Are there any defined criteria, on the basis of which it can be decided how the research model can be broken into multiple studies in the same research paper? Please back up your answers with examples if possible.
It would help me if you could explain why you want to do this. I encourage my students to think about developing an encompassing quantitative model that includes alternatives as a special case And then the publishable paper evaluates them, maybe finding some can be ruled out, but that there are more than one plausible model remains. So what drives this is different theoretical accounts and their potentially different explanations,; I really struggle with your criteria requirement, but probably I do not understand what you wish for.
Kelvyn Jones, in this case, the situation I am facing is a quantitative model, which 5 IVs, 1 intermediate variable, and 3 DVs. Apart from these direct relationships, there is one moderator between 5 IVs and intermediate variable, and another moderator between intermediate and DVs. However, in the latter case the moderator has multiple levels (e.g. positive, negative). This created a dilemma for me whether this part of the model would need to be tested using experimental design, and rest using SEM. Pertaining to this I had asked another question too: https://www.researchgate.net/post/Is_it_possible_to_do_experiment_in_a_condition_where_IV_has_one_level_while_moderator_has_multiple_levels_2_and_three_DVs?_ec=topicPostOverviewAuthoredQuestions&_sg=WG7VfcScJIqeYQLn3zXklwh3qj8Uw0BqdVy6zCNqzWIOEps4TOHsdL8QloA6MJcTl6wNInylFhxss66Q This was the primary reason I asked this question, to understand the criteria of breaking the model into multiple studies.
I understand more now but not why some is SEM and other is experimental. My advice would be to simplify the model; certainly to start with. For example; is there are great deal to be gained by analyzing all three DVs simultaneously in a multivariate model. The gains are the correlations before and after conditioning on prior variables and the ability to test the comparative, differential effects of the variables in an overall model, but this comes at a very large increase in model complexity. I word sort out an estimable model for each DV before putting it all together. You would then be able to do moderation mediation etc for each outcome in a set of regression models http://afhayes.com/introduction-to-mediation-moderation-and-conditional-process-analysis.html
Also, Judea Pearl has been urging us to use graphical models to ascertain what variables need to be 'controlled' - the latest account is this book JUDEA PEARL AND DANA MACKENZIE THE BOOK OF WHY: THE NEW SCIENCE OF CAUSE AND EFFECT http://bayes.cs.ucla.edu/WHY/ which may help you think through what you want.
I'm sending an article by Onwuegbuzie and Leech (2006). It addresses mixed methods research designs, which it sounds like you're trying to construct. Hope it helps!
Kelvyn Jones I read your answer closely again today. If I have understood correctly, and let me paraphrase my thoughts, what you suggested is to take one DV per study and ascertain IVs which are a better fit for that particular DV. This will make the studies simpler and mediation, moderation analysis more manageable. Have I understood your thoughts correctly?
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers