do this?

You just need the league average PER. Which you can get using just the league team average.

[Leftos]

I'm using SQLite databases as the backend for the tool.

Yeah, but doesn't getting the team average PER need each player's PER? Sorry if the question sounds stupid.

[benji]

No, just use the league team average. The thing labeled "league average" at the bottom of the team statistics table: http://www.basketball-reference.com/lea ... _2012.html (Or make your own by summing the teams.)

[NovU]

Great job on your part going distance to get the datas from B-R. It was not an easy task to do so I believe.

Can you copy-paste? Because that's all you need to get data off B-R.

Parsing is faster but if you don't know how to do it, copy-paste is faster. 82games.com is the hard one to get data off of for me because of that.

Copy-paste and then grabbing the data would be harder since it's 2 steps, from application developiing perspective. Unless the application doesn't need to automatically retrieve data from the sites.