|
|
Structural Equation Modeling and Path Analysis
in Dissertation & Thesis Research
Structural equation modeling (SEM) in dissertation and thesis research combines factor analysis, multiple regression, and canonical correlation. Reasons for using SEM can include having both latent and observed variables, independent and dependent variables, and a desire to predict the values of some variables from other variables.
The SEM technique determines whether the dissertation model fits reasonably well to the data and evaluates the contribution of each independent variable to each of the dependent variables. Using structural equation modeling for your analysis also allows you to compare alternative models and evaluate differences between groups.
Other Names for SEM
To avoid confusion, structural equation modeling in dissertation research can also be called causal modeling, simultaneous equation modeling, analysis of covariance structures, and causal analysis. Dissertation research often uses these terms interchangeably. Sometimes, articles refer to structural equation modeling as path analysis or confirmatory factor analysis. In fact, path analysis and confirmatory factor analysis are two different types of structural equation modeling that are sometimes used in dissertation research. Confirmatory factor analysis will not be discussed in this article.
Deciphering Path Analysis
If you are reading articles that involve path analysis, you will come across many confusing diagrams. Here are some helpful tips for decipering these images:
- Measured variables are represented by squares or rectangles and their names are in lower case letters.
- Latent variables, or constructs, are represented by circles, and their names have the first letter of each word capitalized.
- The relationship between variables are indicated with lines.
- If there is no line connecting two variables, this means that a direct relationship between these two variables is not predicted.
- If a direct relationship between two variables is predicted, the line will have either one or two arrows.
- A straight line from one variable to another variable with one arrow indicates the direction of the relationship between the two variables. The variable with a line coming from it is an independent variable, while the variable with the arrow pointing to it is a dependent variable.
- If two variables have an arched connecting line with arrows at both ends, this simply means that the model has not predicted the relationship direction.
SEM Applications
Now that you know how to read a structural model path diagram, let's discuss when you might use structural equation modeling to analyze your data. Let's say your paper has two groups of randomly-assigned subjects. One group of subjects (college students) is the treatment group and receives a pill that is thought to increase learning. A second group of subjects (also students) is the control group and does not receive the pill. A test is given to both groups at the end of two weeks of class instruction and the groups' test scores are compared. A simple analysis of variance can test whether the treatment (the pill) really affects test scores (direct relationship between independent variable A and dependent variable B).
But what if you hypothesized that subjects receiving the treatment experienced an increase in self-esteem as a result of the treatment, and that this increased self-esteem is actually what led to better scores on the test? The relationship between the treatment and the score on the test is mediated by self-esteem level. This poses a different question than an ANOVA or even an ANCOVA can answer.
The path diagram for this example may look something like this:
- Path A = Direct relationship between observable independent variable (treatment) and dependent variable (test score)
- Path B = Direct relationship between observable independent variable (treatment) and dependent variable (level of self-esteem)
- Path C = Direct relationship between observable independent variable (level of self-esteem) and dependent variable (test score)
You will notice that this example used only observed, measurable variables that define a path analysis.
A big benefit of using structural equation modeling path analysis in dissertation research is that it allows complex relationships between variables to be examined. When your paper is concerned with complex and multidimensional phenomena, SEM is the only analysis that tests all of the relationships of the variables completely and simultaneously.
As path analysis and SEM are for very complex designs, learning your way around the statistical analysis software programs required to complete the analysis can be a daunting task. Most standard statistical software packages do not come with structural equation modeling or path analysis programs, so this might be an extra cost.
|
|
