On 15 Jun 2003, mac55 wrote: > > > A simple question...... > > > > > > If you are comparing binary data from two groups with > > > underlying populations that are normal.....I assume you can use a > > > two sample Z test for binomial proportions or a chi squared test > > > on a 2x2 contingency table. > > > > > > What happens if the underlying populations aren't normal? I > > > know the Z test is out. Can you still employ the chi squared > > > distribution. If not, what non-parametric test could you use? > > > > Nothing in the logic of either test requires the assumption that the > > data be derived by dichotomizing a normal continuous variable. The > > Z test only requires that the number of cases be large enough that > > the normal approximation to the binomial is decent. > > You are correct. Let's say the data are skewed
You're talking about "data" comprising a variable that has only two values? In what sense do you then want to mean "skewed"? Do you merely mean that p-hat is close to 0 or 1 for one or both groups? > or that the numbers aren't large enough for the normal > approximation? Is Chi squared a valid test? Valid, possibly; but if the numbers are THAT small, you probably can't detect anything interesting anyway. But you can always try applying binomial distribution theory (which for small numbers may not be able to detect anything either). ----------------------------------------------------------------------- Donald F. Burrill [EMAIL PROTECTED] 56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
