[brettd]

I typically don't use regression, but in my world, transforms of some sort are need to get the fit closer, and residuals are the signal that tells me when and where.

What methods do you use most commonly Wrecktangle? I was about to start my first model using regression, but I'm just wondering what other stuff I should learn before giving it a go.

[Miz]

Well, this is the point of least squares in regression.

Not really, implementing a dimishing returns function is separate from that, as the function changes the data set being fit first, prior to any application of regression.

anyway,

My only point, was not to count a cover by 20 as 20 "action pts" and lump that in with 2 losses by 8 pts and think you are ahead by "4 action pts"

In my opinion, using non-linear regression needs to be founded on a reason first. I mean, it is well known that certain distributions are Poisson-like or log-normal or whatever. If you apply enough polynomials and search for a better fit, you may minimize the absolute error in the fit (not the mean squared error, necessarily) but have not added any more predictive value.

[Cheese1976]

Hey Miz, I've used your formulas to ensure a little strength in a few of my systems for this year in the NBA and just started back-testing last season...dead on target through three random months of back-testing so far

[Miz]

Cheese, I didn't invent the formulae my man. I just read through some posts in here and dug a little on the internet. Glad they are useful for you though.