Hello Manvendra,

SEM is a broad church. However, I am taking from your comment that your ‘general model’ involves multiple constructs that are being regressed on each other (i.e., not just a measurement model). If so, then your output includes regression coefficients and intercepts/means … it’s just regression.

If you have a specific model you are using, yes, you need all the constructs involved since they can impact the regression coefficients elsewhere (via suppression, muliticollinearity, mediation, etc.). Therefore, modeling only parts of the model may not do the theoretical framework justice.

In regards to your comment on SEM not being sufficient to ‘validate’ … (other scholars): for regression coefficients, SEM is sufficient; in fact, because it takes measurement error into account, I would argue it is better than sufficient when compared to non-SEM regression (if using multiple indicators for constructs). The comment from other scholars (I suspect) may be in reference to causality. Because SEM allows the ‘mapping’ of constructs through pathways, individuals sometimes leap to the conclusion, for example, that construct A ‘causes’ construct B, which then ‘causes’ construct C. Although you have shown a significant association (path), there is more required to proving causality than a model which is significant. Causal inference has nothing to do with the methodological approach. It is about one’s ability to theoretically specify what must be a causal pathway because they contain no other confounders (i.e., is there something else which may be causing the relationship?). There is a whole literature on this, but a few references include:

Pearl, Judea. 2009. Causality: Models, Reasoning and Inference. 2nd ed. New York, NY, USA: Cambridge University Press.

Bollen, Kenneth A. and Judea Pearl. 2013. “Eight Myths about Causality and Structural Equation Models.” Pp. 301–28 in Handbook of Causal Analysis for Social Research, edited by S. L. Morgan. Dordrecht, Netherlands: Springer Science & Business Media.

Hope this is helpful, and wishing you well in your analysis,

Tim

Thanks sir, i do believe in same proposition, but of our collegues were arguing on the topic. so i thought let me confirm on this....

You see, SEM is a second generation method whereas Regression is  the first generation. 

For regression coefficient SEM is sufficient.