Curve Fitting Discrepanc​ies--Expon​ential versus Log-Transformed

I need to fit data using an exponential relationship, but I also want to make a statistical comparison of coefficient values between several models. Initially I was using the Curve Fitting tools to perform the fits, i.e.:

for i = 1:NumberofGroups ftype = fittype('exp1'); [coefs{1} goodfit{1}] = fit(x(Group==i),y(Group==i),ftype,options); end

In order to make comparison among the models easier, I set up a linearized version using regstats:

for i = 1:NumberofGroups Model{i} = regstats(x(Group==i),log(y(Group==i)),'linear'); end

I am also using aoctool to intercompare parameters as:

[h, atab, ctab, stats] = aoctool(x,log(y),Group);

and using multcompare to determine significant differences among slope and intercept parameters of the exponential fits.

The two methods (curve fit vs. linearization) produce very different coefficient values with different relationships among my groups. Obviously I am concerned about finding statistical significance for certain parameter groups that are only supported in one approach. Any assistance as to the differences between fitting methods in the two tool sets or advice on interpretation of such differences is greatly appreciated.

Answer 1

[bym]

Have you seen this?

http://www.mathworks.com/products/statistics/demos.html?file=/products/demos/shipping/stats/xform2lineardemo.html