Network Analysis in JASP: Correlations

richmullen

Hi,

I'm running some network analyses in JASP. Looking at the "Show info for this analysis" button, it says that I should be able to select Spearman correlations (my data are ordinal and skewed). Is there a reason this option does not then appear on the Correlation menu, or have I missed something?

Without the option to use Spearman correlations, if I select Auto, the correlation method is automatically implemented, "This will detect continuous, binary and ordinal variables and will use Pearson, tetrachoric or polychoric correlations". How do I know which method has been used?

Thanks in advance for any advice anyone can offer.

EJ

I'll attend our network experts to this!

Cheers,

E.J.

Jill

Hi Rich, 

Unfortunately, it is not yet possible to select Spearman correlations (sorry for the confusion, as it is now stated wrongly in the help file as an option). We’ll add Spearman correlations as soon as possible! 

Relating to the Auto correlation: Polychoric correlations are used when ordinal variables are detected (<7 categories), tetrachoric correlations are used when the variables are dichotomous, and Pearson correlations are used for continuous variables. I hope that helps!

Best, Jill

richmullen

Thank you Jill. Yes, that’s clarified things for me. Much appreciated.

Konstantin

Hello everyone,

I want to run a network analysis in JASP. My data are binary (25 variables) and collected from 1000 participants. Can I perform this analysis in JASP using the EBICglasso estimator and "npn" correlation method? Or would you prefer using an eLasso method? (https://www.nature.com/articles/srep05918).

Many thanks in advance, Konstantin

EJ

I'll attend Jill to this (but I would suggest consulting the literature as well)

E.J.

Jill

Hi Konstantin,

EBICglasso estimation method assumes Gaussian data. For binary data it is recommended to use an Ising model (also discussed in the paper you reference). This can be done by selecting the option IsingFit or IsingSampler under "estimator". IsingFit uses a form of eLasso regularisation (assuming a sparse underlying network) and IsingSamper is an unregularized estimation technique.

Best,

Jill