The computation of the inverse of the F-distribution sometimes fails! For example, for n1=10, n2=1 and
F(x)=0.99 the algorithm doesn't converge (but this will be reported). As for the inverse function different algorithms are used,
you can try the direct way and find that x=6056 will produce F(x) near 0.99. Only for the Gamma and the Chi Square distribution,
both functions are related (that's the reason why these algorithms are quite slow and the last advice on this page will not help). In general the computation of the inverse functions can be critical
if you choose values near 1 (like 0.9999999). The largest possible value for F(x) is 0.99999999999999994 (on my machine).
|
| For the discrete distributions a value as F(100) is usually not
calculated as (f(0)+f(1)+f(2)+...+f(100)).
I have to trust on available algorithms for gamma functions, incomplete gamma functions, inverse beta functions and other nice things.
These algorithms are only exact to a certain precision and sometimes the errors cumulate and the result is not usable. But for "normal" values
like k=20 or F(x)=0.95 I haven't found an error (if the program reports, that it couldn't calculate the desired value, it's not a bug, it's a feature).
|
The algorithms of the inverse of the Gamma and the Chi-Square-distribution
are quite slow, so please wait for a few seconds! When calculating many (maybe 1000) random numbers, your
browser will be unusable for up to a minute, depending on your machine.
|
|
The cumulative distribution function of the Hypergeometric distribution is calculated exactly (simple additions),
that's why I've restricted the computation: If more than 100 additions have to be done,
the calculation is not performed (except for trivial computations)!
| t-distribution: For n>5000 the Normal Distribution is used.
|
| Don't try to type in something like
"@#%", only an error will occur (and no usable result). So please concentrate and
don't try any tricks!
| I've tried my best, but I give no warranty that
the algorithms work correctly under all circumstances. And yes, you can paralyse your computer for a few minutes,
if you type in time-consuming parameters, especially for tables and random numbers.
If you are sure that you have recognised an error, please send me an
e-mail.
Tell me the name of the distribution and the values of the parameters
that have produced the wrong result.
|
| I wasn't able to find out how Javascript functions can be made public for all pages, so with every new
chosen distribution the "functions.js"-script is loaded again (but should be in the cache).
|
| Opera 6.0 doesn't like my scripts. It takes years to print the tables.
The random numbers are placed outside the box.
|
| I don't give any warranty
for the correctness of all calculations! An advice: if you need to calculate
F(x), so calculate also the inverse function of F(x), then the original
value x should appear (and vice versa).
|