Handling Non-normal Distributions in Explanatory Variables

[returningstudent]

Hello,

I would really appreciate some advice on using EViews for my regression on elevation and other geographic data. Upon looking at the histograms of each of my variables, I noticed they are all very skewed. To be more specific, consider the following variable: area at elevation of height <500m as a portion of total area. Inevitably some observations are never <500m, therefore many such regions have a value of 0 for that corresponding variable. As a result, the value of 0 proceeds to inflate the distributions of nearly every such variable, so many explanatory variables are non-normal.

Let me briefly add, I am not try to model anything, I'm merely interested in the coefficients of each variable. I also want to be assured that the coefficients are statistically different from zero given many non-normally distributed variables. I know the Poisson model deals with non-normal distributions, but there are non-integer values in my dependent variable. Would it be better to sacrifice some accuracy and round the real values into integers? My concern is that given the skewedness of my independent variables I may have violated the normality assumption for OLS efficiency and negative predicted values may be present. Or by any chance is there another feature of EViews that can help me estimate under these circumstances that I'm not aware of?

For your reference here is some more information on my estimate

- 8 explanatory variables
- 54 observations
- All values are in percentages; current functional form is level-level, but I'm open to transforming variables

:eviews6: Thank you :eviews6:

[startz]

The distribution of the explanatory variables is not relevant for the distribution of the coefficients.

Consider the case of a dummy variable on the right. It's about as non-normal as you can get and the use of dummy variables is absolutely standard.