I need help in understanding how to test whether a statistic conforms to a
known distribution so I can establish levels of significance. The statistic
I'm using is one that's used to test similarity between two profiles (i.e.,
a variant of the sum of squared differences). This statistic can be
calculated in two ways: (1) assuming uncorrelated profile elements and (2)
with correlated profile elements. The uncorrelated version does follow the
Chi-square distribution and tabled significances exist. No such table exists
for the correlated version - which is the formula I'm using.
So, I've got a large sample of individuals (N = 1108) who completed the
survey. I can calculate the profile similarity coefficient comparing each
respondent against every other respondent (the number of comparisons is
about 1.22 million). I can generate the distribution of coefficient values.
But, how do I test whether the distribution of scores follows the Chi-square
distribution (or some other)? Then what do I do to generate a table of
Does someone know how to do this and can guide me through the process?
Alternatively, can someone point me to references (for non-mathematicians)
that explain how to do this kind of thing?
Edgar F. Johns <EdgarJohns@obik.com>
2906 River Meadow Circle
Canton, MI 48188
Tel. 734.495.1292, Fax 734.495.1981