Dear all,
Kindly clarify my doubts and advice me regarding linking of control variables.
My models consist of three variables X, M, and Y. X is an independent variable, M is a mediating variable, and Y is a dependent variable.
I have selected a group of variables as controls (age, gender, and other demographics).
There are two models for the analysis (using SPSS AMOS)
Model 1 consists of mediation effect.i.e., X-->M-->Y; here I have taken all the control variables and linked them with both M and Y. The model fit seems to be adequate on respecification.
In Model 2, I have considered the direct relation only between M-->Y (now M is acting as an independent variable; I have excluded X from the model and the residual error term on M has been removed). On linking the control variables with both M and Y(as previously present in Model 1) , the model fit seems to be inadequate/irrelevant. However, on removing the controls from M and link them exclusively to Y, the model fit seems to be absolutely fine.
One more observation is that in Model 1 (Mediation), the regression coefficient between M--> Y is found to be -ve but in Model 2, the relation between M (act as an independent variable ) and Y is +ve. Additionally, gender, which is insignificant in Model 1 becomes significant in Model 2.
Please clarify the matter that why Model 2 becomes irrelevant when I link the control group to both M and Y however, previously in Model 1, it works. Why the direction of relationship M-->Y changes from -ve to +ve ?
Your model #1 is the most commonly used, where the control variables are in the same "exogenous" position as X, so that all of them are correlated with X and each of them can have an effect on both M and Y. This allows any of the effects of the control variables to be mediated.
Also, in a mediation model, M is always an independent variable with regard to Y, at the same time that is a dependent variable with regard to X and the control variables. This "indirect effect" of X on Y via M is what mediation is all about.
I really do not understand what your Model #2 is, and especially what was involved in removing the residual error term. Remember, any variable can still have certain amount of measurement error.
Thank you, Sir...
Well, regarding model#2, it states exclusively about the relation between M and Y (or in other words, I can say that M is acting as X i.e. the independent variable).
Now, in that case, if I take the same controls and link them with both M (the independent variable now) and Y. The model fit becomes irrelevant. I need a clarification on this point.
A change in the sign of a coefficient might indicate a suppressor effect. This can happen when you add a predictor to a model when that added predictor is related to another predictor. For example, the b for X in the model Y = X might be positive. You add another variable (C) that correlates positively with X, and now the b for X in Y = C + X is negative. Generally this happens when the X-C correlation is quite a bit stronger than the X-Y correlation, and often C doesn't relate to Y.
The effects of control variables can be complicated, and much has been written about them recently. Here's a few sources:
Atinc, G. M., Simmering, M. J., & Kroll, M. J. (2012). Control variable use and reporting in macro and micro management research. Organizational Research Methods, 15, 57-74.
Becker, T. E. (2005). Potential problems in the statistical control of variables in organizational research: A qualitative analysis with recommendations. Organizational Research Methods, 8, 274-289.
Becker, T. E., Atinc, G., Breaugh, J. A., Carlson, K. D., Edwards, J. R., & Spector, P. E. (2015). Statistical control in correlational studies: 10 essential recommendations for organizational researchers. Journal of Organizational Behavior.
Spector, P. E., & Brannick, M. T. (2011). Methodological urban legends: The misuse of statistical control variables. Organizational Research Methods, 14, 287-305.
In a mediation model, you always need to include both X ->M and M->Y as paths. the only way to pose an alternative to mediation would be to eliminate one or the other of these two paths.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS