You would use a basic multiple regression analysis. In fact, your research question seems perfectly suited for multiple regression. Cheers

Cory Cobb Do you mean a linear multiple regression analysis?. Can i apply a regression when the data(DV or IV) are not normally distributed?

Hi Mosab,

You could transform your variables to be normally distributed, then use linear regression. Assuming they can be transformed to within acceptable parameters for normality, this should be fine.

Hi Mosab,

Yes, I was referring to multiple linear regression. Regarding your question about skewed variables, there are several things you should consider. First, let me modify David's answer as it is not quite accurate. It is not the observed variables that must be normal, it is the residuals that need to be normally distributed. Indeed, you can have non-normal variables and still have residuals that are normal. The best way to check this, assuming you are using an accessible software such as SPSS, would be to run the model with all predictors in it and check the residual plot to examine normality. To do this, before running the actual model, go to the Plots option in SPSS and place the ZPPRED in the section labeled X, and place ZRESID in the section labeled Y. The output will show you how residuals are distributed. If thy are non-normal, you may consider some transformations. However, if they are extremely skewed with lots of zeros clustered at the upper or lower limit, you might want to consider other forms of regression such as Tobit (if the DV is continuous and censored) or zero-inflated models. Hope this helps.

Cory

Your DV should be continuous (does not matter if it is normally distributed or not) for linear regression analysis. To determine how much each of your IV contributes to the variance in your DV you look at the part correlation and square it, then you get the % it contributes to the predication or explanation to your pain variable.

Thank you for your help , your answers sorted out my doubt.

Cory Cobb David Hallford S. Béatrice Marianne Ewalds-Kvist

Just for clarity, I have to point out that the answer provided by Béatrice Marianne Ewalds-Kvist is incorrect. If the regression residuals from your outcome are not approximately normally distributed, then OLS linear regression will provide biased parameter estimates.

Cory

Multiple regression analysis is cool for your said research.

Check for multivariate normal distribution of your data. If not normal go for logistic regression. Binary logistic regression application would imply that your dependant variable is dictomous.