2
$\begingroup$

I need some kind of hypothesis test (or at least a reasonable rule of thumb) that will enable me to validate if observed minimum and/or maximum is "close enough" to the theoretical minimum/maximum. I have randomly generated values from some bounded, continuous probability distribution and I need to ensure that the pseudo-random generator produced values that reasonably stay within the bounds, how can this be achieved?

$\endgroup$
  • $\begingroup$ Is your concern about some source of error (is this implementation right/sufficiently accurate), or is it in relation to the possibility getting a sample (or samples) that are by chance far from what you'd usually see? $\endgroup$ – Glen_b -Reinstate Monica Jun 20 '17 at 1:15
  • $\begingroup$ @Glen_b I was rather thinking about the second thing, but as you wrote it, looking at errors also may answer the similar question. $\endgroup$ – Tim Jun 20 '17 at 7:59
2
$\begingroup$

If you know the density and distribution function, you can use the distributions of the first and last order statistics. If $n$ is the sample size, $f(x)$ is the pdf and $F(x)$ is the distribution function, then the first order statistic (the minimum) has pdf $$f_{1}(x)=nf(x)[1-F(x)]^{n-1}$$ and the $n$th order statistic (the maximum) has pdf $$f_{n}(x)=nf(x)[F(x)]^{n-1}.$$ If $L,U$ are the lower and upper bounds respectively, and $x_{(1)},x_{(n)}$ are the min and max from the sample, then $$P(X_{(1)}\leq x_{(1)})=\int_{L}^{x_{(1)}}f_{1}(x)dx$$ and $$P(X_{(n)}\geq x_{(n)})=\int_{x_{(n)}}^{U}f_{n}(x)dx.$$ You could then base a test of the hypotheses that the theoretical bounds are true on these distributions.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.