I am doing my study on the employees working in private sector banks with a sample size of 300. I wants to test moderating effect of personality traits on the relationship between technostress creators and job outcomes, in correlation analysis i found that, only one dependent variable is significantly correlated, now should run regression analysis with all dependent variable or only one variable which is correlated?
There are several methods to select the proper multivariable model.
- You can run univariable analysis and include the variables with p value of less than 0.1 or 0.2 in the multivariable model
- do a stepwise regression with a forward or backward selection
- include variables depending on their importance to the study
you can test model fit using AUC or Hosmer Lemeshaw method (for logistic regression)
I agree, a HMR could be an option, adding variable according to the theoretical relevance.
For your moderation analyses it is not necessary that the predictor and outcome variables are correlated. The reason for is that, for example, a positive relationship with an outcome may exist under the high moderator valuue, while a negative relationship may exist when the moderating variable has a low level (simple slopes). So if you ignore the moderation, you find a zero relationship.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India