Detecting Differences Between Contrast Groups

In medical research, doctors must evaluate the effectiveness of a new medicine B against a specified disease. This evaluation is often carried out by comparing B with an old medicine A, which has been used to treat the disease for many years. This comparison should include two important statistical summaries: mean and distribution function differences between A and B. The datasets of applied/tested A and B are referred to contrast groups, and the mean and distribution differences are referred to group differences. Because the datasets to be contrasted are only two samples obtained by limited applications or tests on A and B, the differences derived from the datasets are inevitably uncertain. This generates a need of measuring the uncertainty of differences. In this paper, an efficient strategy is designed for identifying confidence intervals for measuring the uncertainty of the differences between two contrast groups. This approach is suitable for most of those applications for which we have no prior knowledge about the underlying distribution of the data. We experimentally evaluate the proposed approach using the UCI, datasets against the bootstrap resampling method and the traditional method, and demonstrate that our method is efficient in measuring the structural differences between contrast groups.