Latent variables like materialism, stress and religiosity need some measurements. Look up questionnaire design on how to measure these variables. Use a path diagram to connect these variables together and join them together - DV, IV, moderating and mediating variables. The basis of connecting them together will be from theory.

So what is your question?

If you really want to use AMOS for your modeling, I'd recommend reading the following book:

Byrne, B. M. (2010). Structural equation modeling with AMOS Basic concepts, applications, and programming (Multivariate Applications Series).

Other than that, I'd recommend to use some more powerful software - like the lavaan package of R.

Hello Oeshwik,

instead of writing the single hypotheses, take a piece of paper (real or digital) and draw a path diagram involving all variables. As the different hypotheses have no connection it is difficult to get an overview. If you have the path diagram you have the rough structure for the model and the next step is to set up the measurement model.

With regard to your questions:

1) Is already answered. For indirect effects, estimate the standard error by bootstrapping

2) I would vote for real testing the model via the chisquare test and inspecting potential sources of misfit. To make such an outcome more reasonable, use 2-3 of the most clearest indicators for each latent variable (and do not conflate reflective and formative indicators). Besides fit, of course, check the plausibility of the coefficients.

3) As aforementioned, select the most salient variables from a theoretical perspective.

HTH

Holger

For the hypothesis 3 can have the variable of religiosity dichotomized into two levels of low and high religiosity and perform a multigroup analysis, you can also add this as a predictive effect on the variable stress, if significant would indicate a difference in the level of religiosity.

Latent variables like materialism, stress and religiosity need some measurements. Look up questionnaire design on how to measure these variables. Use a path diagram to connect these variables together and join them together - DV, IV, moderating and mediating variables. The basis of connecting them together will be from theory.

Thank you for all the insightful answers, many of my confusions are gone already. I have attached my research proposal and model.

Should I try Item parceling to converge my items to 4-5 items for each observed variable? OR should I run CFA with 14-17 items for each of my 4 observed variable and then keep the ones with the strongest loading for the Structural modelling?

I have already surveyed 1200 people in 6 districts using Questionnaire constructs from established researches.

You can plot according to the total factor scores of the tests, or you can try only with the total scores in case they are undimentional. In your model, it would be advisable to perform two separate analyses

Hi

a) first please do not dichotomize religiosity (sorry, Cristian, bad idea). It reduces variance and information to a massive amount and you have to deal with the now binary mediator with applying appropriate estimators (e.g., DWLS), see

MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D. (2002). On the practice of dichotomization of quantitative variables. Psychological Methods, 7(1), 19-40.

Maxwell, S. E., & Delaney, H. (1993). Bivariate median splits and spurious statistical significance. Psychological Bulletin, 113(1), 181-190.

Dimitruk, P., Schermelleh-Engel, K., Kelava, A., & Moosbrugger, H. (2007). Challenges in nonlinear structural equation modeling. Methodology, 3(3), 100-114.

b) with regard to the measurement models, I would--based on my understanding of the attributes in your model, select 2-3 crystal clear indicators and specify a measurement model. Parceling ASSUMES that all items measure the same underlying latent. This, of course, makes only sense, when the model structure including an attribute is a factor model. If instead, the attribute is a composite then add the items together and model the composite.

HTH

Holger

Holger Steinmetz Hi, Thank you so much for further clarification. If I conducted Principle Factor analysis to decrease Items, would that have any issue with validity of the research?

Also all the articles I read on similar research, they conducted a median split on religiosity to make two groups. However, what would be a better way to deal with this in your opinion?

I will go through the provided articles. Once again, Thank you very very much.

Hello,

here is an informative thread: https://www.researchgate.net/post/Dear_scholars_how_to_design_a_Structural_Equation_Modeling_surveys

If you really understand the meaning of a common factor model and "reflective indicators" and you have an idea or assumption about the meaning of your constructs, select 2-3 indicators that all reflect the same underlying factor. Then do a CFA...then add the structural model.

With regard to religiosity: Simply keep it as a variable. I assume (perhaps you can give some clarification) that it is a somewhat "degree of religiosity"? If yes keep it.

Best,

Holger

If you are testing a theory then use AMOS , if your model is conceptual then use smartPls, it's easy to draw model there and currently widely used by researchers around the world .