Linear or quadratic best fit model

382 Views Asked by At

The statistics question asks which would be the best method to determine if your data would best be fit with a linear regression model or a quadratic model.

I would think that the residual plot would be the best method. The other 2 choices are the shape of the scatter plot and the correlation coefficient.

I have seen scatter plots that look very liner and have a $.9$ R value, yet when you look at the residual plot you can see a distinctive pattern or curve.

So, isn't the residual plot not only the best but the ONLY method? I guess I'm no longer sure what the R value really does, since R values of .9 can still NOT be a good linear fit.

1

There are 1 best solutions below

0
On BEST ANSWER

I would go with a q-q plot if given the opportunity, but residual plot is best from choices you are given.

The $R^2$ captures the basic notion of goodness of fit, but it is insufficient here. For example, if your linear regression leaves residuals which are not looking like Gaussian errors, even with a large $R^2$, you can likely do better as there seems to be some inherent effect in the data not yet described by your model...