Date:         Sun, 16 Jan 2000 12:25:51 +0100
Reply-To:     NASSAR NAJI <nassar@CYBERCABLE.FR>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         NASSAR NAJI <nassar@CYBERCABLE.FR>
Subject:      Re: path analysis
In-Reply-To:  <85256867.0076EFE3.00@n21atm0.shu.edu>
Content-Type: text/plain; charset="iso-8859-1"

Shawn,


> To conduct a path analysis using SPSS I was under the impression that one
can just
> conduct several (depending on the number of endogenous variables) multiple
> regressions.
> In this case, the partial regression coefficients (standardized beta) are
> used as the path coefficients.  Is this correct, or do I need to conduct
partial
> correlations or use structural equation modeling techniques?

For me, path analysis has to be conducted when one has a latent
construct -mesured by several variables-
to analyze (mesurement by CFA, and then path analysis). Several multiple
regressions should give similar results if
- the exogeneous variables are uncorrelated (tested by a factor analysis)
- you have good reasons to suppose that there is no direct relationship
between your several endogeneous variables, if there is one, it's due to
some exo. variables which have a significant impact on several endo.
variables.
(tested by partial correlations)

I use multiple regressions, partial correlation and canonical analysis to
explore path analysis (ie generate hypothesis)
Construct + projection/extrapolation ---> Latent Variable Partial Least
Squares
Construct + only path analysis   ---> Variance/covariance technics
No construct + interactions between endo. variables ---> Simultenous
regressions
No construct + no interaction between exo. var.     ---> Multiple
regressions



Regards