Description
Test whether a within-person link from X to Y runs through a same-occasion mediator, letting each person have their own paths and adding their covariance to the indirect effect.
In diary data a cause, mediator and outcome can all vary from day to day within each person, and each person can have their own X-to-M (a) and M-to-Y (b) paths. The average indirect effect is then not just the product of the average a and b: it also includes the covariance between people's a and b paths, which can be a large share of the effect. Scenarios are labelled by level (2-2-1, 2-1-1, 1-1-1); this covers 1-1-1. X, M and Y are all person-mean centred so only within-person variation is analysed. The M and Y equations must be estimated jointly, either by stacking data with one M row and one Y row per occasion and selector dummies, or with multilevel SEM such as Mplus; both give the same estimates. In Bolger and Laurenceau's example the a-b covariance was over half of the mediated effect.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Goal
- Understand why
- Stage
- Discovery, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- State the X to M to Y process and check timing: X and M should precede Y within each occasion.
- Person-mean centre X, M and Y so each carries only within-person variation.
- Either stack the data with one M row and one Y row per occasion and dummies selecting the equation, or use multilevel SEM software.
- Fit the M equation (M on X) and Y equation (Y on M and X) jointly, without intercepts, with random a, b and c-prime slopes and their covariances.
- Allow separate residual variances for the M and Y rows.
- Compute the average indirect effect as a times b plus the a-b covariance, and the total effect as that plus c-prime.
- Test the indirect effect and the a-b covariance with Monte Carlo or bootstrap confidence intervals, and report the share mediated.
Sources
- Intensive Longitudinal Methods (Bolger & Laurenceau, 2013)
Also known as
- Lower-level mediation
- 1-1-1 mediation
- Multilevel mediation with random indirect effects
- Within-person mediation
Categories
- Analytics & Behaviour
- Statistics & Modelling