Michaelis-Menten function fitting is a can of worms.
The main thing is to be aware of that and to have
looked at the literature on it. There is
a lot, going back decades; do not try to reinvent
wheels without reading first.
_Biometrics_ is a good journal here. You may be able to
exploit stuff on http://www.jstor.org
However, in Stata one attractive route is to reformulate
the problem as a generalised linear model with reciprocal
link.
See the exchange in
Generalized Linear Models for Enzyme-Kinetic Data
J. A. Nelder; D. Ruppert; N. Cressie; R. J. Carroll
Biometrics 47(4) (Dec., 1991), pp. 1605-1615
which I haven't read for some years, so the memories
are hazy. As I recall, the main idea is this. You have
y = ax / (1 + bx)
so
1/y = 1 / ax + b / a
Let us define
X = 1/x
and reparameterise
A = 1 / a
B = b / a
We then have
1/y = AX + B
and the right-hand side is then a piece of cake.
The left-hand side we take care of by
using a reciprocal link, another piece of cake
with -glm-.
In Stata terms
. gen rec_x = 1 / x
. glm y rec_x, link(power -1)
Of course, all this is just algebra with the
deterministic curve and says
nothing about error structure.
Nelder I guess recommends using a gamma
family.
Nick
[email protected]
Simon Moore
> I have a reasonably simple hypothesis that the form of relationship
> between the independent variable (x) and the dependent variable (y)
> follows the Michaelis-Menten rational function, f(x) = ax/(1+bx). I
> want to have this in a regression model with the cluster() option: reg
> y f(x) V, cluster(). But I can't see a way of achieving this and
> having reg solve for a and b. I thought maybe a power expansion of
> f(x) might work, but this does not seem appropriate.
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
