Multi variate correlations without underlying models may be a waste of time

I know multi variate correlations are hard to figure and perhaps, without models, not particularly meaningful.