RNDTEST <SURVO_data>,<variable>,<output_line>
makes empirical statistical tests on a series of numbers supposed to
form a random sample from a uniform distribution in the interval (0,1).
Most of these tests are explained in Volume 2 of "The Art of Programming"
by Donald E. Knuth.
The main application of RNDTEST is testing of various random number
generators.
A standard set of tests is performed without any extra specification.
However, If RESULTS=0, no test is performed without explicit specifica-
tions. To select tests in a more detailed form, following specifications
can be given.
RNDTEST: specifications
SUBSAMPLES=<size>,<# of classes>
The sample is divided systematically in subsamples of given <size>
and their uniformity is tested by the standard Chi^2-test by divi-
ding the interval (0,1) in <# of classes>.
Also tests for mean=0.5 as well for the minimum ans maximum values
in subsamples are made.
Default is SUBSAMPLES=0 (i.e. this test is omitted).
FREQUENCIES=<# of classes>,<lower limit>,<upper limit>
The uniformity of the total sample is tested by the Chi^2-test.
Default: FREQUENCIES=10,0,1
MAXLAG=<largest_lag>
The autocorrelations of the series are computed up to the given
maximum lag. Default: MAXLAG=10
RNDTEST: specifications
GAPTEST=<lower_limit>,<upper_limit>,<max.gap>
The lengths of gaps between occurrences of values in the given range
are computed. Default: GAPTEST=0,0.5,10
PERMTEST=<# of consecutive numbers (3,4,5,6 or 7)>
Frequencies of different permutations of relative orderings are computed.
Default: PERMTEST=4
POKER=<# of obs.>,<# of classes>,<lower limit>,<upper_limit>
Default: POKER=5,5,0,1
COUPON=<# of classes>,<max_len>,<lower limit>,<upper limit>
Coupon collector's test Default: COUPON=5,20,0,1
Certain run tests are performed in any case.
R = Simulation and random numbers
Help System (web edition)