
Exercises 417
because it would be necessary to divide by 0 if the chi-square test were strictly
applied. It is also difficult to find chi-square tables w ith more than about 30 or 40
degrees of freedom.
In general, a good guideline is that you should use the chi-square test only when
20 percent or fewer of the expected counts are < 5 and all indiv idual expected counts
are at least 1 (Moore 2001).
It should be obvious that, when using a computational intelligence system for
modeling or simulation, the goal is for the chi-square test to yield the smallest value
feasible. In other words, the goal is for the differences between the modeled and the
modeling ...