When a dependent variable has no any significant correlation between the dependent variables, is it possible that a regression model be established between them?
I think what you mean is can say y and x have correlation coefficient zero and yet still have a regression relationship. The answer is yes. Just consider a downward opening parabola. Best regards, David
Pearson correlation (r) verify only the linear relationship between X and Y, but may be to have other types of relationship, such as nonlinear regressions. A graph between X and Y help you to distinguish any probable correlation.
In summary:
IF any relationship between X and Y => r=0
But:
IF r=0 , we can verify other types of relationships.
If you are talking about Pearson correlation and least squares linear regression then the models are essentially the same. Thus a significant non-zero correlation is synonymous with a significant independent variable in a regression model. It is slightly more difficult to interpret when you have several independent variables but the principle is essentially the same.
Non-significant independent variables in correlation may have some bearings in non-linear regression model as suggested by Dr. Rafat and Dr. Booth. The significance level of Beta value may also indicates the importance of any independent variable though it may be non-significant in correlation. I also agree with Dr. Samules' statement of "It is slightly more difficult to interpret when you have several independent variables but the principle is essentially the same." Here, the basic assumptions of applying regression model for the variables (nature of scale used) is the important factor.
To answer your question, the following two points are to be studied:
1. In order to apply regression analyses between two random variables or more, the statistical assumptions of the model and variables have to be assured, and
2. If the correlation coefficient between the dependent and independent variables is not significant, i.e the estimated regression coefficients are not significant. In this case, the relation between variables by this model is not fitted.