Output for categorical variables

Discuss the jamovi platform, possible improvements, etc.

by DrE » Wed Feb 10, 2021 9:57 am

Dear Jamovi Crew,

I’m a college professor who discovered jamovi a few years ago and has since been spreading the words to my colleagues about it. We not only use it for our own research, but have found it to be the ideal tool to use in our introductory statistics classes, especially with gamlj since our data often required generalized models.

My purpose in writing it to suggest a feature that many may find helpful. It has to do with how the results of certain analyses with categorical variables is output. It’s easiest to explain this with actual data. In Portuguese one can make a diminutive by adding either the suffix “inho/a” to a word or another suffix. The graph below shows how the choice of suffix is dependent on what sound the word end in. It shows that words ending in “s” take a suffix other than “inho/a” while words ending in “o” take “inho/a.”

Image

The results of a binomial logistic regression give these results which include estimates that compare each word final sound to each other sound:

Image

Note. Estimates represent the log odds of "INHO/A = other" vs. "INHO/A = inho/a"

This is very useful information, but what would be helpful as well is an estimate that doesn’t compare each value to every other value, but merely indicates how likely each value of the independent value is to appear with the values of the dependent variable. That is, something that would go along with the bar graph showing that word final “s” predicts “other” and word final o takes “inho/a.”

There is another GUI for R that only does binary regression (http://www.danielezrajohnson.com/rbrul.html). It outputs logodds to show this kind of information:

factor logodds
a 5.972
o 3.713
i 3.556
u 2.306
l 1.219
e 0.119
r -1.407
stop -2.466
z -5.816
s -7.197

Perhaps an option to give this kind of output as an estimate, logodds, or odds ratio, could be included under the Model Coefficients dropdown box in Logistic Regression. In gamlj it could be put under Options menu.

This may not be a difficult thing to incorporate since the code for the GUI I mentioned, that produces this outcome is available here:

http://www.danielezrajohnson.com/Rbrul.R

You may be able to use it to give the kind of outcome I’m suggesting in jamovi and gamlj.

I do appreciate your taking the time to consider this request from a committed Jamovi user.

Best regards,

David Ellingson Eddington
DrE
 
Posts: 6
Joined: Sun Dec 30, 2018 3:00 pm

by DrE » Wed Feb 17, 2021 5:09 pm

Jamovi gives pairwise comparisons between the values of a variable. What would be helpful is deviation contrasts that show how each value differs from the mean, rather than from every other value.
DrE
 
Posts: 6
Joined: Sun Dec 30, 2018 3:00 pm


Return to General