ANOVA's post-hoc tests algorithms

Comments

• EJ

  January 2019

  Hi Peter,

  I'll ask Johnny for more information on these tests. They come from R packages, and I think we've checked them against results from other programs, but Johnny knows more about that. Nice project, by the way!

  Cheers,
  E.J.
• JohnnyB

  January 2019

  Hi Peter,

  So JASP features multiple p-value corrections (Scheffe/Holm/etc) - these are done with the multcomp package (or sometimes with the p.adjust function). This package is used for the standard post hoc comparisons and it provides us with the different p-value corrections.

  For the different types of post-hoc tests, you can look at the JASP source-code (Here is the AN(C)OVA R-code) and search for the following functions:

  • .anovaPostHocTable (uses the multcomp package)
  • .anovaGamesTable (analogous to the oneway function from the userfriendlyscience package)
  • .anovaDunnettTable (analogous to the dunnett.test function from the asd package)
  • .anovaDunnTable (analogous to the dunn.test package)

  These functions have the R-code for the actual computations and should be fairly straightforward. The code also includes some markup stuff for the tables, but if you scroll a bit past that you will find the computations.
  We have verified the results by comparing it to data examples in statistics text books, such as the ones by Andy Field, but please let us know if you find any anomalies

  Kind regards
  Johnny
• pklaren

  January 2019

  Hi EJ and Johnny,

  Thanks for the quick reply, as always!

  I will look into the R packages for the algorithms. Hopefully I will have sorted them out by next Wednesday when my course starts.

  Still a bit confused about the distinction "correction" and "type" in the post-hoc tests screen. I would have thought that Bonferroni and Holms truly correct alpha, and that Tukey and Scheffé are different types of procedure in their own right. Scheffé's test also works on contrasts, but I have yet to figure out how to do that.

  What is a "standard" analysis: repeated t-tests?

  All in all I am happy with the added functionalities in JASP 0.9.2.0, but I do think that the help-function is lagging behind somewhat. My (undergraduate) students, still working with v 0.9.0.0 well find their way around in JASP, but from an educational point of view they should have a context-sensitive help function (also for the more advanced procedures) at their fingertips. Keep JASP transparant!

  Cheers,

  Peter