10-25-2011 08:42 PM
I am trying to run a regression between a dependent variableand several other independent variables.
First I have used STEPWISE, which selected two variables in the regression model. But when I verified the regression’s assumptions, I realized that the residual were normally distributed but there was unequal variance.
I have tried to transform the data but I could not use one single transformation for all the variables and the dependent variable has many values of zero and I could not find any transformation for that.
Later I have used LOESS procedure for nonparametric regression, where I have built several models based only on 3 independent variables that mostly correlated with my dependent variable. The model with the smallest AICC was still the one selected by STEPWISE procedure but now I found out that this method also requires normality of the variance. Another problem is that this method only tells me what are the most important variables but doesn’tgive me an equation.
Now I am trying Weighted Least Square method but I don’tfind an example on how to run it with more than one variable. Also, I am not sure this method would be better or if it will work.
Can anybody help me with a piece of advice on what method to try in this situation? And if the Weighted Least Square method is good, is there a code on how to run it for multiple variables?
I will highly appreciate any kind of help!
04-23-2013 07:17 PM
I am wondering if you ever found an answer to this problem. Dealing with the same situation myself.