I am testing influence of three Independent variables (For example: X1, X2 and X3) on one single dependent variable (Y) through one mediator (M). The relationship has been tested via Process Macro v3 and through SPSS AMOS 21. In both cases the result shows partial mediation. Now, I would like to employ control variables in the above model. Some control variables are in nominal scale such as Gender, Purpose of Visit etc.. while others are in ordinal scale such as education level, income level etc.
Do I need to test control on mediation for each control variable separately? Or is there a way of treating categorical variables as continuous variables?
You need to be careful about using the qualitative variables and apply them to an appropriate scale that measures your actual objective. One tip is to use odd quantity scales. Ordinal variables can be applied to age, temperature, etc.
you should test the effect of control variable by connecting the control variable and dependent variable in the AMOS. Then, the effect of independent variable is going to change.
@Marcus, @Mehdi and @ Young-Rock, I have tested the effect of control variable on my mediation analysis using AMOS by adding it to Independent, Dependent and Mediator variable. Based on literature and convenience of interpretation I have chosen only dichotomous scale such as gender, and ordinal scales as control variables.
Some of the control variables were found to be significant at p
Dear @Deep, Some control variables are closely connected to the dependent variables and even serve to give your model more consistency. So, check all the possibilities of combinations between the variables, considering the best robustness for your model: a larger R2. Also note if there is no multicollinearity between the variables since this factor can bring a bias to its modeling.
@Marcus, Thank you for your valuable suggestions. Model Fit is all Good. I have tested adding control variables to Independent, Mediator and dependent variables. After test, only the control variables with significant relationship are kept in the model.
R-square value is also promising although it has reduced after adding control variables.
A non-parametric method that explicitly maximizes model predictive accuracy may be of interest in this domain. Specifically, classification tree analysis (CTA) may uncover mediation effects where conventional approaches do not, because CTA does not require any assumptions about the distribution of variables nor of the functional form of the model, and CTA systematically identifies all statistically viable interactions.
Here is a citation to a recent article comparing SEM and CTA analyses, should this be of interest:
https://www.onlinelibrary.wiley.com/doi/abs/10.1111/jep.12848?af=R
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling