Bootstrapping was used to evaluate the robustness of the results, as explained subsequently for each type of analysis in Section 3.3. Bootstrap samples were constructed by random selection with replacement of observations, not of subjects. Each observation received a value of 0, 1, 2, … max number of times selected, and then when the weighted covariance calculations were performed, each weight was multiplied by the value received in the randomization procedure. One thousand bootstrap samples were created.
