Logistic regression returns improbable non-sig results

Discuss statistics related things

by mihaidricu » Fri Mar 05, 2021 8:04 am

Hello !

Running a Binomial Logistic Regression gives me results that should not be possible.

I am trying to determine whether qualifying for Round 1, Round 2, National Finals and International Finals (factor Round) in a Science Olympiad determines whether respondents recommend participating in the Olympiad ("yes"/"no").

Now, everyone who made it to the International final has said "yes" so the probability computed is 1.0. However, when this probability is compared against Round 1 (0.35 probability compared to "no") it tells me that the difference is not statistically significant. However, comparing Round 1 (0.35) to Round 2 (0.75) rightfully says it is highly significant.....

Is this a bug of jamovi when probabilities are 100%?

Thank you!
Attachments
Screenshot 2021-03-05 08.44.28.png
Bonomial logistic regression
Screenshot 2021-03-05 08.44.28.png (113.35 KiB) Viewed 537 times
mihaidricu
 
Posts: 17
Joined: Thu Feb 27, 2020 2:21 pm

by jonathon » Fri Mar 05, 2021 10:38 pm

Is this a bug of jamovi when probabilities are 100%?


probably not a bug in jamovi, possibly a property of logistic regression.

a lot of statistical tests don't work when you have no variance in a cell.

i personally don't know enough about this procedure to say more, perhaps someone else will chime in (but i think we're past the point where people are finding bugs in analysis results, so if you think there's a bug in jamovi here, you'll have to do a bit more leg-work to demonstrate it ... i.e. demonstrate a different result in different software)

cheers

jonathon
User avatar
jonathon
 
Posts: 1729
Joined: Fri Jan 27, 2017 10:04 am


Return to Statistics

cron