When researchers want to test mediation in a pre-test–post-test design, the PROCESS macro is usually the first tool they reach for. But PROCESS was never built for designs in which the same participant is measured twice. This post shows exactly where it breaks down, how MEMORE fills the gap, and how to read the output you get back.

Training on mediation analysis for repeated-measures designs using the MEMORE macro
In two-instance within-participant designs, mediation rests on the logic of difference scores.

Why PROCESS does not fit

PROCESS is a path-analysis tool developed for between-subjects designs, where observations are independent of one another. Once you collect data from the same person under two conditions — or at two time points — that independence assumption no longer holds: the two observations share everything about the individual. Your independent variable is no longer something that varies between people but a condition that varies within each person. There is no correct way to hand such a structure to PROCESS; the macro specifies the model over independent observations and computes standard errors accordingly.

Three common workarounds that do not work

  • Stacking the data into long format: presenting two measurements from 100 people as 200 independent rows inflates the degrees of freedom and shrinks the standard errors.
  • Using a change score as Y and the post-test mediator as M: the model will run, but the effect you have specified is not the effect you think you are estimating.
  • Modelling mediation on the post-test alone: ignoring individual differences at pre-test reduces your design to a cross-sectional one.

How MEMORE approaches the problem

MEMORE (MEdiation and MOderation analysis for REpeated measures designs) is a free macro for SPSS and SAS, developed by Montoya and Hayes for two-instance within-participant designs. It builds on the difference-score logic set out by Judd, Kenny and McClelland (2001). Here the total effect is simply the mean difference in Y between the two conditions — the same quantity a paired-samples t test gives you. The mediator, meanwhile, is decomposed into two components: the difference between conditions (M₂ − M₁) and the person’s overall level (the average of the two measurements). That decomposition is what allows the indirect effect to be estimated without violating the within-participant structure.

When MEMORE is the right choice

  • You measured the same participants under two conditions or at two time points
  • The mediator was also measured on each of those two occasions
  • You want to explain the effect of an experimental manipulation through a mechanism
  • You want to test whether a within-participant effect depends on a between-person variable (moderation)

Reading the output

MEMORE first reports the total effect; comparing it with your paired-samples t test is a useful consistency check. Next come the indirect effect and its bootstrap confidence interval. Your decision rule is the interval, not a p value: if it excludes zero, the indirect effect is supported. The direct effect shows what remains of the difference once the mediator is accounted for. Because the sampling distribution of an indirect effect is skewed, approaches that assume normality — such as the Sobel test — are no more appropriate here than in between-subjects models.

A note on reproducibility

  • Keep the number of bootstrap resamples large (5,000, for example) and state it in your write-up.
  • Record the random-number seed. Without it, rerunning the same analysis will not reproduce the interval limits exactly.
  • Report how many participants were dropped because of missing data.

Where MEMORE stops

MEMORE is designed for two measurement occasions. With three or more time points you need to move to multilevel modelling or structural equation approaches such as latent growth models. The macro excludes participants with missing values, so it is worth inspecting your missingness pattern beforehand. And MEMORE cannot repair a design: counterbalancing the order of conditions, guarding against carry-over effects and choosing the interval between measurements are all decisions made before any analysis begins.

  1. Pin down the designHow many measurement occasions do you have? Two means MEMORE; more means multilevel or latent growth modelling.
  2. Prepare the data in wide formatOne row per participant, with Y₁, Y₂, M₁ and M₂ as separate columns.
  3. Install the macroAdd MEMORE to SPSS as a custom dialogue, or call it from syntax.
  4. Estimate with bootstrappingFix the number of resamples and the seed, and keep the output.
  5. Report against the intervalGive the total, direct and indirect effects together with their confidence intervals.

Repeated Measures Analysis with MEMORE

Learn mediation and moderation analysis for experimental and pre-test–post-test designs, hands on. Work with your own data set and see how to report the output in publication-ready language.

Course details →

References

  1. Montoya, A. K., & Hayes, A. F. (2017). Two-condition within-participant statistical mediation analysis: A path-analytic framework. Psychological Methods, 22(1), 6–27. https://doi.org/10.1037/met0000086
  2. Judd, C. M., Kenny, D. A., & McClelland, G. H. (2001). Estimating and testing mediation and moderation in within-subject designs. Psychological Methods, 6(2), 115–134. https://doi.org/10.1037/1082-989X.6.2.115