Can someone explain why league averages are essential to take into account

**Jive** · 11-28-10, 02:37 AM

You have to account for the competition that those paces were set against, for one thing. But as an example, if two teams are playing each other and they both have a pace significantly higher than the league average, chances are very strong that they will play at a pace higher than the average of the two teams. This needs to be built into your calculations. Know what I mean? I'm having difficulty putting it into words. Sorry. Too dang tired.

**dmolition** · 11-28-10, 02:53 AM

Im no expert and i'm still tinkering with formulas, but as i see it, comparing against the league average gives u an understanding of how much better or worse each team is to the rest of the league on any given stat, also most teams in many stats tend to reverse to the mean, so applying the league average to some stats helps you account for times where teams where over-perfoming or under-perfoming.

Im not sure if i explained myself well enough or if this is all there is to it, but hope it can help fwiw.

**MadTiger** · 11-28-10, 07:16 AM

I ran a 15.

Is that great, horrible, or just average?

You won't know unless you know the league average.

**pedro803** · 11-28-10, 09:59 AM

I can't answer your question but I will give my thougts -- it is important to realize (as I am sure most of us do) that league average (LA) and strength of schedule (SOS) are very different

first off SOS will be different for the two teams so it can straight up and objectively and almost on its own indicate an edge for one team or the other

LA on the other hand is the same for both teams by its very definition, so it can't indicate an edge for either team -- so maybe one's first impression is that it can't be useful, ........that is -- how can it be useful if it is the same for both sides (talking about sides here not totals).

But this is where the part about multiple regression being part art and part science comes in, I can remember having a discussion with one of my professors about a model of hers where she had included a factor similar to this, and she couldn't give an exact answer as to how it improves the model other than to say that it seems to somehow steady the baseline -- and like a cook perfecting a recipe we should just experiment with some of these things and test the predictions against real results and hone the model that way.

**roasthawg** · 11-28-10, 12:12 PM

I don't use league averages.

**dbDan** · 11-28-10, 01:10 PM

Isn't a league average kinda like a regression of a single set of numbers (1 factor)?

**benjy** · 11-28-10, 02:01 PM

I'm not familiar with OPace and DPace (just Pace) but I think the formula you are using would break down when looking at two teams that are both on one end of the pace range. For example if you were looking at the expected pace of a game between the two fastest teams is the league you would simply get a result that is approximately equal to either of them regardless of opponent as opposed to an abnormally high scoring result, which one would intuitively expect. By using league average and doing some more math your results can account for this and be more representative.

**pedro803** · 11-28-10, 02:03 PM

yes I agree that it may not or maybe even probably won't increase the power of your model, but my point is you can try it both ways and then backtest

**pedro803** · 11-28-10, 02:05 PM

Originally posted by benjy

I'm not familiar with OPace and DPace (just Pace) but I think the formula you are using would break down when looking at two teams that are both on one end of the pace range. For example if you were looking at the expected pace of a game between the two fastest teams is the league you would simply get a result that is approximately equal to either of them regardless of opponent as opposed to an abnormally high scoring result, which one would intuitively expect. By using league average and doing some more math your results can account for this and be more representative.

yes, that is what I meant when I said sometimes a factor like that can sort of anchor the baseline