Mixed Anova with a Dichotomous DV

[Evan Harrington, Ph.D.]

I'm seeking an answer to the following:

My research design involves a mixed ANOVA with a 4 level Between-Subjects
variable and 3 level within-subjects variable.

I have a couple interval DVs that gave nice results, but I also have a
dichotmous DV that I would very much like to analyze with the same mixed
ANOVA design, and when I do I get the results I hypothesized.

I have Lunney's article from J.E.M. on using ANOVA with dichotomous DV,
but after looking for other examples I have seen almost no one saying it
can be done (due to the violations of the ANOVA assumptions).

My question is really: how defensible is using ANOVA with dichotomous DVs?
It looks like it's working properly and although it violated the
assumption of sphericity, when I examine the Greenhouse-Geisser adjustment
it remains significant (p=.02). Will I be on safe grounds if I simply
follow Lunney's suggestions?

Many thanks!

Evan Harrington, Ph.D.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Bruce Weaver]

Evan Harrington, Ph.D. wrote

I'm seeking an answer to the following:

My research design involves a mixed ANOVA with a 4 level Between-Subjects
variable and 3 level within-subjects variable.

I have a couple interval DVs that gave nice results, but I also have a
dichotmous DV that I would very much like to analyze with the same mixed
ANOVA design, and when I do I get the results I hypothesized.

I have Lunney's article from J.E.M. on using ANOVA with dichotomous DV,
but after looking for other examples I have seen almost no one saying it
can be done (due to the violations of the ANOVA assumptions).

My question is really: how defensible is using ANOVA with dichotomous DVs?
It looks like it's working properly and although it violated the
assumption of sphericity, when I examine the Greenhouse-Geisser adjustment
it remains significant (p=.02). Will I be on safe grounds if I simply
follow Lunney's suggestions?

Many thanks!

Evan Harrington, Ph.D.

Assuming you have a recent enough version of SPSS, you can use the GENLIN procedure to analyze a binary outcome variable.  In the pull-down menus, Analyze - Generalized Linear Models - Generalized Estimating Equations.  I expect the Help files will have an example or two that are not a million miles from your design.