Transforming dummy variables to get rid of model bias in OLS

Discussion created by angelag69 on Dec 23, 2013
Hi there forum! I have a question about the use of dummy variables in OLS and GWR. My model(s) pass all the tests for a properly specified model, EXCEPT the Jarque-Bera test, which indicates model bias. I am using several dummy variables, which clearly are NOT normally distributed. Could this be the problem? If so, would log or exponential (or other) transformation rectify this problem?

I am very certain that there is non-stationarity in the relationships between my dependent and explanatory variables. Is there any chance that this could be causing the model bias, or is it likely to be non-linear relationships, meaning I need to transform my explanatory variables?

Thanks in anticipation!