I was conducting cross-lagged path analysis in MPlus, testing a double moderated mediation model type:
W V X --> M --> Y
I have a large sample of around 1000 participants who participated over three waves. The outcome variable is very low frequent, highly skewed variable. When first tested, the model fitted the data well and most of the effects were significant. However, when i repeated the analyses with bootstrapping, suddenly almost all effects lost significance, even the well-established relationships in the model. The total indirect effects also became insignificant (with BCBootstrap CI). Does this mean that the model really does not hold at all or is there something else going on? Are there certain measures that i need to take given the outcome variable is nonnormal and has a very low variability? Thank you!