We have identified a significant issue affecting 3dMVM whenever the R package phia (version 0.3.1) is installed. This version of phia produces incorrect degrees of freedom for t-statistics under some circumstances in the output, an issue that may not be obvious unless closely inspected.
More specifically, the degrees-of-freedom issue arises only when running 3dMVM with models that contain no within-individual (repeated-measures) factors. In version 0.3.1, phia incorrectly reports the residual sum of squares as the degrees of freedom. (In the example case found, it greatly reduced the degrees of freedom, which led to an artificial increase of p-values, e.g., from p=0.0015 with 33 DFs, to p=0.044 with ~3 DFs.) In other words, analyses are unaffected as long as the 3dMVM model includes at least one within-individual (repeated-measures) factor.
This issue is limited to 3dMVM and does not affect 3dLME or 3dLMEr.
A corrected version of the package (phia 0.3.2) was released in May 30 2025, and updating phia resolves the problem.
How to check your current phia version
Run the following at your terminal:
R -e "packageVersion('phia')"
If the output shows:
0.3.1
you must update the package.
How to update phia
Use the following command at the terminal:
R -e "install.packages('phia', repos='https://cloud.r-project.org')"
If you previously ran 3dMVM with phia 0.3.1
If your previous analysis involves at least one within-individual (or repeated-measures) factor, you should rerun your analyses with the updated phia version to ensure accurate degrees of freedom for correct t-statistic results.
Moving forward
We have now updated 3dMVM on GitHub so that the next AFNI release will detect phia 0.3.1 and stop with a message prompting users to update.
Please spread the word to anyone who may be affected.
Gang Chen