I made up a little story to illustrate Simpson's Paradox. http://alan.sdf-eu.org/simpsons.html http://alan.sdf-eu.org/simpsons.html By carefully contriving the

I made up a little story to illustrate Simpson's Paradox. http://alan.sdf-eu.org/simpsons.html http://alan.sdf-eu.org/simpsons.html By carefully contriving the numbers, I was able to bring out a point that the Wikipedia page underplays. Should we believe the aggregated date or the partitioned data? Since the partitioned data can be partitioned again, potentially causing a second reversal, the partitioned data is the aggregate data with respect to the second partition. Not only do we do not know whether to go with the aggregate data or the partitioned data, we do not even know whether a data set is aggregated or partitioned.