I computed a non-linear model to fit a set of observed data. The model has three parameters (a, b, c) and a very good fit to the data (R = 0.995). The computation results in separate 95% confidence intervals for each of the three parameters. Does it makes sense to separately randomize then each parameter within the 95% confidence range to obtain a range of predicted outcomes based on the model?

Also, what would be the best way to generate random numbers? Is it OK to use the RAND function in Excel (2003 and higher) to do this? Earlier versions of Excel before 2003 apparently had statistical problems with generating pseudorandom numbers, but this seems to have been fixed 2003 and onwards. Since most programming scripts generate pseudorandom numbers that do not meet the strict requirements needed for statistical analysis (computer clock via Mersenne twister algorithm), is there an advantage of using a script over Excel? My programming skills are fairly basic, so is there a recommended software function that would generate higher quality pseudorandom numbers with access e.g. to CryptGenRandom in Windows? Has anybody used packages like "random" in R that seem to give access to true random numbers?

Similar questions and discussions