Estimating Sample Sizes for Ordinal Data

(Letter to the Editor concerning » BMJ 311 (1995), 1145-1148 «)

R. BENDER
Klinik für Stoffwechselkrankheiten und Ernährung
Heinrich-Heine-Universität Düsseldorf
12.12.1995



Dr Campbell and colleagues outline some ways of calculating sample sizes in two group comparisons for binary, ordinal, and continuous outcomes [1]. While the presented formulas for binary and continuous data are well-known and have been repeatedly reviewed in the statistical [2,3] as well as in the medical literature, [4,5] the proposed proceeding to estimate sample sizes for the Wilcoxon-Mann-Whitney U test for ordinal response seems to be the only new issue of the paper. However, the proposed formula, which uses an odds ratio as effect size, is quite complicated. For application a number of steps are required, which are circumstantial and troublesome in practice. In this note I want to point out that the estimation of sample sizes for ordinal data can be much easier.

Let X and Y be the ordinal outcomes in the two groups, then the probability p=P(X<Y) that the response in one group is smaller than in the other contains similar information as the standardised difference of means for continuous data [6]. Hence, it is obvious to use p as effect size. A corresponding formula is given by Noether [7]. The probability p can either be chosen arbitrarily similar to the clinically relevant difference in the continuous case or can be estimated from a pilot sample by means of the Mann-Whitney statistic U [6]. For the data shown in table V one can calculate U=317 and p=317/(21×22)=0.686. Using Noether's formula [2,7] the sample size is m=38 patients per group. The use of an odds ratio as effect size is possible due to the simple relation OR=p/(1-p), which yields p=OR/(1+OR) [7]. This proceeding avoids the fussy and complicated steps proposed by Dr Campbell and colleagues and makes use of already available formulas.


References

  1. Campbell MJ, Julious SA, Altman DG. Estimating sample sizes for binary, ordered categorical, and continuous outcomes in two group comparisons. BMJ 1995; 311: 1145-1148.
  2. Bock J, Toutenbourg H. Sample size determination in clinical research. In: Rao CR, Chakraborty R, eds. Handbook of Statistics, Vol 8. Amsterdam: Elsevier, 1991: 515-538.
  3. Woodward M. Formulae for sample size, power and minimum detectable relative risk in medical studies. Statistician 1992; 41: 185-196.
  4. Florey C du V. Sample size for beginners. BMJ 1993; 306: 1181-1184.
  5. Lachin JM. Introduction to sample size determination and power analysis for clinical trials. Contr Clin Trials 1981; 2: 93-114.
  6. Wolfe DA, Hogg RV. On constructing statistics and reporting data. Am Statistician 1971; 25: 27-30.
  7. Noether GE. Sample size determination for some common nonparametric tests. J Am Stat Ass 1987; 82: 645-647.