I have been conducting a mediation analysis with 7 major variables represented in the attached file.
Firstly I have run the model in AMOS in which it showed poor model fit.
secondly, I thought of SPSS, but there are so many variables, so it has been difficult to conduct analysis on the model.
Then thought of Hayes PROCESS Macro, but which model should we select, it's not clear.
Can anyone help me to proceed with statistical analysis?
Looking forward to valuable suggestions.
@Abhishek Shukla As I see in your model, you have two serial mediation and one simple mediation models. I not not aware whether you tested the model in AMOS using Structural Equation Model (taking latent variables) or Path Analysis (taking composite variables) but AMOS should have worked if your constructs are correlated enough with each other. Anyway, I still suggest to test your model as a whole and try it on Smart-PLS as it works on PLS approach and extracts more of the variance. In Process Macro, you would have to work on each mediation relationship separately hence would result in higher effect sizes without taking suppression effect of other variables into account. Still, if you wish to proceed with process, use model 1 for simple mediation and model 6 for serial mediation (with two or more mediators). Below is the attached template for Process Models.
Imran Anwar Thank you so much for the suggestion. I have conducted the analysis in AMOS through path analysis, but the model doesn't fit well, therefore, looking for some other way to conduct the analysis. I will try with smart-PLS....Thank you for the help
I have used SMART-PLS but it's showing n/a for estimates and in path, it's showing o.oo. as I have attached the image.
Kindly suggest..why it's happening or what we can do best for it.
Dear Abhishek,
1. The model shows that there is only a mediation variable and you don't have "multiple mediation", which I thought your question is about serial mediation So this model can be run in AMOS.
2. For poor model fit in AMOS, you could look to the modification indices for the covariances between errors, or figure out which item or items are causing the trouble i.e remove that items with poor loadings if possible.
3. In addition, for testing your hypotheses it is possible to use SPSS and you can apply Sobel test for all mediation paths.
4. It seems the items of "TOT-CCG" and "TOT-DEP" are combining the different scale, variables/dimensions which might be the reason for PLS error. If I am right then run the (TOT-CCG) and (TOT-DEP) as second order factor. However, Haywood cases (negative variances), high missing data, High multicollinearity, smaller or larger sample sizes etc. are other reasons for getting n/a as a result for parameters.
5. Most importantly, I think your conceptual model ( word file ) and PLS model (figure above) are different.
Best,
This model is very complex. If it doesn't run well, it's very difficult to find out why. Which is why you'd better start by getting the latent factors straight. Just fit one or a few related factors at a time and see if the indicators load well. You may need some adjustment there or correlations of error terms.
Another source of error can be very different scales (like one variable on a 0-1 scale and another on a 0-10000 scale). Transforming items to roughly the same scale helps. Once you have small pieces running, then slowly expand your model by adding other parts until you have your full model.
I do agree though that the model in the picture above is very different from the earlier conceptual model.
- Badges
- Science topic