His additional response to your other question:
"My understanding is that the Tukey-Kramer approach (which is what you get with the pairwise comparisons when the sample sizes are unequal) has markedly better Type I error protection properties than does the homogenous subsets approach with harmonic means when sample sizes are unequal. So no, what I've read would go the other way."
------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------
Original Message:
Sent: Tue February 28, 2023 04:54 PM
From: Rick Marcantonio
Subject: Multiple Comparisons vs. Homogeneous Subsets in SPSS Oneway ANOVA
A statistician responded:
"The HSD pairwise comparisons use a function of just the two sample sizes involved in each pairwise comparison, while (as noted in the footnote) the homogeneous subsets calculations use a common value based on the harmonic mean of all of the group sample sizes. The formulas are in Appendix G in the algorithms manual."
------------------------------
Rick Marcantonio
Quality Assurance
IBM
Original Message:
Sent: Tue February 28, 2023 12:36 PM
From: Donald Sharpe
Subject: Multiple Comparisons vs. Homogeneous Subsets in SPSS Oneway ANOVA
Hello. I'm getting discrepant results between Multiple Comparisons vs. Homogeneous Subsets. In one case (attached) ambiverts appear not to be different from undifferentiated (p. 124) in Multiple Comparisons but the Homogeneous Subsets suggest otherwise. I have other examples. Any ideas? Thanks.
------------------------------
Donald Sharpe
------------------------------