EXTENDING THE FRAGILITY APPROACH FOR ASSESSING STATISTICALLY SIGNIFICANT TESTS INVOLVING COUNTS AND PROPORTIONS

The Fragility Index, proposed by Walsh, et al., is a metric defined as the minimum number of subjects within the smallest group whose status is needed to change from non-event to event, in order for a significant test of... [ view full abstract ]

The Fragility Index, proposed by Walsh, et al., is a metric defined as the minimum number of subjects within the smallest group whose status is needed to change from non-event to event, in order for a significant test of independence in a randomized controlled trial to lose significance. There is increasing advocacy for the use of the Fragility Index along with the p-value when reporting the results of a significant test. However, the Fragility Index, as computed by Walsh et al., is focused exclusively on the standard treatment-control 2×2 randomized control trial. We propose extending Walsh's approach to fragility, to other common scenarios involving hypothesis testing about counts and proportions. Examples suitable for the Business Statistics class are presented.