I have analysed some data for a research project using multiple linear regression. However, normality assumptions for this method were not met in my data (and could not be resolved using transformations of the data) and hence, I have decided to bootstrap confidence intervals for the regression coefficients.

Should I report the original significance estimates from the linear regression model I set up in a regression table? My intuition says "no", since obviously I have chosen the bootstrap because normality assumptions on which traditional significance estimates are based were not met. Therefore, it seems "dangerous" or at least misleading to report t-values, dfs and p-values of the regression coefficients.

My approach until now was to report the original estimates for b, ß, and the SE of b and in the last columns include the limits of the bootstrapped CIs. However, I am unsure about reporting standards (in general, but also specifically regarding the American Psychological Association's publication manual, 6th edition) and I wonder whether there are statistical reasons unknown to me that imply I should report the t(df) and p values?

Cheers, bunsen



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.