I'm currently trying to estimate the following equation:

$$y = \text{const} + \beta \frac{x^\gamma}{x^\gamma+\rho^\gamma}$$

So, I have to estimate 3 parameters for every variable plus a constant. A natural choice would be to estimate the equation with Nonlinear LS. However, as usual my estimates depent greatly on the starting values. I also tried to estimate the equation via Maximum Likelihood, as the y-variable follows a Gamma distribution. Nontheless, as the number of independent variables goes up(in my actual model I have 13 x-variables and for each I have to estimate 3 parameters) the ML-algorithm can't converge.

Has anyone an idea, how to handle such a regression? Up until now I tried to estimate the model like this in R.

logLikFun_Gamma_optim1 <- function(param) {
  alpha1 <- param[1]
  beta1 <- param[2]
  rho1 <- param[3]
  gam1 <- param[4]
  rate1 <-  param[5]
  sum(dgamma(y, shape = alpha1+beta1*((x1^gam1)/((x1^gam1)+(rho1^gam1)))
             , rate = rate1, log = TRUE))
}

maxLik(logLik=logLikFun_Gamma_optim1,start=c(alpha1=1,
                      beta1=0.5,gam1=0.51,rho1=0.5,rate1=0.007),fixed='rate1',
                      method='CG',control=list(printLevel=3))
  • I have added a formula in mathml code for clarity. I hope that is ok. – Martijn Weterings Oct 10 at 12:45
  • 2
    What is unclear to me in this question is what you mean with "estimate 3 parameters for every datapoint" Does every datapoint get different estimated parameters? How does that work, what kind of estimation are you referring to? – Martijn Weterings Oct 10 at 12:45
  • 2
    I agree with @MartijnWeterings about the 3 parameters for each point ... I think it's 3 parameters overall. Also, your formula has a single x, but you say there are 13. It would also be helpful to know more about the data set (what is the sample size? What are the variables? etc) – Peter Flom Oct 10 at 12:55
  • 1
    If $y$ follows a gamma distribution then what does your equation $y= const + \beta \frac{x^\gamma}{x^\gamma+\rho^\gamma} $ mean? – Martijn Weterings Oct 10 at 13:05
  • What does ML stand for (maximum likelihood, machine learning, ....) – Martijn Weterings Oct 10 at 13:09

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.