topic Re: model building in Statistical Procedures

model building

kurofufu — Sun, 05 Aug 2012 01:19:27 GMT

I have a list of variables and built a regression model on them. The R-square is 0.1. Are there any ways to improve the predictive power of the model without finding more variables?

Re: model building

PGStats — Sun, 05 Aug 2012 01:43:54 GMT

At r-square = 0.1, your model has essentially no predictive power. If you give some description of the variables and of the way the observations were acquired (sampling/experimental plan) someone might be able to suggest a better statistical model or at least, alternative data exploration methods and modeling approaches.

Re: model building

kurofufu — Sun, 05 Aug 2012 02:27:09 GMT

"At r-square = 0.1, your model has essentially no predictive power."

Are you sure? The CMS-HCC model used by CMS to pay private insurance companies has R-square of 0.09. The model my company is using has R-square of 0.1, but it already helped the company save a lot of money.

Variable transformation is one way which possible help improve the model's predictive power. I would like to know what other options I can have without acquiring more variables.

Re: model building

PGStats — Sun, 05 Aug 2012 02:50:51 GMT

It is true that strength of evidence requirements vary a lot across disciplines. - To go beyond regression, you might want to consider the model building techniques provided by data mining tools - PG