Research Thought Bubble #2: There Must Be A Shortcut!
After reading several sets of journal articles on possible forms of stat analysis, I find that I am back at square one. The process of fleshing out my thesis has been iterative; that is, I find myself saying that "aha! I have set of constructs worthy of testing using method X" only to find myself reading through the mountain of literature available on these constructs and method X, Y, and Z. I'm frustrated and dumbfounded with the truth that I will never really know enough. I am also daunted by the idea of never finishing the damned thesis because I am not confident in the research design. I can be such a perfectionist when my reputation is at stake, and I can be unforgiving of myself when it comes to giving birth to an aberration to an already struggling field of research. It's this sense of dread over betraying my integrity as a researcher that keeps me from saying "AHA! I HAVE A SOUND RESEARCH DESIGN!".
I'm exploring the possibility of creating a new model based on several correlations that have surfaced in the field. According to this YouTube & StatWiki dude, I could use Exploratory Factor Analysis (EFA) to test how a group of concepts fit nicely into several models. From what I understand, I ought to run Confirmatory Factor Analysis (CFA) once I find a "sound model" to test. With this kind of analysis, I can say that this model may explain a certain phenomenon to a certain degree without discounting the possibility of other models explaining that same phenomenon. That's quite a mouthful. On the other hand, I could opt for Structural Equation Modeling (SEM). This one will allow me to test the feasibility of a theory. Thus, I need to make sure that there's enough literature out there on the concepts or theory for me to say that I have a theory to test. Also, I will need to check if the correlations reported in the literature are significant enough for me to consider this kind of method.
I thought that SEM might actually bring them all together. I've seen how these concepts are correlated with one another, and how they can mediate one another in the literature. Models have been generated based on these correlations. There are theories of theories of theories on these concepts. Lots of measures on the concepts are available for me to use as I please. My problem here is that there are theories that account for the variables I want to test., but that they don't sufficiently explain the model I have in mind. Based on the literature, the operationalized definitions of these concepts seem to indicate that they are related to one another. When measured, they are correlated to one another...
Wait, I think I just realized something here. My method is SEM! I've got the established correlations, the degree to which they affect one another, and the theories to back it all up. The question is, how do I operationalize my variables? Do I say that one is categorica/dihotomousl (i.e. trait-like) and that the other is continuous (i.e. differential scales). So that was my dilemma all this time! I needed to look for literature that showed me how to manipulate these variables in such a way that I can make the model I had in mind work out!
I'm exploring the possibility of creating a new model based on several correlations that have surfaced in the field. According to this YouTube & StatWiki dude, I could use Exploratory Factor Analysis (EFA) to test how a group of concepts fit nicely into several models. From what I understand, I ought to run Confirmatory Factor Analysis (CFA) once I find a "sound model" to test. With this kind of analysis, I can say that this model may explain a certain phenomenon to a certain degree without discounting the possibility of other models explaining that same phenomenon. That's quite a mouthful. On the other hand, I could opt for Structural Equation Modeling (SEM). This one will allow me to test the feasibility of a theory. Thus, I need to make sure that there's enough literature out there on the concepts or theory for me to say that I have a theory to test. Also, I will need to check if the correlations reported in the literature are significant enough for me to consider this kind of method.
I thought that SEM might actually bring them all together. I've seen how these concepts are correlated with one another, and how they can mediate one another in the literature. Models have been generated based on these correlations. There are theories of theories of theories on these concepts. Lots of measures on the concepts are available for me to use as I please. My problem here is that there are theories that account for the variables I want to test., but that they don't sufficiently explain the model I have in mind. Based on the literature, the operationalized definitions of these concepts seem to indicate that they are related to one another. When measured, they are correlated to one another...
Wait, I think I just realized something here. My method is SEM! I've got the established correlations, the degree to which they affect one another, and the theories to back it all up. The question is, how do I operationalize my variables? Do I say that one is categorica/dihotomousl (i.e. trait-like) and that the other is continuous (i.e. differential scales). So that was my dilemma all this time! I needed to look for literature that showed me how to manipulate these variables in such a way that I can make the model I had in mind work out!