I have been preparing an analysis of a data set from which I have extracted a number of covariance matrices and used the Log-Euclidean algorithms to calculate distances, geodesics etc. This data set contains multiple observations from the same individuals. I did a bootstrapping analysis to understand the variability in the data set using 1000 resamplings of the observations, not the individuals. Pairwise distances within each resampling were found to be consistently larger than the same distances within the observed data. Why this apparent scaling up? Are there any tests I could conduct that would shed light on this question?