ABSTRACT
This chapter provides an overview of current applications of computer intensive methods for mediation analysis. David P. MacKinnon, C. M. Lockwood, and J. Williams compared a large number of single sample and resampling methods to assess the mediated effect and found several bootstrap methods had more accurate confidence limits than the single sample methods. Resampling methods, especially randomization tests, may handle small samples better than alternative tests and may provide more accurate results than traditional tests in this situation. Tests based on a random sample of permuted data sets are called approximate randomization tests, and the permutation test is sometimes called an exact randomization test because it is based on all possible data sets and is therefore exact. Randomization tests for the mediated effect are complicated because two separate equations are involved in the analysis of three variables. As a result, the number of potential data sets is even larger than that for the correlation coefficient.
