Dunnett's test is appropriate for multiple comparisons where the means of r new treatments are compared with a control treatment mean. this test controls the experimentwise type I error rate for a nominal significance level α. The limitation in their use is the difficulty of obtaining computationally cumulative distribution function and quantiles of the test statistic, since the possible correlations between treatments have broad amplitude. This paper presents an algorithm for probabilities related to bilateral Dunnett test for balanced data using numerical methods to solve the integrals by Gaussian quadrature. The algorithm showed accurate results when compared with values reported in the literature and with values obtained in three computational programs.
multiple comparison procedures; numerical integration; experimental statistic
