stepwise regression - Statistica General Discussion - Statistica - Dell Community

stepwise regression

stepwise regression

This question is answered

Hello 

I would like to do using statistica a stepwise regression.

I have a small sample (n=10) and the data are not normally distributed.

Does exist in Statistica a method (procedure) to do 1) a stepwise regression for non normally distributed data and small sample 2) a multiple regression to find the best predictor for non Gaussian data ?

Many thanks for your response.

Best

GdM

Verified Answer
  • Hello,

    Statistica does not have a have a stepwise regression module for non-normally distributed data. When dealing with a sample size that small, stepwise regression should probably be avoided (for most types of data). Since automated regression model selection methods choose variables based on importance and contribution, a sample this small will heighten your chances of choosing an "important" variable based on data that might not properly represent the population. A rule of thumb for stepwise regression is to never use less than 10 observations for every 1 variable you are beginning your model with (http ://people.duke.edu/~rnau/regstep.htm). 

    One possible solution for your second question (best predictor) would be using the Feature Selection module found in Statistica. Regardless of the method chosen, you might still run into issues choosing the best predictor if your sample size is limited to 10.

    Best,

    Collin

  • Hello,

    I will include an example of stepwise regression in Statistica using data with a normal distribution. Going through this example should provide a good foundation for using stepwise regression with other data.

    Stepwise Regression Example

    Best,

    Collin

All Replies
  • Hello,

    Statistica does not have a have a stepwise regression module for non-normally distributed data. When dealing with a sample size that small, stepwise regression should probably be avoided (for most types of data). Since automated regression model selection methods choose variables based on importance and contribution, a sample this small will heighten your chances of choosing an "important" variable based on data that might not properly represent the population. A rule of thumb for stepwise regression is to never use less than 10 observations for every 1 variable you are beginning your model with (http ://people.duke.edu/~rnau/regstep.htm). 

    One possible solution for your second question (best predictor) would be using the Feature Selection module found in Statistica. Regardless of the method chosen, you might still run into issues choosing the best predictor if your sample size is limited to 10.

    Best,

    Collin

  • Hello,

    I will include an example of stepwise regression in Statistica using data with a normal distribution. Going through this example should provide a good foundation for using stepwise regression with other data.

    Stepwise Regression Example

    Best,

    Collin