Comparing three different measures' performance within a model?
I'm comparing how well three different measures of one construct perform within the model I use for my research. Within the model, the construct I am investigating is a moderator variable. The model is comprised of two predictors, two mediators, and one outcome variable, with the moderator affecting the predictor -> mediator path.
Two of the three measures have subscales that I'd also be interested in investigating.
What is the best way to analyze this? I want to preserve as much power as possible and be computationally efficient as I'm working in R.
Join Matchmatician's Affiliate Marketing Program to earn up to 50% commission on every question your affiliated users ask or answer.
- 189 views
- Wiener process probability
- Multiple comparisons problem
- Determining which excel T-Test to use
- Explain how to get the vertical values when $n = 10$, $p = .5$, $\mu = 5$ and $\sigma^2 = 2.5$
- stats- data analysis
- How do I meaningfully display Kruskal Wallis Data when I have a lot of zeroes?
- In each of the situations, state whether the indicated model can be regarded as a Generalized Linear Model (GLM) and give reasons for your answer.
- Intro to stats symmetry vs skew and missing data values questions
The offered bounty is low for the level of the question and the time one needs to spend to write a good answer.
Thanks for the comment! I'm new here and was not familiar with the bounty system. Adjusting now.
Based on your description, it seems like you have a moderated mediation model. One possible approach to analyze this would be to conduct a series of multiple group analyses, where you compare the fit of the model using each of the three measures of the moderator variable.
Specifically, you would create three groups in your data set, with each group representing a different measure of the moderator variable. You would then fit your moderated mediation model separately to each group and compare the fit statistics. I am not sure how much detailed response you are looking for, so I did not reserve your question, but I hope this will help you.
@Kav10, yes, exactly! Moderated mediation is what I figured as well but wasn't too sure about how to compare the measures. Might it also make sense to run an SEM with the moderator as a latent construct comprised of all items of each measure and see which items/measure performs best? I like the comparing group suggestion though--that is one that hasn't been suggested and I'll look more into.
Running an SEM with the moderator as a latent variable ... works too but the use of statistical models also depends on the actual research question you are trying to answer. I would say, try my idea first and then construct the SEM. Let me know if my ideas were helpful.
Will do. Is there a way you can still collect the bounty for this question? I'm open to other ideas that users might have but your response was very helpful.
Yes, you can buy me coffees, each coffee is worth $4. Click on my Username and it will give you the option. I am glad it was helpful. I hope it works too.
Thanks @Kav10 for being patient with the delay. I just tipped. Thanks for your help, will keep you posted (still collecting data, but should be complete within the next few days).
Thanks for the tip. Keep me posted.