Imagine you perform the same measurements, on the same samples, using two different techniques. From the experimental results of each of these techniques you can fit a complex valued function, which typically has the form y(x) = A*(ix)^B, with A and B (Real) fitting parameters.
I am trying to find out the proper way to compare the two function, to see if the methods are equivalent. From what I understand, ANCOVA seems the approach to follow, but for what I understand requires a linear relationship, and also I am not sure if it extends to C.
Hello Massimiliano Berardi I think in software like GraphPad Prism, define these two functions.
Then compare the curves and estimate their parameters.
Dear Massimiliano,
I think that the problem is a little simpler than you are considering. Although we say that regression in all its variants in linear, this does not preclude examining data generated in a non-linear way.
If I understand the question as described, you simply wish to describe whether the outputs of the two functions match. This agreement can be quantified using correlation, or intra-class correlations (ICC) depending on your needs.
It is possible to consider agreement in a non-linear fashion, but you will have to be explicit about what you mean agreement to mean in those cases.
If you actually meant to compare the functions themselves, this is a question of analysis, and not statistics. IN other words, statistics can certainly tell you about agreement between two approaches as calculated... but doesn't describe the world of possible relationships between general functions.
I hope this helps.
Massimiliano Berardi wrote: "I am trying to find out the proper way to compare the two function, to see if the methods are equivalent."
Like Bruce E Oddson, I take that to mean that you want to generate fitted values of Y from your two models and then determine how closely they agree. If that is what you want to do, perhaps the approach I suggested in a recent MedStats post will be helpful. Here's the link:
- https://groups.google.com/g/medstats/c/YR-2uLLacrc/m/NeAfgisZAgAJ
HTH.