Re: Chi-square for multinomial data

Hi Clara,

Is there a way to scale the two so that I can use the test?

Suggestions:

1. Based on what you know, make a reasonable guess as to the N for the
percent data, and multiply to produce estimated frequencies. Then use
a chi-squared test to compared the distributions.

2. Alternative: try the same thing using a lower limit for the likely
N. If the comparison is statistically significant that way, it will
also be statistically significant assuming a larger N.

3. Alternative: try both a lower and upper bound estimate, calculate
chi-squared for both. Perhaps you will get the same result
(significant or non-significant difference) both ways.

should I just present both data sets in terms of a percentage and base
the goodness of fit on a visual comparison of the percentages?

You can do that, too. In fact, if you do a chi-squared test and find a
significant difference, this can help you see which cell(s) contribute
most to the difference.

Hope this helps.
--
John Uebersax PhD

.