some thoughts and comments/feedback

Discuss the jamovi platform, possible improvements, etc.

by davidr » Tue May 16, 2017 7:44 pm

First, just want to start by saying thanks for building this software. It's excellent. I couldn't really get into using JASP because it's too Bayesian focused, and us frequentists seem to be an afterthought. I also appreciate the fast pace of development and incorporating so many user requests. And, the syntax mode is AWESOME!

Second, I switched to R earlier this year. I only open SPSS to ensure new analyses I write in R provide the same output as SPSS. You've done an excellent job keeping all of this sort of stuff behind the scenes (e.g., no changing the default contrasts for Type 3 SS!). I'm starting to have my students switch from SPSS to R and I have been telling them to start with jamovi if they feel anxious about the lack of an SPSS-type GUI in R. So far, they like it quite it bit.

But, this brings me come comments that I hope can be addressed moving forward. Some are general features, others are more for clarification.

1) I'm on macOS 10.12.4 running javmovi I can only load CSV files. Although, I've only tried loading CSV, Rdata and SPSS's SAV files to be fair. It would be nice if SPSS files could be imported, and SPSS and Rdata files could be exported alongside OMV and CSV. This would make collaborating with people using R, SPSS, Stata, etc. much easier. I recently switched to the 'rio' package for data import/export. It has nice defaults and supports pretty much every file format a psychologist ever needs.

2) I've been comparing some 2x2 between-subjects ANOVA output between SPSS, jamovi, and my own R script. I'm particularly interested in consistency in how each handles missing data, along with the statistical output. jamovi seems to struggle with missing data natively. For example, I loaded dataset I've been analyzing in R. Ran the exact same anova in the jamovi GUI as I ran in R using jamovi::anova(). jamovi spit out a bunch of blank stat tables, while the same syntax ran perfectly within R. In addition, I cannot figure out what "post-hoc tests" are being ran in jamovi. More specifically, which packages and commands are being used to compute the posthocs (using the postHoc command in anova()). I think it's drawing on both 'lsmeans' and 'multcomp' for this but I cannot recreate the 2x2 interaction output in my own R script using these two packages but I get identical main effect posthoc output using "summary(glht(model1))". Maybe there could be a separate page in the Developer's Hub or something that goes into more detail about the R commands or functions used? I also noticed the posthoc output provides t-scores. It would be nice if it gave Fs, eta-squareds, and CIs. I know F = t^2 but when less savvy users try moving from SPSS to jamovi, they might not pay attention to this and think jamovi is giving the wrong output.

3) Related to point 2, given the increased popularity in simple effect analysis and simple slopes, would it be possible to add these to the ANOVA and regression sections? Btw, awesome job adding hierarchical regression to jamovi!

Posts: 1
Joined: Tue May 16, 2017 7:09 pm

by Ravi » Mon May 22, 2017 3:13 pm

Hi davidr,

Thank you for all the kind words! Let me address your questions one by one:

  1. We will definitely add support for more data import types (e.g., SPSS, XLSX) in the future. For now you can copy your data directly from the spreadsheet and paste it into jamovi. We do acknowledge that the import functionality is important (you don't want to still need SPSS to access your old data files :slightly_smiling_face: ).
  2. Can you send me the .omv file in which this "empty table behavior" occurs so we can get to the bottom of this? You can just add attach it to this topic or send it to me in an email: selker [dot] ravi [at] gmail [dot] com. You can find the exact code we use for post-hoc tests over here: ... #L357-L373. It works like this:

    Code: Select all
    dat <- data.frame(var = factor(rep(c("A", "B", "C"), each = 20)),
                      dep = c(rnorm(60, .5)))

    model <- aov(dep ~ var, data=dat)
    referenceGrid <- lsmeans::lsmeans(model, ~var)
    summary(pairs(referenceGrid, adjust="bonferroni"))

    For the other requests, it's best to add a feature request on our github page:
  3. We want to keep our main analyses as simple and clean as possible. For more specialized functionality it's best to make it available as a module in jamovi. I definitely think that simple effect and simple slopes analysis would be a great addition, so let's hope that someone starts working on a module. It's our goal to create a community of people implementing analyses, in a similar fashion as R, and make it as easy as possible for people to create modules. In the end, everybody has their own idea of what's important to include and we want jamovi to reflect these different ideas.
User avatar
Posts: 162
Joined: Sat Jan 28, 2017 11:18 am

Return to General