The same set of statistics can produce opposite conclusions at different levels of aggregation.