If a theory proposes that x relates to y but there are so many moderating variables, that when considered, the relationship between x and y becomes non-significant, does it make sense to argue that we should change the x variable?
If the answer is yes can someone please reference some papers to support this?
I know there will always be boundary conditions we need to consider, but the theory in its current state has SO many boundary conditions that I think x needs to be tweaked to an extent to eliminate some obvious boundary conditions.
Thanks!
Could you please be more specific about the theory and the empirical situation you describe? Generally, there is no universal limit on the number of arbitrary elements a theory may contain. However, if you can propose a simpler theory which is as good as the old one in predicting empirical outcomes, lex parsimoniae will be on your side.
I'm not sure what you mean by "boundary conditions." You also mention moderating variables, which would involve interaction effects between the X variable and other measures -- is this what you mean by boundary conditions?
Alex - thanks! Thats a great point about parsimony.
David - Yes When I say boundary conditions I mean moderating variables.
I'm writing a critique about a theory that has been published so I am not going to disclose the theory.
Thanks for the help!
Assuming that you begin with an association between X & Y, then this "simple main effect" is not very informative because you don't know what produces the relationship. Introducing other variables that interact with X in its effects on Y is a way to explain that relationship. So, it is possible to consider this to be a more complete specification of the mechanisms that link Y and Y, which is not necessarily a weakness. Of course, if X interacts with nearly every relevant influence, then this is not much of a "specification."
To "tweak" X so that it doesn't interact with a third variable in the relationship between X & Y, you have to eliminate either the covariance between X and that variable, or the higher level covariance between both X and Y and that third variable. So, you would need hypotheses about why your modifications would eliminate at least one of these connections.
Dear Alex,
I propose that you use a Problem Solving Process to evaluate the situation. Depict you process to make a problem diagnosis. Don't forget your goal, never. You worry tell you more than you imagine. Map facts. Remap them. Defer judgements until you have a real picture about what is going on.
OK?
I always think it is best to start with reasonably reliable main effects, and then look for moderators and mediators. If you have a very weak or unreliable main effect consider starting with better variables.
Dear Alex,
If there are too many preconditions (ad hoc hypothesis) that have to be satisfied, for theory X to be true, you might be prone to think that theory X should be abandoned, in favour for another, more progressive/productive theory. If such a one does not yet exist, you might be the one to start developing it...!
According to Lakatos (1978), too many such ad hoc hypothesis might be an indication of a degenerating research programme. See Lakatos (1978) philosophy of science (and specifically about his arguments about the logic behind how research programmes work) for a more full understanding.
Lakatos, I. (1978). The methodology of scientific research programmes: Philosophical Papers (Volume I). Cambridge: Cambridge University Press.
I would like to thank everyone for their responses. They were all extremely helpful!
- Badges
- Science topic
- Similar topics
- Psychology
- Organizational Psychology