2
$\begingroup$

The JackStraw is a method to get honest p-values for correlations between variables and principal components derived from those variables.

JackStraw paper

There is a close relationship between PCA and clustering: K clusters gives you a signal in the top K-1 principal components, as outlined in this paper on PCA for detecting population substructure in genetic data. There is a more thorough and less sloppy discussion in this CV thread.

Does anyone know of a JackStraw equivalent that gives honest p-values for correlations between variables and cluster indicators? Preferably for an arbitrary clustering method and an arbitrary two-sample test?

$\endgroup$
2
$\begingroup$

This new preprint, "Statistical Significance of Cluster Membership with Applications to High-Throughput Genomic Data" answers your question. It tests association between variables and their computed cluster centers: https://www.biorxiv.org/content/early/2018/02/23/248633

The developmental version of the jackstraw R package includes these new functionalities. There's a dedicated and refined function for k-means clustering, as well as one that should work with a range of arbitrary clustering methods: https://github.com/ncchung/jackstraw

$\endgroup$
  • $\begingroup$ We are trying to build a permanent repository of high-quality statistical information in the form of questions & answers. Thus, we're wary of link-only answers, due to linkrot. Can you post a full citation & a summary of the information at the link, in case it goes dead? $\endgroup$ – T.E.G. Apr 1 '18 at 13:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.