Fractional polynomials and nonlinear least-squares estimation

[Andrzej Niemierko]

Is there a fundamental reason why fractional polynomials (either -fp- or -mfp-) are not allowed when using -nl- (nonlinear least-squares estimation)?
Thanks,
Andrzej

[Maarten Buis]

There is a pragmatic one: -nl- often does not converge without good starting values. -fp- and -mfp- fits many different models, so I would expect a fairly large proportion of those not to converge, which throws off the entire algorithm.

[Andrzej Niemierko]

Thanks Maarten. Is there any other approach to accomplish what

fp <x>: nl <x>

would do?