Yes. This is usually what people look for in a mediation model. However, it may be helpful to check what you mean by a direct path. The total effect is usually denoted c, the indirect effect is ab (i.e., a x b representing the combined effects of X->M and M->Y) and the direct effect c' such that:

c = c' + ab

So if the total effect is completely accounted for my the mediating path then c' would be zero. In practice partial mediation is more common. Its possible you meant total rather than direct effect. This too can happen and is known as inconsistent mediation (where the direct and indirect effects are in different directions and approximately cancel each other out).

According to Baron & Kenny's (1986) criteria for mediation, both the a and b paths should be significant (I.e., the path from X to M and the path from M to Y). This makes sense to me since the mediated (indirect) effect is defined as the product a*b. If the a path is not different from zero, then also the indirect effect (product a*b) would not be different from zero.

Hey Thom Baguley Christian Geiser

Thank you for your comments. By direct path I mean, during my SEM path analysis, my hypothesis which state that perceived usefulness is correlated with attitude was not supported (Direct path between the perceived usefulness and attitude).

After conducting my path analysis, I have calculated the direct, indirect and total effect of my model and found perceived usefulness is a mediator between the external variable and attitude.

Christian Geiser based on what you just said, since in my SEM path analysis I have not found correlation between the perceived usefulness (being considered later on a mediator) and the attitude, should I consider that my finding concerning the mediation effect later on as biased? Or this can happen? Thank you.

When the a path is not significant then the typical conclusion would be that X does not influence M and hence there's no mediated effect. Note however that the a path may be significant even when the zero-order correlation between X and M is not significant. This is because you also have "perceived ease of use" as a predictor of attitude in the model along with "perceived usefulness." Therefore, the a path represents a partial regression coefficient that may be significant even when the zero-order correlation coefficient is not (e.g., due to a suppression effect). You want to base your assessment on the significance of a, not the significance of the zero-order correlation (this isn't entirely clear from your question/description).

I suspect that "perceived ease of use" and "perceived usefulness" are highly correlated with one another. Because of the high correlation, you may see a suppression effect (significant path coefficient between perceived usefulness and attitude even though the zero-order correlation between perceived usefulness and attitude is not significant). If this is true then you may want to check out the literature on suppression effects in path models.

Hello Jawad Harb. Please forgive me for commenting on something other than the question you raised. In the original question, you wrote:

"After checking the indirect effect (mediation effect) I have found that perceived usefulness is a mediator between the external variable and attitude."

By definition, a mediator is on a causal pathway between two other variables. So your statement is tantamount to saying that X causes M and M causes Y. This way of describing the results of mediation analyses is very common, but IMO, it is rather careless, especially when one has observational data (as opposed to experimental data). For alternative and more careful ways of describing your results, see section 6.6 in the paper by Fiedler et al. (2018). I think this article should be required reading for all mediation analysts. HTH.

Fiedler, K., Harris, C., & Schott, M. (2018). Unwarranted inferences from statistical mediation tests–An analysis of articles published in 2015. Journal of Experimental Social Psychology, 75, 95-102.

https://www.sciencedirect.com/science/article/abs/pii/S0022103117300628

Article Unwarranted inferences from statistical mediation tests – An...

Hey Christian Geiser

Thank you so much for highlighting this for me. I had no idea before what does a suppression effect means.I have checked some literature about these topics. I have looked at the output again. The path coefficient between the perceived ease of use (PEOU) and the perceived usefulness (PU) is strongly correlated (0.47).

I have tried to create a model including perceived usefulness, attitude and behavioural intention. I have found that perceived usefulness is highly correlated to both (0.739 with attitude and 0.779 with behavioural intention with a p value equal to 0.000).

I have looked again on the correlation matrix, I can see that the correlation between the perceived usefulness (PU) and attitude (ATT) and behavioural intention (BI) is negative (which I think this is a sign of a suppression effect). Can I conclude that perceived usefulness is a suppressor variable? how will this change the results?

After reading your comment, you mean that if the PU was a surppressor variable, should I still accept the hypothesis that state that there is a positive correlation between the PU and ATT or BI? I am in the right direction? Bruce Weaver Thom Baguley please feel free to comment in case you are familiar with this 😊. Thank you all for your time.

It looks like you posted the covariance matrix along with the unstandardized regression (path) coefficients. If you could post the correlation (standardized covariance) matrix and standardized path coefficients, it would be much easier to see what's going on in your model.

Hey Christian Geiser ,

Thank you for your reply and sorry for the confusion. kindly check the new output, will this help now? thank you.

I think you posted the correlation matrix of the coefficients. What we need to look at is the correlation matrix of the latent variables (which shows the zero-order correlations between the latent variables). By comparing the zero-order correlations among the latent variables with the standardized path coefficients, you can usually pretty easily figure out whether you have a suppression effect.

Hey Christian Geiser

I am sorry for the late reply, this was due to the time difference between the US and Australia. I was contacting Stata in the US to enquire about the "Correlation matrix of the latent variable which shows the zero- order correlation between the latent variables". Unfortunately, they did not understand what is this matrix, neither my professors at my University. Do you mean we need to look at the evidence of the relationship using the correlation matrix of the raw data ?

However, I have managed to produce a matrix that shows the Fitted covariances of observed and latent variables (standardized). Will this help? if not, is there any other way to do it? Thank you so much.

Jawad Harb It does not look like suppression to me if this is the latent correlation matrix. The signs of all latent correlations are positive and so are the signs of all path coefficients. As you had mentioned in your initial post, it looks like your b path [PU --> ATT] is not significantly different from zero (stand. coeff. = .338, p = .259). Therefore, I'd say it is questionable whether PU can be seen as a mediator. The indirect effect [PEOU --> PU --> ATT] may have reached statistical significance because the path PEOU --> PU is quite strong (.51). But this could be misleading if PU does not account for a significant amount of variance in ATT beyond what PEOU explains (the direct path PEOU --> ATT [.62] is also quite substantial). I hope this helps!