I am working on clinical synthetic data and I would like to learn more about metrics to compare synthetic vs measurements distributions.
As there are methods to generate synthetic distributions with given constraints extracted from measurements,
I wonder whether you have pointers for metric definition apt to assess whether a synthetic distribution can be used as surrogate of a measured one.
I think that classical metrics have been often used to do that, but I can't find a review assessing the errors which have been introduced. I will be grateful for any suggestions. Andrea
PS E.g. I was wondering whether a vector containing classical distribution descriptors like mean, median, skewness, kurtosis for each distribution could be used taking their correlation as a value of similarity (i.e. if correlation = 1 there is high probability that the two distribution could be used as surrogate of one another).... just thinking loud. I could not find anything similar used so far.
