rmedsem: Statistical mediation analysis for covariance-based, partial least-squares and Bayesian structural equation models

Mediation analysis is one of the most widely used techniques in the social and behavioral sciences. Conducting mediation analysis manually is prone to errors, highlighting the need for accessible and flexible software tools. The open-source and free R framework offers many excellent software packages, most of which support mediation analysis with only observed variables. To address this limitation, we have developed an R package, `rmedsem`, which estimates mediational hypotheses with both observed and latent variables. `rmedsem` is the only software implementing mediation analyses within the frameworks of covariance-based SEM, partial least squares SEM, and Bayesian SEM. The package extracts the necessary coefficients from an estimated SEM model (using `lavaan`, `cSEM`, or `blavaan`) specified with the well-known `lavaan` syntax. It then follows a decision tree based on the approaches by Baron and Kenny (1986), and Zhao et al. (2010), to determine the presence and type of mediation. One limitation of `rmedsem` is its current restriction to linear mediation models.