Structural Equation Modelling for a Moderation and Mediation Hypothesis.

This forum made possible through the generous support of SDN members, donors, and sponsors. Thank you.

jazminrice

New Member
Joined
Mar 6, 2024
Messages
2
Reaction score
2
Hello,

I have concluded data collection for my first study and I am stuck with data collection at the moment.
I am looking into the working mechanisms of mindfulness and have explored two hypothesis:



1. Moderation Hypothesis: Mindfulness moderates the relationship between stress and work place outcomes.
2. Mediation Hypothesis: The relationship between mindfulness and workplace outcomes is mediated by psychological capital.



1709772469181.png


1709772475629.png



My measured workplace outcomes are: burnout, turnover intentions, job satisfaction and work engagement.

I have run moderator and mediation analysis in SPSS using Hayes process macro.
10 models in total, a mediation and a moderation for each of the 5 outcomes


I have found significant results and no issues came up.
Now my supervisor suggested trying out SEM to put all variables into one model.
I am a bit hesitant about this - mainly because of its difficulty and the fact that I am running out of time with my publications.

What are your opinions on this? Is SEM necessary or can I stick with two separate analyses - since I am testing two separate hypotheses, I thought it would make sense..

Any comments, opinions or suggestions will be highly appreciated!

Thank you!

Members don't see this ad.
 
I would wonder if I had the power to test such a model and whether it makes theoretical sense to do so. In my mind, I think it would be fine to keep them separate since it seems you're not explicitly testing a full theoretical model.
 
  • Like
Reactions: 1 user
Members don't see this ad :)
Yes, from what I remember about PROCESS, I think that methodology should suffice
 
It sounds like your supervisor is making two requests simultaneously, either of which could make sense. If you have the theoretical rationale and statistical power to run the combined model, you could choose to do that. However, whether or not to run an SEM is separate from this decision, as it describes how you define your variables rather than how your constructs are connected. For SEM, you are using your data to create latent variables, which are basically an approximation of your construct with as much individual variability taken out as possible. You can do this using different measures that fit within the same construct definition (e.g., your workplace outcomes) or by creating parcels with items from a specific study measure.

The bigger question that I'd be asking is whether you - or someone in your lab - has experience with running these types of models. Best practice is to run a CFA with all of your latent variables before running an SEM, and from my understanding you can't run a CFA in SPSS (I might be wrong... I primarily use R). Would likely help to clarify with your supervisor what their goals are for the additional analyses.
 
It sounds like your supervisor is making two requests simultaneously, either of which could make sense. If you have the theoretical rationale and statistical power to run the combined model, you could choose to do that. However, whether or not to run an SEM is separate from this decision, as it describes how you define your variables rather than how your constructs are connected. For SEM, you are using your data to create latent variables, which are basically an approximation of your construct with as much individual variability taken out as possible. You can do this using different measures that fit within the same construct definition (e.g., your workplace outcomes) or by creating parcels with items from a specific study measure.

This is generally what is meant when reviewers say 'theoretical rationale'. People should have hypotheses about relationships of variables before specifying them in a path model. Also, 'SEM' is a broad term that encompasses measurement models and structural models (i.e., technically a path model is 'SEM').
 
  • Like
  • Love
Reactions: 1 users
This is generally what is meant when reviewers say 'theoretical rationale'. People should have hypotheses about relationships of variables before specifying them in a path model. Also, 'SEM' is a broad term that encompasses measurement models and structural models (i.e., technically a path model is 'SEM').
Definitely agree with your definition of theoretical rationale. My understanding was that SEM = model with latent variables. Is this incorrect? What would fall within SEM?
 
Definitely agree with your definition of theoretical rationale. My understanding was that SEM = model with latent variables. Is this incorrect? What would fall within SEM?

Yeah, it's super confusing. That is 'SEM' (i.e., structural model with latent variables), but some folks encompass testing any structural model as 'SEM' (e.g., Rex Kline). Also, if you think of H. Marsh's ESEM, for instance, it's explicitly testing a measurement model, but technically also imposing a structural model given that there are relationship between observed (i.e., the items) and latent variables.
 
  • Like
Reactions: 2 users
Theoretical rationale aside, SEM allows you to test all your hypotheses at once, forgoing the need to run multiple tests which increases the likelihood of Type I error. It also allows you to see if there are issues within either model that might affect independent interpretations. You note that you are hesitant to do this, because it's difficult. That's, honestly, not a great reason. Finally... presumably your supervisor has a PhD, and they're your supervisor for a reason. They've likely wrestled with things like this in the past... I'd trust, and learn, from them :)
 
  • Like
Reactions: 1 users
I started with PROCESS in SPSS for moderation and mediation. I eventually switched to SEM with mplus. SEM is just way more fun and less restrictive. Switching over is not too hard either imo. Pm me if you need help getting started.
 
This is generally what is meant when reviewers say 'theoretical rationale'. People should have hypotheses about relationships of variables before specifying them in a path model. Also, 'SEM' is a broad term that encompasses measurement models and structural models (i.e., technically a path model is 'SEM').
I wish more people understood this spirit, this principle.

"There is one thing even more vital to science than intelligent methods; and that is, the sincere desire to find out the truth, whatever it may be."

- Charles Sanders Pierce
 
  • Like
Reactions: 2 users
Thank you everyone for your input. I have discussed with by supervisors and we all agreed to go for SEM so I am trying to teach myself how to use amos at the moment. Wish me luck :D
 
  • Like
Reactions: 1 users
Best of luck to you.



Whatever you do, get out of Louisiana.
 
Last edited:
If I do get to Heaven, I hope to Hell that Justin Wilson welcomes me there.

 
Top