Exact tests in different dichotomous data analysis problems
Dibaj, Seyedeh Shiva
MetadataShow full item record
In many studies researchers have to deal with binary outcomes. Applying inferential methods that are based on large sample approximations can give unreliable results, especially if the sample sizes are small. In recent years, with fast improvements in computer power, exact methods have increasingly became popular in the categorical data problems. In these methods, the exact sampling distribution of the test statistic is utilized for all inferential purposes. In this thesis, we propose exact tests for some common dichotomous data analyses problems. First, we develop an exact unconditional multiple testing procedure for all pairwise comparisons between k binomial proportions ( k > 2) which controls the family-wise error rate. The second method deals with the problem of testing homogeneity of proportions when samples are collected in an inverse sampling scheme. Finally, we present an exact unconditional test for the comparison of proportions in samples drawn from finite populations.