One can only test hypotheses about observable data sources. Some hypotheses, however, may require mentioning non-observable data-sources. For example, suppose there is a polling centre with three voting stations, and two polling stations show candidate A as winner with 60.1% and 58.7%, respectively, whereas the third shows B as a winner with 55%. This polling centre looks suspicious, assuming that voters are assigned uniformly to polling stations, because we would have expected comparable voting preferences across all three polling stations. If voters are not uniformly distributed among polling stations but alphabetically, it may very well be that several families all voting for B were registered with the same third polling place, providing a benign explanation of the discrepancy. For a statistical analysis, the data source of valid invalid votes for each polling station is observable, but to become a testable hypothesis, non-observable information about on how voters are allocated to polling station must also be considered.