ABSTRACT
In this chapter, we consider three methods that may be applied to data on a trinomial scale. In some situations, the three scale points constitute an equally spaced ordinal scale. Examples include procedure rates based on paired organs, and gene frequency without the assumption of the HardyWeinberg equilibrium (HWE). The mean of such a variable is equivalent to a paired difference based on trinomial data, accordingly a confidence interval may be derived from one of the recommended methods for paired differences introduced in Chapter 8. However, the regions of the parameter space that are plausible for these contexts are quite different to that for the difference between proportions. When evaluated over appropriate regions of the parameter space, these approaches turn out to have much better performance than methods that analyse the data as if from a Gaussian model, or methods that disregard the paired structure of the data. The chosen method, based on the Tango interval, is readily implemented in MS Excel. Two extensions are discussed, to clinical trials with paired organ related outcomes, and to estimation of the degree of firstmove advantage in chess.
