Taklo Simeneh Yazie, the decision to include variables in a statistical model should be based on your hypotheses coming from theory.

Hello Taklo,

You haven't indicated what your specific research aims and research questions are, so it's hard to give advice.

If what you are asking is, what conditions justify removing an independent variable from a regression model, then the answer depends on the goals of the model-building.

As Mehmet's answer suggests, many people would not bother with a model modification, but would instead report on the variables which did and the variables which did not appear to make noteworthy contribution to the explanatory power of the model. After all, if a set of IVs was thought to be sufficiently viable to evaluate in the first place, then attention should be given to each in the analysis and interpretation.

In some applications, e.g., data mining, there really is no specific theory or rationale guiding the seeking of structure, so different guidelines might apply. But for purposively chosen IVs, context, background, and theory matter.

Good luck with your work.

Suppose we assume your goal is prediction. Then p-value methods (such as the stepwise family) are very poor methods, see Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating | Ewout W. Steyerberg | download (b-ok.cc)

and

The attached papers, for further suggestions. BTW David Morse gives good advice. Best wishes, David Booth

Floating values

Thank you Booth, materials you shared are quite interesting