Hello everyone, I am conducting a mediation analysis in JASP software. First, I checked for a general correlation with the IV-mediator correlation being significant at p =.015. Then, when I ran the mediation analysis, the path coefficient between IV and the mediator showed a significance of p = .011. Does anyone know why it's different?
I don't know if it matters but direct and total effects were significant and indirect effect was insignificant. Thank you for your help!
Possible reasons for the (slight) differences in the p-values could be:
- differences in the number of cases (sample size) between your bivariate correlation analysis and your path (mediation) model and/or the way missing data is handled in the two types of analyses (e.g., listwise deletion vs. FIML)
- the structure of the path model. If the model is overidentified [df > 0], it may not reproduce the observed data perfectly, leading to differences between the observed and model-implied data structures. A correlation analysis, in contrast, typically implies a just identified (saturated) "model" (df = 0).
- the type of correlation coefficient and/or significance test used to test the correlation coefficient.
- differences in the test of significance for an unstandardized path coefficient vs. a (by definition) standardized bivariate correlation coefficient. (For a saturated path model that is estimated with maximum likelihood estimation, the standardized bivariate path should equal the bivariate Pearson correlation unless there are additional predictors of M in the model.)
- potential inclusion of covariates (or, more generally, additional predictors of the mediator M in addition to the IV) that would lead to a multiple (rather than bivariate) regression for M.
- Badges
- Science topic
- Similar topics
- Filing