I'm measuring binding constants my system and I appreciate the usual methods of using replicates to measure standard errors and using those errors to calculate propagation of error. I'm curious if bootstrapping is a reasonable alternative to calculate uncertainty?
My understanding of bootstrapping is that you may estimate variance (and thus standard error of the population mean) iff your measurements are independent and have the same population distribution, in which case a number of sampling-with-replacement calculations can be done. I suspect that this method of estimation would be less desirable with small n values because of the effect that outliers or large standard deviations may have on the calculation. If you have a large dataset then it shouldn't be an issue.