Does anyone know how to calculate the t-value in intermediate variable ,especially in structural equation modeling,for example when we want know.is a relationship between A and C by intermediate B variable. especially when a one of them is not significant.
Hamid, I guess that you want to estimate the indirect effect? In order to estimate the significance of the indirect effect (product of two path coefficients), you need to calculate the standard error of this product and then devide the product by its standard error. One possibility is to use the Sobel test (for large sample sizes), another is bootstrapping.
Kristopher Preacher provides all information you need on his website:
http://quantpsy.org/sobel/sobel.htm
If one the the coefficients included in the indirect effect is zero, then the product is zero, too, and therefore you would not need to test the significance of this product.
Some more information is included in this article, which you will find on Kristopher Preacher's website:
Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods, 40, 879-891.
HTH, Karin
Dear Karin
thanks for your answer,I put on a model in the attachment, according to this model ? I want to know :relationship between the TAB.ETEL variable and TAMOL.AY variable is signification?and as you can see that the first half of model is not significant ,and generally how we can calculate the mediate variable significant?
Hamid,
there are 3 indirect effects of TAB.ETEL on TAMOL.AY: one mediated by ETAMAT, another mediated by REZEYAD and the third mediated by TAAHOD. The sum of these indirect effects is the total indirect effect of TAB.ETEL on TAMOL.AY. There is only one insignificant path, so one of the indirect paths will not be significant, but the other two may be.
Are you using LISREL? From the type of graph I guess that this is so. If yes, LISREL can estimate all separate indirect effects as well as the total indirect effect. Just include the command EF in our output line:
OU EF
EF invokes the printing of the estimated total and indirect effects in the output file.
Another problem seems to be that the model does not fit the data too well. Check the modification indices, maybe you forgot to include a direct path in your model or the covariances betwee the residuals of the mediator variables.
Regards,
Karin
Dear Karin
Thanks for your useful guidance,Yes I used LISREL soft. you say that LISREL can be estimate all separate indirect effects and total indirect effect.according to the LISREL output attached file could your more explain about indirect effects?
Hamid, this is quite a small model, which consists of one direct effect and only one indirect effect. But anyway:
First, you need standardized parameter estimates, not unstandardized estimates as you have. Then, the direct effect of YADGH to EP is the direct path from YADGH to EP (standardized: -.04, n.s.). The indirect effect is the product of the parameter estimates of the paths YADGH to ESTRA, and ESTRA to EP (standardized: .47*.39).
LISREL calculates the significance of the indirect effect only for unstandardized parameters (marked in blue). This should be the same for the standardized effect. With a t-value of 3.74, this indirect effect is significant.
Enclosed is a LISREL output with the unstandardized parameter estimates, the standardized parameter estimates, and all direct and indirect effects. The relevant parts in the output are marked in yellow and green. You will find the path model with standardized parameter estimates below the output.
Any further questions?
Kind regards, Karin
Dear prof Engel
Hi
Thanks for your useful guide ,professor Engle I have again a question
What will happen, when a model's fitting parameters are not fitting enough?
What will happen, when the model's fitting indicators , have not fitting enough or the model in some indexes have a good fitting (for example in NNFI,CFI,RFI,.....)and in other some have not,and this condition what should we do ?
Hamid, if the model does not fit the data, the model structure is not correct. Check the modification indices: Did you forgot to include a path which you fixed to zero in your model? Check also the direction of your paths, are they correct? If the model does not fit, you cannot interpret the parameter estimates as they may be biased.
Regards, Karin
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling