Standardized vs. Unstandardized Path ... PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Chantal Hermann posted on Friday, April 03, 2015 - 8:33 am
Hello,

I have a question that I am hoping you can please help me with. I have a sample of 248 participants, all continuous variables. I have estimated the following model.
MODEL:

IAT2 ON IAT SESAO1 PASTO1;
SESA2O1 ON IAT SESAO1 PASTO1;
PAST2O1 ON IAT SESAO1 PASTO1;
IAT WITH SESAO1;
IAT WITH PASTO1;
SESAO1 WITH PASTO1;
IAT2 WITH SESA2O1;
IAT2 WITH PAST2O1;
SESA2O1 WITH PAST2O1

*I hypothesized that IAT and SESAO1 would significantly predict PAST2O1.

For the undstandardized results:
Estimate S.E. Est./S.E. p
PAST2O1 ON
IAT 0.642 0.334 1.920 0.055
SESAO1 2.400* 0.961 2.497 0.013
PASTO1 0.102 0.059 1.726 0.084

For the Standardized results:
Estimate S.E. Est./S.E. p
PAST2O1 ON
IAT 0.110* 0.051 2.129 0.033
SESAO1 0.246* 0.093 2.660 0.008
PASTO1 0.220 0.114 1.919 0.055

The confidence intervals for the IAT:
Unstandardized effect: Lower 2.5 = -0.013, upper 2.5 = 1.298
Standardized effect: Lower 2.5 = 0.009, upper 2.5 = 0.210

Should I be interpreting this as a significant or non-significant effect?

Thanks,
Chantal
 Bengt O. Muthen posted on Friday, April 03, 2015 - 11:49 am
You can probably trust the CI for the unstandardized coefficient better because the standardized coefficient distribution may be skewed so that the usual symmetric CI is not suitable. See our FAQ:

Standardized coefficient can have different significance than unstandardized

Actually it would be interesting to me to see how this comes out in your example, so you can send us the input and data and we can look at it.
 Chantal Hermann posted on Saturday, April 04, 2015 - 6:50 am
Hi Bengt,

Thank you for the quick response! Yes I will send my data to you to check out.

Thanks,
Chantal
 Bengt O. Muthen posted on Monday, April 06, 2015 - 5:13 pm
A general comment might be useful. These discrepant significances for unstandardized and standardized coefficients seem to happen when a linear model is not suitable in the first place. Examples include a DV with a strong floor or ceiling effect and a DV that is multimodal.
 Filipa Alexandra da Costa Rico Cala posted on Wednesday, June 14, 2017 - 4:31 am
Dear Dr. Muthen,

For my PhD research, I am running a model composed by 10 latent variables, in which I want to estimate the direct effect of 9 independent variables to my dependent variable, and also want to estimate one mediation effect. For estimating this mediation effect, I run the bootstrap. My dependent variable is categorical (and has also a skewed distribution), so I used the WLMVS estimator. Therefore, for writing the paper, should I use the standardised or the unstandardized results for reporting the direct effects from all independent variables to the outcome variable, and also for reporting the indirect effect?
Many thanks in advance for your help
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: