I am searching for some literature or references, or also only some tips for the topic of test data generation that is realistic and what it means to the data to be realistic.

Example Questions that are arise is how to generate data for time series, generate trends, realistic distributions, correlations and correspondence between data instead of always using random values. An additional question for me is that although almost everything is normal distributed, sometimes a normal distribution looks very unrealistic, so what is a good way to bring some noise to it?

  • 1
    $\begingroup$ Who in the world told you that "almost everything is normal distributed"? That's one of the most incorrect things I've ever heard. $\endgroup$ – shadowtalker Feb 23 '16 at 11:50
  • $\begingroup$ @ssdecontrol: Maybe I exaggerated, but many natural / social / health processes are kind of normal distributed. $\endgroup$ – ScientiaEtVeritas Feb 23 '16 at 12:10
  • $\begingroup$ Not entirely -- my impression is that in those fields they frequently just assume approximate normality, regardless of whether it actually holds, so they can use statistical tests mechanically, plugging in numbers, without thinking, understanding any of the underlying assumptions, or having to argue that those assumptions are even approximately valid in the situation in which they are being applied. $\endgroup$ – Chill2Macht Apr 25 '17 at 17:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.