Dear Researchers,
A colleague of mine is working on a structural model (using SmartPLS) that has 3 IVs and 2 DVs and one mediating variable. My question is:
1. Is it appropriate to add/sum the 3 IVs and 2 DVs and do a mediation analysis ? or,
2. Is it required to take one IV and one DV at a time and do the mediation analysis? In this case there will be 6 mediation models to test. But it would not test the entire model then. And individual mediation is no guarantee of collective mediation and vice-versa.
3. If mediating variable contains two dimensions (which has been operationally defined) then what is the statistical criteria of treating the sum of two variables as a mediating variable? Is it necessary that mediating variable should be unidimensional or single construct?
4. Is there any other best practice for this type of problem?
Thanks
Sanjay Singh
If you are using smartpls, it does the whole job at once, you may not need to do that multiple times or in stages. The path-model is not arbitrarily defined, this is based on in-depth literature review, and you can draw the whole model in smartpls (both version 2 and 3) at once. Mediation may have more than one dimension and can be formative or reflective construct. For details, you may consult the suggested book that explains the recommended practices:
https://us.sagepub.com/en-us/nam/a-primer-on-partial-least-squares-structural-equation-modeling-pls-sem/book237345
Hello Sanjay,
Using Mplus software you could specify all paths in one go, you would not have to do this one by one. I would recommend using Mplus at some stage of your research. Of course it is not free...
Good luck with your work
There are two ways to approach this question - mathematically and substantively. The mathematical answer would best be addressed with Mplus, so I focus on the substantive answer, which has at its center a question you do not answer: What is the research question?
If I have three IV that I am equally concerned with (say, family income, gender, and ethnic identity), and one potential mediating measure (say, reading ability) for two DV (say, college test scores and engagement with school), I can conceive of a situation in which ALL I care about is the mediation of gender, holding family income and ethnic identity as covariates. I can also conceive of a situation in which I care about the mediation of all three IV but do not have an interest in modeling the intersection of those three IV. I can also conceive of a situation in which I need to know about the intersections (but it would probably then become a moderation analysis, which has different mathematical issues than simple mediation).
I am a strong advocate of forming the model to address the research question, and choosing the mathematical structure that best fits that model. If you are driven by the math, your results may not make any sense to your question.
IMHO
If you are using smartpls, it does the whole job at once, you may not need to do that multiple times or in stages. The path-model is not arbitrarily defined, this is based on in-depth literature review, and you can draw the whole model in smartpls (both version 2 and 3) at once. Mediation may have more than one dimension and can be formative or reflective construct. For details, you may consult the suggested book that explains the recommended practices:
https://us.sagepub.com/en-us/nam/a-primer-on-partial-least-squares-structural-equation-modeling-pls-sem/book237345
You can perform several mediation tests simultaneously in SEM or Pls-SEM. Just convert your framework into the graphics, input data and execute.
Kindly find mediation analysis in attached file with two predictors,
Article Making Sense of Mediating Analysis: A Marketing Perspective
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling