clascalmc.1.txt

CLASCALMC(1)
============
:doctype: manpage


NAME
----
clascalmc - compares nested CLASCAL models using Monte-Carlo tests


SYNOPSIS
--------
*clascalmc*
 [*-g* 'uint']
 [*-n* 'uint']
'exp'
'sol0'
'sol1'


DESCRIPTION
-----------
Given *clascal* solutions 'sol0' and 'sol1' for experiment 'exp' and
assuming that the model of 'sol0' is correct, *clascalmc* uses a
Monte-Carlo method to estimate the probability that fitting the models
of 'sol0' and 'sol1' to new data would yield a likelihood ratio better
than the likelihood ratio of the given solutions. It then prints its
estimated lower bound and upper bound on this probability to standard
output. If this p-value is less than a desired alpha (1 minus the
confidence level), one can reject the model of 'sol0' in favour of the
model of 'sol1'.

For the test to be valid, the models must be nested, i.e., the model
of 'sol1' must be a superset of the model of 'sol0 .'  Thus,   

  * the number of dimensions in 'sol1' must be greater than or equal
    to the number of dimensions in 'sol0', and furthermore,
    non-spatial models may only be compared to other non-spatial
    models;

  * the number of classes in 'sol1' must be greater than or equal to
    the number of classes in 'sol0'; 

  * if 'sol0' has shared specificities, then 'sol1' must have either
    shared specificities or specificities for each class; and

  * if 'sol0' has specificities for each class, then 'sol1' must also
    have specificities for each class.

The options are as follows:

*-g* 'uint':: Use 'uint' to seed the random number generator for
the Monte-Carlo samples.

*-n* 'uint':: Use 'uint' samples to generate the Monte-Carlo
estimate. This value must be at least as large as the reciprocal of
the desired alpha (e.g., 10 for alpha = 0.1, 20 for alpha = 0.05, 100
for alpha = 0.01, and 1000 for alpha = 0.001); extra iterations
increase the precision of the estimate.  The default value is 100.


SEE ALSO
--------
clascal(1)


AUTHORS
-------
John Ashley Burgoyne <john.ashley.burgoyne@mail.mcgill.ca>