mediation.analysis.fnc

Target

Performs analysis of "causal" mediation mediation library, following the procedure defined by Baron RM, Kenny DA (1986). "The Moderator-Mediator Variable Distinction in Social Psychological Research: Conceptual, Strategic, and Statistical Considerations." Journal ofPersonality and Social Psychology, 51 (6), 1173-1182. 1

Starting from mediation package jobs database (Job Search Intervention Study).

We want to test a model of mediation to demonstrate that the effect of the independent variable (attending or not subject to a program for the unemployed) on the values of the dependent depress2 (depression values unemployed) treat is mediated by the Variable mediator job_seek (perceived self-efficacy in the search for employment).

The user must create a list (in this example (mediation) with at least 3 elements: dv (the dependent variable on which you want to evaluate the effect), iv (independent variable on which the direct and indirect effect will be evaluated through the mediating variable) and m (mediator variable).

If you want the model also contains other variables these should be included in item 4 of the list with the name of covariate.

If a test of interaction of the independent variable and the mediator is desired it can be included in the function call with vi.m.interaction = T argument.

If a covariate mediation modulated is desired include mod.med argument, which should be a list with the name of the moderator variable ant two values on which you want to test modulation. If no one is included, the user must include NA as unique value. In this case the function will do it on the first and third quartiles of the moderator variable (Q1 vs Q3).

BASIC MEDIATION

mediation=list(dv='depress2',iv='treat',m='job_seek')

mediation.analysis.fnc(jobs,mediation)

Two figures are created automatically. The left presents the estimated two regression models values and estimates for mediation parameters: ACME (Average Mediation Causal Effects), ADE (Average Direct Effect) and the total effect. The figure on the right, offers bootstrap confidence intervals calculated for the three effects. As we can see, all intervals contain the value 0 thus conclude that none of the estimated effects are given.

res_mediacion

MEDIATION WITH TREATMENT MEDIATIOR INTERACTION

mediation=list(dv='depress2',iv='treat',m='job_seek')

mediation.analysis.fnc(jobs,mediation, vi.m.interaction=T)

MEDIATION WITH COVARIATES

mediation=list(dv='depress2',iv='treat',m='job_seek',

covariate=c('age','econ_hard'))

mediation.analysis.fnc(jobs,mediation)

MODULATED MEDIATION WITH AND WITHOUD CUT POINTS

When the investigator suspects the effect of mediation can be modulated by a covariate variable measured before treatment may carry out an analysis of moderate mediation by that variable.

Next we assume that the age variable can lead to different parameter values estimated mediation as the subject is 20 or 60 years. To this list we must include the mod.med argument with the name of the variable and the two values that you want to test modulation. If you do not specify any value (then including the NA value) function will do it on the first and third quartile of the variable (if this is numeric) or on the first two levels if the variable is a factor.

mediation=list(dv='depress2',iv='treat',m='job_seek')