Mplus Discussion >> Sample size multiple vs. single-mediator models

Topics
Last Day
Last 3 Days
Last Week
Tree View

Edit Profile


Sample size multiple vs. single-media...

Mplus Discussion > Structural Equation Modeling >

Message/Author

Sabrina Twilhaar posted on Wednesday, April 10, 2019 - 2:17 am

I tested two mediation models: one with an observed mediator and one with a latent mediator (3 indicators). X=binary, Y=continuous, 130 obs, bootstrap CI. As both mediators are related (r=0.58), it may be better to test a single model with two mediators instead of separate single-mediator models. But is this also a good idea given my limited sample size? Below the results of the different models:

Model A: observed mediator M1
path a: beta=-0.24
path b: beta=-0.11
path c': beta=0.30
indirect effect: estimate=0.23 (95% CI -0.11, 1.01)
direct effect: estimate=2.51 (95% CI 1.22, 4.14)

Model B: latent mediator M2
path a: beta=-0.50
path b: beta=-0.53
path c': beta=0.07
indirect effect: estimate=2.21 (95% CI 0.70, 5.25)
direct effect: estimate=0.55 (95% CI -2.25, 2.31)

Model C: two mediators M1, M2
path a1: beta=-0.50
path b1: beta=-0.68
path a2: beta=-0.25
path b2: beta=0.25
indirect effect M1: estimate=2.85 (95% CI 0.87, 9.73)
indirect effect M2: estimate=-0.52 (95% CI -3.37, 0.08)
total indirect effect: 2.33 (95% CI 0.73, 7.30)
direct effect: 0.43 (95% CI -3.93, 2.37)

How can I decide whether model C suffers from the limited sample size? I see that the CIs for the indirect effects are wider, so the estimates are less precise. What else should I look at?

Sabrina Twilhaar posted on Wednesday, April 10, 2019 - 5:16 am

Sorry, I think I got the answer. I conducted a Monte Carlo simulation study to determine the power of model C. I hope I did it the correct way, but the output shows a % sig coeff of 0.88 for the indirect effect of M1 and 0.07 for M2, which shows that my analysis is not underpowered, right?

- probability to reject null-hypothesis when it is (indeed) false (ind M1) is 0.88
- probability to reject null-hypothesis when it is actually true (ind M2) is 0.07, which I interpret to be low or is that not low?

Bengt O. Muthen posted on Thursday, April 11, 2019 - 10:23 am

The M1 power of .88 is fine. The M2 power of .07 is too low - I don't know why you say the hypothesis for M2 is true; the true indirect effect is not zero.