I am working on my thesis and I had to do a multiple regression. I have only 45 samples and I checked that there were not problems of multicollinearity (VIF < 3) and the residuals are normally distributed.

The overall model is statistically significant (p=0.001) but all the predictors are not significant except the constant. How is it possible? Which type of analysis can I do? Thanks a lot for your support. Dam

  • $\begingroup$ You may not need much dependence to render all the predictors non-significant. In fact it's even possible to get significant regression but no significant predictors with completely independent predictors (because the regions of joint non-significance and all individual variables being non-significant don't correspond exactly. e.g. with two standardized variables the region of joint insignificance is the inside of a circle while the region of individual insignificance for all variables is the interior of a square. (See the second image in Christoph Hanck's answer to the second duplicate) $\endgroup$
    – Glen_b
    Jul 19, 2019 at 11:02
  • $\begingroup$ I believe there's another post with this information but I didn't locate it $\endgroup$
    – Glen_b
    Jul 19, 2019 at 11:21


Browse other questions tagged or ask your own question.