[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Calculation of total effects and the significance test of indirect effect
At 02:05 AM 9/30/2004 -0500, you wrote:
The correlation between X and Y can be due to several things: direct
effects, indirect effects, common causes, correlated causes. Estimates of
the total effect of X on Y will differ depending not only on the other vars
in the model, but how you think they are interrelated. For example (hard
to draw diagrams unfortunately!) if you thing A affects B, and A and B
affect C, then B and C are correlated because B affects C and also because
they share a common cause, A. But, if B affects A, and A and B both affect
C, then B and C are correlated because B affects C directly and also
indirectly (through A). Even though your estimates of the direct effects
of A and B on C are the same in either model, your estimates of the total
effect of B on C will differ.
I am writing to ask a few questions regarding the calculation of total
effect = direct effect + indirect effect.
Suppose I am interested in the relationship between X and Y and Z. I also
have demographic variables to consider such age and sex.
1) My first question is,
If I want to calculate the total effect of X on Y, do I have to control
for Age and Sex or not?
In other words, is Rxy the total effect or Rxy.age,sex the total effect?
Turning more specifically to your problem, you'll (a) have omitted variable
bias if you leave out age and sex (b) part of the correlation between X and
Y that is due to common or correlated causes (age and sex) will instead get
attributed to the direct or indirect effects of X on Y.
In short -- if age and sex are theoretically relevant variables, then yes,
they should be in there, and your estimates of the total effects will be
affected by their presence or absence. Given your description, Rxy.age,sex
may be the total effect of X on Y, but again the estimates of total effects
are dependent not only on what vars are in the model but how you think they
are interrelated (e.g. it wouldn't be the total effect if you thought that
Z affects X rather than X affects Z).
2) My second question is,
For a simple model, it is not that hard to do by hand. But, there is a
freebie student edition of Lisrel that might meet your needs:
As far as I know, the SEM package softwares (e.g. Lisrel or Amos)
automatically calculate the direct and indirect effect of included components.
Unfortunately, I now do not use such softwares but just SPSS, and need to
calculate the total, direct and indirect effect between X Z and Y.
I will appreciate it much more if you can suggest any reading materials to
grasp these topics more clearly.
I've always liked Otis Dudley Duncan's book "Introduction to Structural
Equation Models". My adaptation of his arguments can be found at
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
WWW (personal): http://www.nd.edu/~rwilliam
WWW (department): http://www.nd.edu/~soc
* For searches and help try: