Statistical Analysis For Future Predictions

[urinesane]

Since the trade thread was closed I thought we'd move the discussion here about data sample sizes and their validity.

I wrote a paper on this in college, close to 25 years ago, geeze, time flys...my modles which I use in DFS are based on the previous 4 games and calculate several splits ect...and are accurate enough to profit. 4 games is all you need. 27, not 21 games is more than enough. ChatGPT isn't some kind of savior, it is language model that predicts words or actions based on the scraping of the internet, it doesn't make them right because it presumably scrapes inaccurate information which may be in abundance. Just saying... But I am correct in this.

What was the class, also just curious, what grade did you get on that paper?

Also, I didn't realize we were talking about DFS, I could have sworn that we were talking about the future implications of making major moves based on a small sample size of games.

You are correct that ChatGPT is not completely accurate or without faults, but that doesn't mean it's always wrong either. When you say a comment like ""Many predictive models use as few as four games to extrapolate data across 82 games" and then have nothing to back it up (other than your personal DFS model that apparently makes a profit) I tend to side with ChatGPT.

I'll gladly eat crow if you provide a handful of predictive models from professionals (not just some guy on a forum that gambles) with proven results. Otherwise you're just a gambler with a system (and an inability to change your opinion when faced with logical counter arguments).

Typically you want around 10%,... about 8 games, but there are tricks you can play with splits which reduces accuracy over 82 games but the drop off isn't proportional to the reduction in sample size.

Ok, please use your model to make predictions for two sample players. We will call them Player A and Player B. Keep in mind that they are the same exact age, play the same position, AND these games happened over the same period in the same season.

Player A:
Game 1: 25pts, 5 reb, 7 ast, 2 STL, 10/19 FG 5/7 FT (in a Win)
Game 2: 28pts, 2 reb, 8 ast, 2 STL, 10/17 FG 7/9 FT (in a Win)
Game 3: 23pts, 4 reb, 10 ast, 1 STL, 9/14 FG 5/6 FT (in a Win)
Game 4: 38 pts, 4 reb, 7 ast, 2 STL, 13/23 FG 10/13 FT (in a Win)
Game 5: 20 pts, 6 reb, 8 ast, 3 STL, 13/23 FG 4/7 FT (in a Win)
Game 6: 27 pts, 2 reb, 11 ast, 2 STL, 9/20 FG 7/11 FT (in a Win)
Game 7: 27 pts, 2 reb, 11 ast, 1 STL, 9/20 FG 7/11 FT (in a Win)
Game 8: 10 pts, 5 reb, 13 ast, 0 STL, 4/6 FG 2/3 FT (in a Win)
Game 9: 26 pts, 2 reb, 5 ast, 4 STL, 8/18 FG 8/10 FT (in a Loss)
Game 10: 28 pts, 4 reb, 14 ast, 5 STL, 11/20 FG 3/6 FT (in a Win)

Player B:
Game 1: 11pts, 2 reb, 9 ast, 4 STL, 4/14 FG 0/0 FT (in a Loss)
Game 2: 16pts, 7 reb, 10 ast, 2 STL, 7/9 FG 1/2 FT (in a Loss)
Game 3: 36pts, 7 reb, 7 ast, 0 STL, 13/17 FG 4/7 FT (in a Win)
Game 4: 14 pts, 0 reb, 5 ast, 2 STL, 5/14 FG 1/1 FT (in a Win)
Game 5: 9 pts, 7 reb, 5 ast, 1 STL, 3/10 FG 2/2 FT (in a Win)
Game 6: 8 pts, 3 reb, 8 ast, 2 STL, 3/10 FG 7/11 FT (in a Loss)
Game 7: 11 pts, 2 reb, 4 ast, 0 STL, 3/5 FG 0/0 FT (in a Loss)
Game 8: 36 pts, 2 reb, 6 ast, 2 STL, 13/21 FG 4/4 FT (in a Loss)
Game 9: 11 pts, 3 reb, 6 ast, 0 STL, 3/6 FG 2/2 FT (in a Win)
Game 10: 9 pts, 0 reb, 2 ast, 2 STL, 3/7 FG 2/2 FT (in a Win)


Now I know that you only require 4 games and 8.2 games would be plenty, but I was generous and gave you 10 games (2% more than you said you typically need to profit in DFS).

Using your predictive model could you give us an idea of what front offices should do with these two players longterm?

[Note30]

So the mod went and locked the discussion and you went and reopened it under a different name? What's the point of moderation then.

[minimus]

Lock this

[urinesane]

So the mod went and locked the discussion and you went and reopened it under a different name? What's the point of moderation then.

I didn't re-open a trade thread. The thread was closed because it turned into a feelings fest and wasn't about trades/basketball anymore. This thread is a topic talking about statistical analysis used for predictions in sports, which is relevant to this team, basketball, and this forum.

You can always choose to read other threads if you aren't interested in the topic, but it's not violating any rules that I am aware of (nor is it using personal attacks to make a point).

[thinktank]

A sample size needs to be random to be valid (and then it needs to be large enough). End of discussion.

[PeeDee]

The sample needs to be random, not the sample size. The sample size needs to be big enough to be considered statistically significant in order for it to be representative of the whole. Professional analyst out.

[urinesane]

The sample needs to be random, not the sample size. The sample size needs to be big enough to be considered statistically significant in order for it to be representative of the whole. Professional analyst out.

The main reason I kept the same age/position/time frame is because I didn't want it to seem cherry picked from different players at different stages in their careers or different eras of the NBA.

Also, please keep in mind that this is to see if SO_MONEY's statistical analysis that apparently can be based on 4 - 8.2 games can carry any water (spoiler: it can't). The main point being that there is no way to predict trends based on such a small sample side or use it as a good reason to make longterm decisions.

I mean, it's obviously a trap, but I didn't want to totally cherry pick different players from different times/ages/positions etc.

Also, in this case it would be easier to put a person in a GM's seat and ask them which player they would want to move forward with longterm (if at all), based on stats from a 10 game sample in the same season with two players of the same age (which would be ridiculous to do).