Hello Gemechu,

Strictly speaking, path analysis (when all variables are manifest/observed) is nothing than a set of one or more OLS regression models. So, the terse answer is, yes.

The relevant standardized coefficients are:

Model 1: C = Beta11*A + Beta12*B.

Model2: D = Beta21*C. (Note that "Beta21" is just the correlation of C and D.)

"Indirect effect" of:

A on D is: Beta11 * Beta21;

B on D: Beta12 * Beta21.

Only a direct effect of C on D ( = Beta21).

You could use these three estimated coefficients (plus the unanalyzed correlation of A & B) to reproduce the observed correlations among A,B,C, and D in order to determine how successfully the model you hypothesize accounts for the relationships observed in the data. This could be by simple comparison (e.g., r-observed vs. r-estimated) or by standardized root mean square residual (SRMR).

Good luck with your work.

Hello Gemechu,

Strictly speaking, path analysis (when all variables are manifest/observed) is nothing than a set of one or more OLS regression models. So, the terse answer is, yes.

The relevant standardized coefficients are:

Model 1: C = Beta11*A + Beta12*B.

Model2: D = Beta21*C. (Note that "Beta21" is just the correlation of C and D.)

"Indirect effect" of:

A on D is: Beta11 * Beta21;

B on D: Beta12 * Beta21.

Only a direct effect of C on D ( = Beta21).

You could use these three estimated coefficients (plus the unanalyzed correlation of A & B) to reproduce the observed correlations among A,B,C, and D in order to determine how successfully the model you hypothesize accounts for the relationships observed in the data. This could be by simple comparison (e.g., r-observed vs. r-estimated) or by standardized root mean square residual (SRMR).

Good luck with your work.

Thank you David Morse ! This is very helpful!

Very interesting question... worth following.