BuzzerBeater Forums

However, consider what the price of such a trainee would be after a suitable number of games where the trainee has demonstrated they are not a foul menace. I submit that the price would be higher - that is, buying a trainee with no game history comes with a discount that reflects that risk.

I do think a less aggressive player should be sold at a premium. But I also think that the seasonality effect more than offsets this premium. After 2-3 weeks, most rosters are full and basic economics tell you that when there's less demand, the price of a good falls. All in all, the value of any rookie, aggressive or not, is lower 3 weeks into the season than at the beginning of it.

It makes no sense to have a hidden skill that can make a player useless, and discussing how you can prevent yourself from ending up with one of these duds misses the point. The BBs have indicated the intent of the hidden aggressive skill is to have pro's and con's, and I think a more useful direction of the discussion is whether it is balanced (which it's obviously not), and how it can be better balanced. We'll see if the current changes are sufficient.

Yeah that was exactly my point.

My spanish is not great but still good enough to understand. And again, I don't agree with you, at all. Whatever statistical model or framework you use, it is impossible to draw conclusions from a single observation analysis. It's just statistics 101.

If you had understood as you say, you would say to which concrete point in my model you disagree. But you don't so I know you are lying either to yourself or to me.

I don't know what statistics 101 says. Statistics 301 says that you have to make the most out of the information you have, no matter how scarce it is.

If you had understood as you say, you would say to which concrete point in my model you disagree. But you don't so I know you are lying either to yourself or to me.

I told you what's wrong with your model quite a few times already. But I'm going to try one last time: the number of observations you use is way too low. I don't have any issues with your statistical development, using both a Poisson distribution and then Bayes' Theorem, but the lambda you use is based on one single observation, which makes the whole model totally irrelevant. As I explained a few times already, your lambda should be computed on at least 5 observations, which is still a pretty low number to be honest.

Compute your lambda with 1 observation = the model is statistically irrelevant, so you're going to make decisions based on irrelevant data. That's pretty bold to claim that other managers are stupid after considering this.

Compute your lambda with 5 observations = the model is more relevant but you're already 2-3 weeks into the season and your rookie has lost some of his value.

This shows why you were wrong in your first message.

Besides that, considering that 25% of the players are aggressive is a pretty bold assumption too. Have you processed the data in order to come up with such a number? Or is it just a guess? Because that number has a lot of impact on your conclusions. Garbage in, garbage out.

I don't know what statistics 101 says.

Yeah I can see that. That's an issue.

Statistics 301 says that you have to make the most out of the information you have, no matter how scarce it is.

1 observation is not "scarce information", it's basically no information at all. Find me one Statistics lecturer who says that you can conduce a decent statistical analysis based on one observation. I'll wait.


Last edited by Veoz at 5/7/2022 9:10:50 AM

Let's take an extreme example.
There are a thousand white or black balls in a closed box.
I have to estimate how many balls are white.
I can make an additional assumption: all balls are white or all balls are black
I take one ball and it happens to be white. I put the ball back into the box.
(of course I cannot see the colour of the balls inside the box)
I can be 100 sure that the number or white balls is exactly 1000.
If you don't understand the example or you can't see the relation to my Spanish post, please retake Statistics 101.

Find me one Statistics lecturer who says that you can conduce a decent statistical analysis based on one observation.

Any first lecture on estimation will include some examples with a sample of only one element, or even zero elements.
Do you know that you can estimate a value with no sample?