jasp-stats / jasp-issues

This repository is solely meant for reporting of bugs, feature requests and other issues in JASP.
59 stars 29 forks source link

Network module: different results in jasp 0.14.1 and 0.15 #1434

Open CogniMath opened 3 years ago

CogniMath commented 3 years ago
vandenman commented 3 years ago

Hi @CogniMath thanks for reporting this issue! Would it be possible to share the .jasp file? Note that to upload jasp files you have to zip them first.

CogniMath commented 3 years ago

Hi, thank you for your message. I attach the jasp file. I selected network, then split by area, selected the network plot and the centrality plot, and finally moved the items q1 to q21 to the variables. These are the date from our research DatosNew2021v1.zip .

tomtomme commented 9 months ago

@CogniMath & @vandenman I tested with 0.18.3 The error message:

bootnet crashed with the following error message: Correlation matrix is not positive definite.

When I switch correlation method from "auto" to anything else, I get results. Is this intended with this kind of data? Is "auto" a sensible default, if it does not work always?

tomtomme commented 8 months ago

@CogniMath So I just found that this error about different results of the network-analysis across versions appeared in the past. Vandenman explains the details here https://github.com/jasp-stats/jasp-issues/issues/1338 This addresses your issue hopefully.

It does not explain however the question why the correlation method "auto" is default, when it does not always work.

tomtomme commented 5 months ago

@vandenman I guess it is to late to fix this for 0.19? A colleague of mine also stumbled upon this and had to switch to Npn instead of Auto.

tomtomme commented 1 week ago

The error message is now different with flatpak beta 0.19.2:

Summary of Network bootnet crashed with the following error message: some categories of variable `q13' are empty in group 1; frequencies are [30, 30, 17, 7, 0]

Is this intended behaviour now? The message seems to imply that I have to remove q13. When I do that, I am back to square one aka the original error message... so this seems only partly solved.