Reihaneh Moniri Typically, you test an entire simultaneous-equation model in SEM that contains the relationships of all variables in a multivariate regression, path, and/or factor (latent variable) model. The global hypothesis is often that the model as a whole (with all proposed relationships) fits the data, or more formally, that the model-implied covariance and/or mean structure is equal to the covariance/mean structure in the population.

In addition, you can formulate much more specific hypotheses about individual variable relationships. For example, "X is indirectly related to Y via the mediator M" , "X positively influences Y with a path coefficient of 0.3", "X is a stronger predictor of Y than is Z", "Z moderates the relationship between X and Y", "X is unrelated to Y", or "X is a more reliable measure of factor F than is Z" are all possible hypotheses that could be tested with SEM.

The questionnaire items can be taken from the measurement side from previous international studies and then tested first with exploratory factor analysis and then moved to confirmatory factor analysis in order to test the study hypotheses. See Hair, JF, Hult, GTM, Ringle, CM, & Sarstedt, M. (2014) . A Primer on Partial Least Squares Structural Equation Modeling.

@Moniri if you're determined about your hypothesized Conceptual Model (hypothesized relationships) then you can test the entire model using SEM (as suggested by @Christian). You can go through attached papers to get some idea how research question have been transformed into hypotheses and based on the hypotheses, a conceptual framework has been developed and test using SEM