PHD Discussions Logo

Ask, Learn and Accelerate in your PhD Research

Question Icon Post Your Answer

Question Icon

4 months ago in Statistical Modeling By Radhya Kumari

Can i use response surface methodology for power law models?

RSM is all about fitting quadratic polynomials. But my system follows a power law—y = a·x?^b·x?^c. Do I have to abandon RSM, or can I make it work?

All Answers (1 Answers In All)

By Akash Answered 2 months ago

You can absolutely make it work—logarithms are your friend. Take the natural log of both sides: ln(y) = ln(a) + b·ln(x?) + c·ln(x?). Boom, it's linear. Now design your experiment in the log-space of the factors, use a standard RSM design (like Central Composite), fit a linear regression to the transformed response, and then transform back. You're still using RSM; you're just being clever about the coordinate system.

Your Answer