Hello and thanks for reading.

tldr at the end

For context:

I work for a company that sells analytical instruments for chemistry analysis. Big part of chemical analysis consists of comparing the response of a sample with the response of standards and this response is not aways linear. For a given analysis we use log/log to get a linear response equation. log (y) = a log (x) b.

The thing is, many users will run replicates for each point to build this regression.

I'm trying to understand what is considered more correct in this context. If it would be to firstly calculate the average of the replicates and than take the log of the mean value or the other way round.

From my research, I did understand that if the data is not normally distributed you should aways apply the log first because the log of a lognormal distribution will be a normal distribution and then you can calculate the mean.

(Tldr)

What I don't get is: will replicate measures of data that follows a non linear correlation be intrinsically a non normal distribution? That would grant that you should aways log first and take the average later. If not depending on the dataset it wouldn't matter what you do first