BuzzerBeater Forums

So far, with 20 data points, I have a model that fits very well. An r-squared of almost .99. It says the following very weird results:

-The BB defined position does not seem to matter for capping purposes.
-The model works much better without an intercept (.99 vs .46 r-squared).
-Jump shot and rebounding matter a lot for skill capping. The rest of the skills have some contribution, but are pretty minor in comparison. Some skills are even so minor they are hardly worth mentioning.

Now, I posted this information for fun, just to show what we could be learning if I had a ton of data. I don't really believe that if you trained everything but jump shot and rebounding that your player would never cap. R-squared, while it gives nice fuzzy feelings, means almost nothing. And given the number of variables that I threw into the model, it is pretty close to being underspecified (in other words, complete junk).

I still need a lot more data.

To give more information on the model: I am starting with the assumption that the BBs used a similar formula for skill capping as they did for salary. It makes sense to me and that's probably why we see a good correlation between salary and skill capping. With more data I can start to get a feeling whether my assumptions are good ones or bad ones and adjust accordingly.

Great work so far. I'm very interested in seeing the results. I would love to help out but unfortunately I have never trained a player long enough to hit the skill cap :(

Yeah, thanks for taking initiative. I think this can be quite helpful.

Well, I can tell you that I was pretty convinced that the cap is linked to salary too. The two guys I have in the all of fame were trained a lot in js, they stopped to pop when they got lvl 17-18 in js with a salary around 65-70k/week, their potential was all-star*
Now I think that the cap is still related to the salary but in a different way; if you train a lot one single skill you'll get the cap sooner and training one single skill a lot means also to have an higher salary that's why it is still related to the salary. If you train a good SF hardly you will get the potential cap, that's my opinion, I wait to see the results of the study

Well, I can tell you that I was pretty convinced that the cap is linked to salary too.

My theory is that it is not linked to salary. Only that the cap correlates well to salary under most conditions. I could give an example of two different allstar potential players: one who capped at 40K and one who capped at 90K. That seems to be a fair distance off.

Now I think that the cap is still related to the salary but in a different way; if you train a lot one single skill you'll get the cap sooner and training one single skill a lot means also to have an higher salary that's why it is still related to the salary.

The part about multi-skilling is indeed true. Maybe your theory is right, but I am going to start by exploring my 1st theory and see where it leads me. I will make sure to get player salaries from now on - in case I need to test a different path.

There is a lot of buzz that salary is not a cap, but combined skill point is, so if you could, test both theories to see which is more accurate. I don't know which one your doing now as I didn't read the first post entirely but if your only doing one as of now, try to analyze both.

Last edited by oriolekid at 08/27/2010 21:48:32

There is a lot of buzz that salary is not a cap, but combined skill point is, so if you could, test both theories to see which is more accurate.

I will keep all possibilities in mind, but as I said I will test my theory first. I don't want to elaborate on the theory too much at the moment... Although like I said salary is not a part of it. Combined skill point is starting to get closer, but it is still not the complete story.

Anyhow, more to come once I have the data. And of course if my theory is wrong I will start to test others. ;-)

Again, a just for fun update. I now have 40 data points. Here is what I see now:

-Position has no impact on capping
-The model is exponential, similar to the salary formulas, but the coefficients are quite different. This means that the higher the skill, the faster you approach the cap per skill-up.
-Jump shot, jump range, inside defense, rebounding and passing all contribute significantly to capping (and at similar levels)
-Outside defense and inside shot contribute in a very minor way
-All other skills have zero impact on capping.

The number of data points was pushed up considerably when jbmcrock pointed me to a federation post where people were talking about their players who had capped. So thanks to him for that.

The data is still skewed in my opinion. I mean, the model fits really well, still with a .99 r-squared. But it only does an ok job of estimating allstar players. And even then, sometimes there are weird results. Of course, this is almost certainly because most of the data points come from allstar potential players, but that's another story. I also have too many PFs and Cs in the model and not enough PGs, SG, SFs. Also, weird players who capped with a lot of outside d or inside shot would really help.

But... as always, I will take the data from any player if he is capped. I will also continue the updates if the model changes or I get a significant jump in the number of data points.

Outside defense and inside shot contribute in a very minor way

That is both very interesting and very bizarre.

That is some very interesting and powerful information if what you say holds true for a larger sample. Especially this for my training:

Outside defense and inside shot contribute in a very minor way
-All other skills have zero impact on capping.

Excellent work as usual PaperPusher.

Last edited by idorux at 08/30/2010 18:52:30

good job keep it up

Last edited by LA-Nir at 08/30/2010 19:01:24