Abstract
Social science research deals with the highly complex phenomenon guided by various latent and interrelated constructs. As these constructs are invisible and not directly measurable, social scientists employ highly sophisticated multivariate analytical techniques that can simplify a complex data pattern to derive meaningful conclusions. Structural equation modeling (SEM) is a methodology used to represent, estimate, and test a network of relationships among variables. It is an integrated model for hypothesis testing and constructs' validity. This chapter deals with the various aspects of the SEM methodology, starting from describing the statistical algorithms used, the role of theory, measurement and structural model, stages in testing model validity, etc., meticulously explained in simple terms to provide with a basic understanding of this intricate technique.
TopBackground
SEM is an old statistical technique that has undergone progress over years, yet is not completely explored by researchers. The history of this technique started with the work of Spearman (1904) by introducing factor analysis method which forms basis for measurement of latent variables. Later with Wright (1918) path analysis approach of correlations were applied to determine relationships between parameters to estimate direct, indirect and total effects. With time, psychometrics research utilized theory testing through proposed hypothesis among abstract constructs (Tarka, 2018). Hence, theoretical assumptions are essential for analysis rather than theory building.
Role of Theory in SEM
To set ne should not go with gut feeling about relationships between variables, specification of parameters, etc. For this, a sound theory is essential. Theory can be thought of as a set of relationships providing consistency and comprehensive explanations of the actual phenomenon. This theory can also be published in the literature, previously developed models and established relations from various studies. A theory serves as a conceptual scheme based on foundational statements assumed to be valid for developing a model. It is very important that the SEM model be based on a theory because all the relationships must be specified before can be estimated. Therefore, theory is needed to determine measurement and structural models, for modifications to the proposed relationships, and many other aspects of evaluating a model. Theory also helps in establishing causation.
SEM vs. Multivariate Techniques
- 1.
In SEM, constructs can be represented as unobservant or latent factors independent relationships.
- 2.
Allows estimation of multiple and interrelated dependence relationships incorporated in an integrated model
- 3.
Provides an explanation of the covariance among the observed variables.
- 4.
Pursues to signify hypotheses regarding structural parameters about the variance, mean and covariance of observed data as the hypothesized model.
- 5.
In other approaches, direct and indirect effects based on pre-assumed relations are tested, but this is unlikely in SEM.
Key Terms in this Chapter
Latent Variables: These are the variables that are not directly measured or observed; a set of observed variables are used to define or infer the latent variable or construct (e.g., intelligence). Variables either if they are observed or latent, they can be defined as independent variables or dependent variables based on researchers’ interest.
Exogenous Variables: These are latent, a multi-item equivalent of an independent variable that is determined by factors outside of the model.
Construct: A concept with additional meaning evolved for scientific purposes.
Model: This is a statistical statement, expressed with equations or a diagram, about the hypothesized relationships among variables based on theory and research. In short, a model is the representation of theory.
Endogenous Variables: These are latent, a multi-item equivalent of a dependent variable that is determined by constructs or variables within the model.
Observed Variables: These are the variables that are measured directly, or also indicator variables; that is we actually measure these by using tests, surveys, and so on (e.g., grades).