you do not need SEM for simple correlations of individual items. When you start operating with latent variables/constructs (e. g. self-efficacy, optimism, beliefs etc.) and their interactions then SEM will serve you well. I cannot be more precise because you asked a very general question. If you would care to elaborate on your research design, I may be able to help you further.
This thread is also helpful: https://www.researchgate.net/post/What-is-the-difference-between-a-regression-analysis-and-SEM
The structural model refers to the relationships among latent variables, and allows the researcher to determine their degree of correlation (calculated as path coefficients).
Structural equation modeling (SEM) and confirmatory factor analysis (CFA) are useful when you have measurements (observed variables, indicators, items) that are less than perfectly reliable in the sense of classical test theory (i.e., when the measurements contain random error). SEM and CFA allow you to separate true score (reliable) variance from variance due to measurement error and to examine relationships between latent variables that are free of random measurement error. The idea is that the correlations and/or regression coefficients should be less biased after controlling for random error, thus allowing you to study associations between variables in a more accurate way. A prerequisite for most SEM and CFA models is that you have multiple indicators (at least 2, preferably 3 or more) per construct to identify the model and separate true score (reliable) variance from error variance.
Instead of relating it to research design, please relate the decision to question at hand. Simply, when we want to test a variable for its role as a determinant (Independent variable), linking mechanism (mediator) and contingency (moderator), we can use SEM.
Hi! It is usually related to research objectives of the study, the nature of the constructs, research model that intends to include multivariate analysis, multigroup analysis and so on. Thanks