My dataset is not very big (between 100-200). the residuals are not normally distributed. so:
1. Is there any other statistical method similar to multiple linear regression but suitable for this case?
2. If not, what can be the solution?
Thank you
I have also encountered such a problem very recently. The data violated the assumption in particular, normality. I transformed it into a logarithm and expo, but still no change. I am thinking of going for poison regression, as the data nature is count.
The answer depends on the residual plot. See KUTNER ET AL APPLIED LINEAR STATISTICAL MODELS for full information. Available in the z-library. Best wishes David Booth
It really depends on the data you are analyzing. A few alternatives that come to mind: 1) transform a variable, 2) add additional variables, 3) add a quadratic term, 4) use a generalized linear model, 5) use a robust regression technique.
Look at the severity of the violation. It may be just negligible given the large sample size.
@salvatore.where.did you get recommendation 4. If he needs a generalized linear model he should be using it a long time before this step. We need to always teach that the type of regression always depends on properties of the DV. The Biostatistics folks stress this more than any of us. See Rosener Fundamentals of Biostatistics. If we leave students thinking that OLS is all there is we have done a terrible job. I was guilty of this for years before Sir David Cox pointed out logistic regression. Business professors seem not to be Aware of this at all.i got a nice paper once by using the right regression but business journals wouldn't touch.it. as a finance reviewer once wrote, "esoteric statistical procedures have no application in
Finance." The times they are a'changin I certainly hope. David Booth
1) plot the residuals. Are they clearly not normal, or is this just the outcome of a normality test? I would trust the graph more than the test.
2) plot the data for each independent variable. Can you see any relationships?
3) What do other researchers do who have reported data similar to yours?
4) Are there other problems? Correlated independent variables comes to mind.
5) Are all independent variables significant in the model?
6) Are there stripes in the residuals. Such might indicate a missing IV.
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling