JASP Friedman Conover test doesn't match R (posthoc.friedman.conover.test)

[rachael-reavis]

• JASP version: 0.9.1

• OS name and version: Mac Sierra 10.12.6

• Analysis: Connover's post hoc test (for Friedman)

• Bug description: I'm not sure if it's a bug or just something I'm doing wrong, but when I run friedman.test in R I get the same result. However, I get different t-stats & p-values for the Connover's follow-up using posthoc.friedman.conover.test -- I tried finding the underlying R code that JASP is using, but I couldn't figure out where it was.

• Add JASP files as a zip:

• Screenshot:

• Expected behaviour:

• Steps to reproduce:

  1. Go to '...'
  2. Click on '....'
  3. Scroll down to '....'
  4. See error

[JohnnyDoorn]

Hi Rachael, Could you please provide a .jasp or csv file? If you prefer, you can send it to me at j.b.vandoorn(at)uva.nl Kind regards, Johnny

[JohnnyDoorn]

Got it, thanks Rachael - I will look into this!

[JohnnyDoorn]

Hi Rachael,

Wanted to let you know that this is now fixed in the next JASP release. The issue was a package upgrade that we were not aware of. We used part of the PMCMRplus package, which is the sequal to the PMCMR package. It seems that the Conover test differs in results between these two packages, and I believe the output from PMCMRplus is the correct one. JASP now gives the same results as that one again. Please let me know / reopen this issue if you have any additional questions/comments!

Kind regards Johnny

[rachael-reavis]

Thanks!

On Fri, Mar 29, 2019 at 10:01 AM JohnnyDoorn notifications@github.com wrote:

Closed #314 https://github.com/jasp-stats/jasp-issues/issues/314.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/jasp-stats/jasp-issues/issues/314#event-2239324678, or mute the thread https://github.com/notifications/unsubscribe-auth/AKBIDL2B4yrl70hK8QL7a8P8BIgoiAQUks5vbhysgaJpZM4bQKjY .

[ksp200425]

Hi, I have a question on this. Has there been any issues with this update and PMCMRplus? I haven't been able to get the same results for the Conover test between R (with PMCMRplus) and JASP.

[JohnnyDoorn]

Hi @ksp200425,

Do you have a numeric example? When I conduct a Conover test in JASP and PMCMRplus I get the same results, for instance for the Bush Tucker Food example from the data library:

JASP file (change extension to .jasp): BushTuckerFoodConover.zip

Running it in R:

dat <- read.csv("Bush Tucker Food.csv")
# convert to long:
y <- unlist(dat[, 2:5])
gr <- rep(letters[1:4], each = 8)
id <- rep(1:8, 4)
PMCMRplus::frdAllPairsConoverTest(y, gr, id, p.adjust.method = "holm")

    Pairwise comparisons using Conover's all-pairs test for a two-way balanced complete block design

data: y, groups and blocks

  a      b      c     
b 0.0054 -      -     
c 0.0082 0.7991 -     
d 0.2565 0.2071 0.2565

P value adjustment method: holm

Kind regards Johnny

[ksp200425]

Ok I see, I did get it right in R, but for some reason got something different in JASP. I'm sorry if there is something I'm missing! .

[JohnnyDoorn]

My apologies, I was using the development version of JASP (available here), which already included a fix to bring the JASP results up to date with PMCMRplus, which also had a bug fix in their latest release. The new version will officially release somewhere in the coming weeks though!

[ksp200425]

Oh, Thank you so much! I will try it again with the development version.