Backward elimination

[David Dubrovski]

Hey I have this regressive model here and need to use backwards elimination to remove non significant variables. Further, the task says that I should use ordinary standard error and a two tailed t-test on a significance level of 1%. I should also keep the constant even if its not significant.

So my first thought is that since the p value of both scan2d and scan3d are below 1% the variables are significant, so the model should be as it is?

[Carlo Lazzaro]

David:
your conclusion looks wise to me.
That said, the sky-rocketing values for R-sq may reflect the evidence that your last model has nothing to do with the original one (https://www.stata.com/support/faqs/s...sion-problems/).

[Nick Cox]

You're predicting mussel weight from two variables whose names suggest that they are at least loosely similar. I would not rule out there being a simpler model. Perhaps you would be better off trying to predict log mussel weight? Why not show your data with the following command?

dataex musselweight scan?d