I fully agree with the former comments and like to add that I can imagine that the motivation to create such models stems from a confounder-like-situation ("X and Y are only related because of Z" (which is a confounder as z causes Y and correlates with X). The formulation in quotation marks sounds like a mediator situation and, hence, may lead to the specification of a mediator model.
Such a model should, however, contain Z as a further predictor and not as a mediator.
In my point of view, a typical result of the causation-phobic language in the behavioral sciences ;)
Nothing in AMOS will stop you from doing so. However, the usual conceptual purpose of mediators is to reflect mechanisms of action of the exogenous variables, implying that differences in the mediators are caused by differences in the predictors (though the actual estimate of causal effects has much more to it than that). So you need to think of mediators as effects.
If I stretch my imagination, I can think of possible studies where age and gender might be endogenous variables, but they would definitely be a rare exception in the fields I work in.
A mediating variable is one that is on the causal pathway. Ethnicity cannot cause age, and with the exception of some hormones, a drug cannot cause a sex!
I fully agree with the former comments and like to add that I can imagine that the motivation to create such models stems from a confounder-like-situation ("X and Y are only related because of Z" (which is a confounder as z causes Y and correlates with X). The formulation in quotation marks sounds like a mediator situation and, hence, may lead to the specification of a mediator model.
Such a model should, however, contain Z as a further predictor and not as a mediator.
In my point of view, a typical result of the causation-phobic language in the behavioral sciences ;)
I also agree with the former comments, but I have the impression that the Author of question might mean moderators rather than mediators. Deepa, did you mean for example that something is influencing age or gender, or rather that the relation between other variables is different for men and women or for people in different age? If yes, you meant moderation, and yes, you can do it in AMOS (with some limitations, but generally yes)
If you are simply looking at the effects of a drug on some outcome and adding these demographic variables either as controls or as moderating variables, it may be much better to use simple OLS regression. If you want moderation you can use a series of interaction terms. Also, you can implement a series of dummy variables for all kinds of marital status and get a finer-grained analysis. AMOS does not always like to converge with many dichotomous variables. And if you truly want mediation (but I do not think you do) then you could do hierarchical regressions that introduce the predictors in stages; that will not show full mediation, but it give you an idea about these variables as intervening or "third-variable" predictors that could signal a spurious relationship in the data.
I believe you had an interaction effect (moderator) in mind. Simple regression in SPSS would be suitable; however, remember to center age first. For example, if age goes from 17-93 years old. I could provide the syntax for you if you needed.
April, if you allow me this side note: you mention centering in a way that moderation analysis would be harmed without it. That turned out to be a myth:
Echambadi, Raj, & Hess, James D. . (2007). Mean-centering does not alleviate collinearity problems in moderated multiple regression models. Marketing Science, 26(3), 438-445.
Moderation analysis with product terms inherently implies some multicolinearity. This cannot be eliminated by centering, unfortunately, as Echambadi and Hess show.
In social science research, mediator would be derived from theoretical argument. Ie like --love --- happiness. Thus, it very imposible to use demographic as mediator, but more likely for moderator. Can read some MacKinnon's papers.
Basically you are using variables with different scales of measurement. Amos is poorly designed to handle this situation. I would recommend using instead, in order of preference, Mplus, LISREL with polychorial correlations & asymptotic variance covariance, or EQS. Mplus is better suited to avoid violation of multivariate normality when using variables with mixed scales of measurement (e.g., continuous, ordinal, categorical), though.
sorry for again nitpicking a bit. The site you refer to, explains / recommends the old Baron/Kenny approach combined with the Sobeltest. There are newer and better approaches, see
Hayes, A. F. (2009). Beyond Baron and Kenny: Statistical Mediation Analysis in the New Millennium. Communication Monographs, 76(4), 408-420.
Many thanks. Yes I know. There are some other reccomendations from Preacher, and also from Schrout & Bolger. But, I think better to know both approaches.
Thank u all.. my dear friends. I tried with AMOS, the efficiency of the model will not show much changes in including and excluding demographical variables. I observed that in my study those variable have not much impact.