Help determining start values of coefficients for a nonlinear model

#1
Hi everyone. For a dataset consisting of three quantitative variables, H, M and W I have to build a non linear model of this form: E(H)=b0+b1*M+(W/(b3+b4*M)).
I tried using the "nls()" function in R, but I don't know how to determine the start values of the coefficients, b0, b1, b3 and b4. Can you help me determine them? And if you succeed, can you also tell me how did you obtain them? I provide the data file (within the dataset: W = wl, B = bm, H = ho). Thank you for any help.
 

Attachments

Last edited:

katxt

Well-Known Member
#2
This thread looks familiar.
One easy way is to use Excel Solver.
Set up 4 variables with what you think is reasonable (or 1 1 1 1 if necessary)
Make a column of predicted values
Make a column of errors^2 and get total sum of squares.
Use Solver to change the variables to minimise total SS
I started with 1 1 1 1 and got
-116.593 4.212815 0.031834 0.003913 for an OLS solution.
Any help?
You could just use the values -138+4.5M+(W/(0.08+0.003M)) from your other post
 

katxt

Well-Known Member
#4
This is just least square error. Make sure you have the Solver addin installed (Google it if you haven't).
1642443863834.png
You don't need named ranges if you don't want to.
Set solver to minimize G7 by changing G3:G6
This is the solution that most non linear software will give, but this sheet doesn't give you p values of SEs.