LIS 504 - Bad statistical analysis

Standard statistical tests sometimes lie. This happens especially when the number of observations is small. For example, suppose two people are asked to rank a list of only two items. Clearly, quite by chance, they will rank them the same 50% of the time, and their ranking the lists the same is in no way significant. Nevertheless, when we use SPSS, for example, to calculate the rank correlation coefficient, we get the absurd result that the identity of ranking is significant at the .01 level:


Last updated November 1, 2000.
This page maintained by Prof. Tim Craven
E-mail (text/plain only):
Faculty of Information and Media Studies
University of Western Ontario,
London, Ontario
Canada, N6A 5B7