Statistical Analysis of Mediation Variables: A Review

Many statistical methods have been developed to detect the impact of mediation variables. The most classic means of testing for mediation is the procedures proposed by Baron and Kenny (1986), in which three regression models were employed. More recently, structural equations modeling is considered the more appropriated method. Other related methods include path analysis, latent variable models, longitudinal models, multilevel models, and computer intensive method. Related issues include mediation in categorical data analysis, meta analysis and causal inference for mediation models etc. After providing an overall view on the statistical methods that can be used for mediation analysis, a review of application software will be provided. Finally, a few examples with specific, step-by-step applications will be illustrated.