Thanks Dr. Pierluigi Traverso ! The article is interesting. Nowadays, I see some research papers publishing non-linear regimes in sensing. Hence it is still a puzzle for me.
Nachiket Gokhale not only the linear range is important for calibration, it happens that the evaluation of parameters such as limit of detection and quantification are also based on a linear method. That is why linearity is important, because it is the basis for other analytical performance parameters. I know that there are calibrations for non-linear responses, but then you would have to see how validated are the procedures to evaluate the other analytical parameters.
In the aspect of data analysis, you can fit any non-linear data perfectly with a polynomial fitting but such fitting is not confident at all - any time you repeat the experiment, even the power of the fitting function may be different consequently. Given that many sensors do not have a very clear mechanism for their response, especially when the reaction involves more than one electron transfer and some chemical steps or absorption/adsorption, linear fitting is always the simplest and thus adequately reliable empirically, and possibly useful in practice.
Non-linear repsonse should be handled very careful, either the physiochemical mechanism behind is sufficiently clarified or the fitting is indeed special (e.g. sigmoidal) and highly reproducible. Normally we try to avoid working on such ambiguous results. PS, those papers you noticed might not be very reliable.