The direction of effect is not enough for prediction. The variables in the equation should lead to significant effect in order to conclude the direction and magnitude.

The direction of effect is not enough for prediction. The variables in the equation should lead to significant effect in order to conclude the direction and magnitude.

The direction is up to a researcher while building a statistical, theory based model. So the direction is not assessed through statistics but through its compliance with knowledge. Non-significance indicates for example poor predictive utility of the variable among any other issues such as multicollinearity etc.

Best wishes

If we have variables which have a insignificant direct effect but significant indirect effect should we conclude that the mediating variable is fully mediating the relationship and the variables impact only via the mediating variable?

I have 9 out of 10 variables having significant indirect effects on the dependant variable. 1 variable has a significant direct effect on the dependent variable and no significant indirect impact. 

Hello Shweta,

Both paths, from the predictor to the mediator, and from the mediator to the dependent variable should have significant effects. 

Best wishes

Hello Shweta,

I think that it would benefit you greatly if you read, Beyond Baron and Kenny: Statistical Mediation Analysis in the New Millennium by Andrew F. Hayes (DOI:10.1080/03637750903310360 ) The article discusses several methods for interpreting the effects from a SEM, such as:

(1) the causal steps approach popularized by Baron and Kenny (1986), that is where the each of the paths in the model is estimated and the decision is made  whether a variable functions as a mediator if certain statistical criteria are met.

(2) the product of coefficients approach (commonly known as the Sobel test)

(3) the empirical M-test (also known as the distribution of

products approach) which has been  advocated by Holbert and Stephenson (2003)

(4) bootstrapping

After taking a look at this these I am confident that you will be able to choose the best method for your work.

Best regards.

Kirlinger and Pedhazur recommend the use of criterion of meaningfulness such as 0.05 and deleting the direct path  if the direct effect is less than 0.05. In fact that is what we use use in modifying an existing theoretical model and suggesting a new model  that is consistent with reality in which we reproduce the simple relation (r) without the direct effect(s)

Thank you everyone for the valuable inputs. 

 I read through the article suggested by Kenisha Samantha Russell Jonsson and that has clarified my doubts. One does not need to only look at total effects- indirect effects significance is important to look at. Thanks everyone.