To be more specific, my experimental design is based on 3-way (4*2*2) ANOVA. Three independent variables: factor 1=type of microorganism (4 levels), factor 2=time (2 levels), and factor 3=moisture level (2 levels).
The interaction effect (microorganism*time*moisture level) was significant, but I do not know which interaction is significant.
So, in your opinion, which way is the best to use SPSS syntax in my case?
/EMMEANS=TABLES(microorganism*time*moisture_level) COMPARE(time) COMPARE(moisture_level)?
or
/EMMEANS=TABLES(microorganism*time*moisture_level) COMPARE(time)?
or
/EMMEANS=TABLES(microorganism*time*moisture_level) COMPARE(moisture_level)?
Your help is highly appreciated.
Thank you so much
Hi Sayed Ibrahim Ali
I performed the following SPSS syntax as you suggested:
/EMMEANS=TABLES(microorganism*time*moisture_level) COMPARE(microorganism*time*moisture_level).
But there was a Warning (The symbol * was found in the EMMEANS subcommand where a right parenthesis was expected, Execution of this command stops)
Then I tried the following syntax, and it worked:
/EMMEANS=TABLES(microorganism*time*moisture_level) COMPARE(microorganism) COMPARE(time) COMPARE(moisture_level)
Thank you again
Hello A. M. A. Al-Khdri. If I follow, you want to carry out interaction contrasts. E.g., you want to test one of the two-way interactions at each level of the 3rd variable. Is that right? If so, I fear that it is not terribly straightforward in SPSS (IMO). The 2008 article by Howell & Lacroix shows some examples, but the method they use requires fairly advanced understanding of how to generate contrast codes. I think it would be much easier to get the desired results using the MANOVA command that they show in some of the supplementary files. You can see their article and get the supplementary materials (i.e., the Appendix link) here:
- https://www.tqmp.org/RegularArticles/vol08-1/p001/index.html
Meanwhile, do you have access to any other stats packages? I know that the contrasts (I think) you want to look at are much easier to get using Stata. And I imagine the same is true of R or SAS.
HTH.
Thank you so much, Bruce Weaver. I will think about doing the same analysis using R. Appreciate your comment.
Is it the case that you're looking for something akin to "test one of the two-way interactions at each level of the 3rd variable." ? It may be possible that what you want is easy with SAS via slice (in a GLM model), or in R via emmeans with the | operator. But it may be the case that you'd have to construct the contrasts you want.
In agreement with Bruce Weaver here, but just to add: "The interaction effect (microorganism*time*moisture level) was significant, but I do not know which interaction is significant." It is worth noting that the microorganism*time*moisture level being p < .05 doesn't require that any simple interactions are statistically significant. this is one reason why interaction contrasts are a better approach, however, generally finding if they are statistically significant post-hoc is uninteresting as you can always find a contrast that is highly significant.
A better approach in my view is to use the interaction contrast coefficients to capture a pattern that you can interpret (ideally a priori) and then look to see what proportion of variation in the cell residuals it accounts for. This can be computed in various ways but the correlation of the interaction contrast weights with the residualized cell means is a statistic known as r_alerting and its square is the proportion of variation in means that the contrast explains (after stripping out lower order effects). This can often be very high (90+%) if the observed pattern matches the contrast coefficients closely.
This is more persuasive if you predicted the pattern a priori, but its still a useful explanatory tool post hoc in my view.
e.g., see https://psych.wisc.edu/Brauer/BrauerLab/wp-content/uploads/2015/01/Abelson-Prentice-1997.pdf
Thanks, dear Thom Baguley, and I really appreciate your valuable feedback.
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation