Fully Mediated Model Of Mediation

Mediation is a hypothesized causal chain in which one variable affects a second variable that, in turn, affects a third variable (Baron & Kenny, 1986). The intervening variable, M, is the mediator. This third variable (M) affects the strength or influence of the relationship between the independent and dependent variables. The independent variable has a path (a) to the mediator. The mediator has a path (b) to the dependent variable (outcome). In a partially mediated model there can also be a (c) path which leads from the independent variable to the dependent variable (outcome). There are two types of mediation, fully mediated and partially mediated. A fully mediated model is one in which there is not a direct path from the independent variable

In a partially mediated model there is a direct path from the independent variable to the dependent variable (Bullock & Ha, 2011). Mediation first took off when Baron and Kenny published their manuscript in 1986. Since then it has been cited 72,357 times (2017). Titled “The Moderator-Mediator Distinction,” they set out to conceptually define what these variables were and how could be applied to social psychological research. Although mediation has been around before Baron and Kenny’s paper, many researchers today refer to the “Baron and Kenny” method when approaching a potential mediation model. As suggested, there are three regression equations to test for mediation, and these three factors must be true. First, the independent variable must affect the mediator. Second, the independent variable must affect the dependent variable (Baron & Kenny, 1986); others have suggested this second assumption can be violated, this will be examined later on. Third, the mediator must affect the dependent variable (Baron & Kenny, 1986). It is then a perfect mediation when the independent variable has no effect on the dependent variable when the mediator is controlled. When using multiple

Past analysis have shown that word learning gains were only stronger for English only students that despite summer loss in the treatment and control groups the program still had an effect. The quasi-experimental study showed small but significant effects (Lawrence, Crossen, Pare-Blagoev, & Snow, 2015). Within this study 28 schools were part of the randomized trail. Treatment was specified as participating in the word generation program. Again, we see in this particular study that the school itself was randomized in the treatment group rather than than the classrooms within schools. Note this is not a concern, but just a limitation. The effect sizes were as followed; Math d=1.13, Science d=.47, Social Studies d=.38, English d=.44, averaged together with an overall effect size of d=.62 (Lawrence et al., 2015). The main concern here is within the effect size for English which is the specific area they are testing with the Word Generation program and it is reported with an effect size of d=.44. Looking at the difference of means between the pre-test and the post-test vocabulary was reported having an average pre test of 18.57 and an average post test of 19.89, the difference is .71 (Lawrence et al., 2015). The effect size calculated from raw scores is .17. Traditionally in psychological research we would like to effect sizes ranging from .20=small, .50=medium, and .80=large (Cohen, 1992). According to Lipsey (1998)