Hi RGaters,
Thanks to everyone who contributed to my question on the difference between internal/external rewards and intrinsic/extrinsic rewards.
Are there references on how/ways to determine (empirically/mathematically) the number of moderators to include in a management research? Is there a reasonable limit to the number of moderators to include in a model? I presume there is. How does a researcher objectively determine this limit? Or is the number of moderators included in a model subjectively determined.
For example, there is a paper on control variable use in management research by Antic et al. (2011) which explains best practices for control variable use and reporting. Is there a reference on best practice on moderator variables use? Are there articles recommending best practices on moderator use and determining the number of moderators to include in a model or management research?
This question is not about understanding moderating effects as the answer to this question has already been provided by MacKinnon and colleagues (e.g. MacKinnon, 2011; Fairchild and MacKinnon, 2009), and several other researchers. I need to be able to scientifically support and justify the number moderators I decide to include in my model.
Reference
Atinc, G., Simmering, M. J., & Kroll, M. J. (2011). Control variable use and reporting in macro and micro management research. Organizational Research Methods, 1094428110397773.
I am not sure if there is any but I think the appropriate number of moderators is dependent on your research question or research aim. Perhaps you should justify the importance of assessing the moderators in your study.
I agreed with Lee submission but you may want to use the theoretical significance of the moderators and outcome variable as a guide in building the model. Another suggestion will be to check other past studies on the key variables based on the research questions and objective.
I second Lee and Oyediran answer because there is a determination on your side on what is the aim of your research and whether the literature indicates to you on the need to include or exclude certain variables as a moderating variables. You readings could guide you in further determining that.
Dear Soon, Kolawole and Mohd
Thank you for your contributions.
Best,
Tobi
Key points have been made, but I will put it slightly differently, that the number of predictors should be reflective of the number of competing explanations for the variable(s) of interest, and as such, it is a conceptual or theoretical consideration. But the upper limit of that number will be determined by a power analysis, or more simply, your remaining degrees of freedom. There are formulas used for such analyses, and regular programs available, but traditionally many have been lazy and used just general 'rules of thumb' or heuristic guidelines to determine this (e.g. in multiple regression, it has often been said that it is best to keep the number of predictors to no more than one tenth the number of cases). Also, the higher the level of the data, that whether it is ratio vs. ordinal, you will be able to employ more predictors. But whatever the upper limit, the more parsimonious you can make your model, the better off you will be, as Einstein said, keep your model simple, but no simpler than necessary (I can't remember the exact wording).
I agree with Christopher G. Hudson.
The research question that emanates from a review of the literature should place limits not only on the number of a moderator variables but the most salient moderator variables tested. In any design, there will be numerous extraneous variables that could be built into the design. However, the choice of which to include is based on the theoretical model being tested and those variables identified in the literature that might be potential moderators.
And as Christopher wisely noted, the number of moderators added to a design is a decision to be made in relation to sample size and expected effect size. See Lipsey (1999) Design Sensitivity: Statistical Power for Experimental Design.
I think we are all in agreement that the question of which moderator variables to use is a theoretical issue rather than an empirical one.
Another technical point to consider is when you introduce multiple moderator variables (which are the equivalent of interaction effects) is whether you want to include the higher order interactions between your various moderators. If so, the complexity of the model (and its interpretation) will grow very rapidly.
Thank you David.
I understand that adding and treating moderators could become complex quickly. Moreover, I agree with you that the addition of variables (IVs, DVs, Moderators and Mediators) should be theory driven.
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory