Further analysis was done using mixture models, assuming that the observations were generated from two distinct distributions. The mixture models fit the data well, and pointed out further differences.
As might be expected, durations from the broadcasts were more "regular," coming predominantly from distributions with concentration in a middle range of values