Hi everyone,
I've run an analysis of a latent 3-way interaction in Mplus (Y on XWZ).
The estimate for this path is 0.000.
S.E. is 0.000.
Est./S.E. is 2.034.
Two tailed p-value is 0.042.
So, yay! Significant interaction, right? But with a slope of 0.000, how can this be significant? Could I possibly have perfectly symmetric effects at +1SD and -1SD? How can this be?
Thanks in advance :)
Hello Chris,
The answer is that the values for both the coefficient and its standard error round to ".000" when only three decimal places are used. For example, the value 0.000479 rounds to 0.000 when only three decimal places are used. How the two variables that are being linked by the coefficient are scaled has everything to do with the resulting value being so small (when given in unstandardized form). For example, the unstandardized regression coefficient for predicting weight from height for adult males might be 0.4, if wt = kg and ht = m. However, change the scale for ht to km, and the regression coefficient changes to 0.0004 (which would round to zero with three decimal places).
The fact that the test statistic (coefficient / se), which should behave as a t-statistic, exceeds 2.0 is the clue that the coefficient may be considered to be different from zero (at the .05 level of significance).
Good luck with your work.
Hi Chris,
I fully agree with David's answer. It might help if you standardize your variables in the Define option. You might then receive some results that can better be interpreted.
All the best,
Marcel
You should add the model graph for latent interaction models that include the effects of the independent variable, the moderating variable, and another interaction variable from the two previous ones.
Personally I would not use z scores are you are then losing the natural observed metric of the data; notice that David Morse changed metres to kms -so trying dividing the X by 10, 100, 1000 or something appropriate and convenient.
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running