An evaluation of the Mantel-Fleiss validity criterion for the Mantel-Haenszel statistic

  • Younghae Chung (Dept of Computer Science and Statistics, Dongshin University) ;
  • Charles S. Davis (Dept of Preventive Medicine and Environmental Health, University of Iowa)
  • Published : 1998.04.01


In testing the partial association between two variables after controlling for the S levels of a third factor, the Mantel and Haenszel (1959) statistic is often used. Since the statistic is based on the asymptotic distribution of the sum X of S hypergeometric variates, a guideline for the minimum requirements for the application of the statistic is useful. Mantel and Fleiss (1980) developed a criterion based on the guideline for the Pearson's $X^2$ statistic. The criterion requires the distance from the expected value to the closer bound of X to be at least five. The Mantel-Fleiss (MF) criterion was studied through a simulation using the hypergeometric sampling scheme. The criterion is not satisfactory. The size of statistic exceeded nominal 0.05 level nearly 1/5 of the cases even when the criteion is met. However, the results show that the statistic is much more unstable and conservative when the criterion is not met.



  1. Biometrika v.61 On the small-sample properties of the Mantel-Haenszel test for relative risk Bennett, B.M.;Kaneshiro, C.
  2. Biometrics v.26 On McNemar's test for the 2?2 table and its power function Bennett, B.M.;Underwood, R.E.
  3. Journal of Royal Statistical Society B. v.26 The detection of partial association, I: The 2×2 case Birch, M.W.
  4. Biometrics v.10 Some methods for strengthening the common χ² tests Cochran, W.G.
  5. American Statistician v.33 On the alias method for generating random variables from a discrete distribution Kronmal, R.A.;Peterson, A.V. Jr.
  6. Biometrika v.66 Small sample properties of the Mantel-Haenszel test Li, S.H.;Simon, R.M.;Gart, J.J.
  7. American Journal of Epidemiology v.112 Minimum expected cell size requirements for the Mantel-Haenszel one-degree-of-freedom chi-square test and a related rapid procedure Mantel, N.;Fleiss, J.L.
  8. Journal of National Cancer Institute v.22 Statistical aspects of the analysis of data from retrospective studies of disease Mantel, N.;Haenszel, W.
  9. Statistics and Probability Letters v.8 Toward a universal random number generator Marsaglia, G.;Zaman, A.;Tsang, W.W.
  10. Psychometrika v.12 Note on the sampling error of the difference between correlated proportions or percentages McNemar, Q.
  11. Simulation and Computing v.17 A Monte Carlo study of three odds ratio estimators and four tests of association in several 2×2 tables when the data are sparse, Communications in Statistics O'Gorman, T.W.;Woolson, R.F.;Jones, M.P.;Lemke, J.H.
  12. Fundamentals of Biostatistics (2nd Ed.) Rosner, B.A.
  13. Journal of Royal Statistical Society A v.145 A comparison of alternative tests for the 2×2 comparative trial Upton