Investigating the multimodality of multivariate data with principal curves

We propose a simple method to assess the number of subpopulations in multivariate data by projecting the data on its principal curve and then applying Silverman’s bandwidth test to the resulting univariate sample. Our results indicate that this method works well even in high-dimensional settings with relatively small sample sizes, provided that the number of subpopulations is not large compared to the number of dimensions.