You run a mediation analysis in SPSS using PROCESS (Model 4) to test whether sleep quality (M) mediates the relationship between caffeine intake (X) and daytime fatigue (Y). Here is the key part of the PROCESS output:
Indirect effect of X on Y: Effect = 0.18 BootSE = 0.07 BootLLCI = 0.05 BootULCI = 0.33
Direct effect of X on Y (controlling for M): Effect = 0.10, p = .210
Total effect of X on Y: Effect = 0.28, p = .010
Which interpretation is best?
No mediation, because the direct effect is not significant (p = .210).
Mediation is supported because the bootstrap confidence interval for the indirect effect does not include 0; the pattern is consistent with full (indirect-only) mediation.
Mediation is not supported because the total effect is significant, which contradicts mediation.
Mediation is supported only if the Sobel test is significant; bootstrap results are not valid.