2

I have fit a non-linear model with 3 parameters to a data set.

I have a preconceived notion that one of the parameters will be one of 2 values. So I would like to compare 2 fits of the same model using the two theoretical values for the fixed parameter.

Because the 2 fits are not nested and hence DF is same in both fits, my vague understanding is that I cannot use an F-test on the models SS's. Is this true?

Also, R^2 does not seem appropriate because it is a non-linear model.

What other options should I look at?

Thanks

omian
  • 91
  • 1
    Is one of them a "null" value? If not, you could just directly compare likelihoods (if you don't have to formally test). – Glen_b Aug 28 '13 at 13:56
  • No, neither are null values. Could you provide a link for information on comparing likelehoods please? What if I do want to formally test? – omian Aug 28 '13 at 13:58
  • Actually, what is meant by null value here? A fixed parameter of zero or null in the sense of a null hypothesis? – omian Aug 28 '13 at 14:04
  • A null value in the second sense you gave. As for information on comparing likelihoods I mean simply this: 'calculate two likelihoods; the one that is bigger is more likely to have produced the sample'. If you like you could compare say AIC, but when the $n$ and $p$ are the same, it's no different from comparing likelihood. – Glen_b Aug 28 '13 at 20:57

0 Answers0