[RsR] Robust ANOVA with no interaction

[Rand R Wilcox]

There is a paper regarding the strategy of testing for no interaction and using an additive model if the hypothesis of no interaction is not rejected:

Fabian, V. (1991). On the problem of interactions in the analysis of variance.
  Journal of the American Statistical Association, 86, 362--367.

The conclusion is that this strategy is not recommended roughly because it is unclear when power is sufficient high to justify using an additive model. 

For this reason, a method assuming an additive model has not been added to WRS2. If despite the results in Fabian there is some reason to include a function for an additive model, let me know and one will be added.

Best,

Rand

Rand Wilcox
Professor
University of Southern California
3620 McClintock Ave
Los Angeles, CA  90089-1061

Access to some reprints:
http://www.tandfonline.com/eprint/ypyq8ZKENfFXERFghYq7/full

________________________________________
From: R-SIG-Robust <r-sig-robust-bounces using r-project.org> on behalf of Worthington, Thomas A <thomas.worthington using okstate.edu>
Sent: Tuesday, March 3, 2015 12:45 PM
To: r-sig-robust using r-project.org
Subject: [RsR] Robust ANOVA with no interaction

Dear All

We are trying to use the WRS2 package to run a robust ANOVA but am having some issues with model specification

The dependent variable is velocity and the predictors are temperature (Temp) and total suspended solids (TSS).  We would like to run a 2-way robust ANOVA using 't2way' however one of the groups of the interaction is missing. We therefore ran a 1-way robust ANOVA using the dummy variable Temp_TSS e.g.

t1waybt(Velocity ~ Temp_TSS, data=data, tr=0.2, nboot=10000)

The result suggested that the interaction was not significant . Therefore we would like to run a robust ANOVA with both main effects but no interaction (due to the missing group). We tried

t1waybt(Velocity ~ TSS + Temp, data=data, tr=0.2, nboot=10000)

but this only tests for first variable stated (in this case TSS). Is this possible to run an ANOVA with just the main effects but no interaction?

Thanks
Tom

_______________________________________________
R-SIG-Robust using r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-robust