Hi Rebean,

I am going to assume that you have a sample size of at least N=200, otherwise even a correctly specified model may not lead to an acceptable solution.

I am also going to assume that the endogenous variable of interest was collected on a dichotomous scale and has not been recoded from a continuous or ordinal scale.

It is possible to analyze models in SEM (and specifically AMOS) that contain categorical variables. This requires the use of Bayesian Estimation instead of Maximum Likelihood Estimation (MLE) which is the default in AMOS (MLE assumes multivariate normality of observed variables). However, it is still assumed with Bayesian Estimation that each categorical variable is based upon a continuous variable that is normally distributed (e.g. agree/disagree as an indicator of level of agreement).

I would encourage you to check out "Structural Equation Modeling with AMOS" by Barbara Byrne which provides some, although limited, description on the treatment of categorical variables in AMOS.

Also, IBM has a number of tutorial videos on their website related to the analysis of ordered-categorical data that might be helpful:

http://amosdevelopment.com/features/ordered_categorical/index.html

Good luck!

Mike Rudolph, PhD

University of Kentucky

From your block diagram , i have observed the following: (1) the system is one which is "open loop" (see my definition of CLM vs OLM).(2)The architecture you illustrate is similar to that used for genetic algorithms ; (3)the "system" you define appears to by comprised of at least 4 subsystems (designated by F1, F2,F3, and F4). (4)the control variables of the system appear to be 4 groups of of 3. This yields 12 variables . (5) Feedback appears to be present only in the individual subsystems , and 6) design of the "dummy dependent variable " box would be enhanced by closure of a loop around it . Good luck in your pursuit -- if you find this of use, i would be happy to provide any other insight that i may have . Thank you for giving my this opportunity to assist you endeavors.

Hi Rebean,

I am going to assume that you have a sample size of at least N=200, otherwise even a correctly specified model may not lead to an acceptable solution.

I am also going to assume that the endogenous variable of interest was collected on a dichotomous scale and has not been recoded from a continuous or ordinal scale.

It is possible to analyze models in SEM (and specifically AMOS) that contain categorical variables. This requires the use of Bayesian Estimation instead of Maximum Likelihood Estimation (MLE) which is the default in AMOS (MLE assumes multivariate normality of observed variables). However, it is still assumed with Bayesian Estimation that each categorical variable is based upon a continuous variable that is normally distributed (e.g. agree/disagree as an indicator of level of agreement).

I would encourage you to check out "Structural Equation Modeling with AMOS" by Barbara Byrne which provides some, although limited, description on the treatment of categorical variables in AMOS.

Also, IBM has a number of tutorial videos on their website related to the analysis of ordered-categorical data that might be helpful:

http://amosdevelopment.com/features/ordered_categorical/index.html

Good luck!

Mike Rudolph, PhD

University of Kentucky

Logistic Regression can be achieved in AMOS.

The discussion on Categorical Response Variables here is very good: http://www.structuralequations.com/AmosTutorials.html

Dear all, thank you very much for your quick and valuable responses!...

@Michael; I have a sample of 600 observations, and my dependent variable is binary, it is not continuous (1= has the intention, 0= has no intention).

Does this mean that I can still use the Bayesian Estimation?

Hi Rebean,

Very good. Your sample size should be sufficient.

I apologize for any confusion generated by my first response. The endogenous variable does not need to be continuous, only it is assumed that if it is categorical there is some underlying continuous variable that is normally distributed from which the categorical variable is derived.

For example, in your case, where your variable contains categories of "has intention" and "has no intention," it can likely be assumed that even though the variable was measured/collected as a dichotomous variable that your subjects actually have varying degrees of "intention." And, assuming that if we were to hypothetically measure "intention" as a continuous variable and expect that it would be normally distributed, then it is appropriate to use Bayesian Estimation to estimate your model.

Mike