# Linear Regression with effect size changes of categorical variables with continuous

#### farre1pd

##### New Member
I have three categorical variables (disked, herbicide, Lands) each is binomial (0,1). For example, within disked there is a group in which it does not occur (0) and where it did occur (1). I am performing a forward stepwise-approach in which I evaluate each categorical variables in the order presented and add statistically significant variables to the overall model. I have found all three variables have a statistically significant "effect size" or group value compared to when actions are not present (0 instead of 1). When I add a continuous variable in the next part of my step-wise process, values for these groups change and some are not shown to be important categorical variables any longer. The continuous variable added also shows a statistically significant effect.

What is occurring between the continuous and categorical variables when this is occuring?

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Re: Linear Regression with effect size changes of categorical variables with continuo

Most people frown on any stepwise variable selections in model building. Unless this is an assignment where you are required to use the process, I would recommend building models based on your familiarity with the subject content. It will take longer, most likely, but it will take away the automated version, where the software is not privy to all of the information you are.

Most likely the continuous variable is just a better predictor of your outcome. Does it as a predictor make sense to you? Many times you will have what you may think is the final model, then incorporate another variable that shakes the thing up. This is not uncommon. Perhaps plot your models and watch its fit on your data or the amount of variance you are explaining.

#### farre1pd

##### New Member
Re: Linear Regression with effect size changes of categorical variables with continuo

I agree with you about the model building; having a predefined set of hypothesis driven models and comparing, this project just called for a little different approach. The prediction does make sense to me overall but I fear the low model r-squared (0.07) and large residuals of continuous data values is what is really causing this continuous variable to be so important.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Re: Linear Regression with effect size changes of categorical variables with continuo

Also, you are just writing about significance. It would also be interesting to seen the actual effects on the model's slope or how big the beta coefficients actually are. Do you have a big sample size, if so these may be significant predictors, but not contextually significant.