Lecture 10: Testing scales and confirmatory factor analysis Flashcards
Confirmatory factor analysis
Brown, T. (2006
What is CFA?
CFA can help us to confirm latent data structures and improve the fit of our model.
In EFA we took a part of our data and want to find out what simplified model lies underneath the data. With CFA we can now take the model we found from EFA and test it on the rest of our data. –> To find out how well the model actually describes data.
Why is CFA useful?
- Testing reliability
- Test of measurement invariance
- Testing of validity of measurement model
What is a parent model?
Where you have two factors
What is a nested model?
Where you remove the second factor, and let the items load on only one factor
Less complex model of the parent model
Purposes and advantages of CFA
The specification of CFA is strongly driven by theory or prior research evidence.
If the CFA model is misspecified (e.g., failure to specify the correct number of factors, pattern of factor loadings, etc.), a poor-fitting solution will result.
What is the goodness of fit?
We are comparing our different models based on their goodness of fit indicators
- This could e.g. be chi-square test
Structural Equation Modeling (SEM)
Tabachnick et. al., 2013
Definition of SEM
SEM is used to show the causal relationships between variables. –> vizualization
SEM is a collection of statistical techniques that allow a set of relationships between one or more IVs, either continuous or discrete, and one or more DVs either continuous or discrete, to be examined.
SEM diagrams
The first step in an SEM analysis is a specification of a model
- Measured variables (observable variables) are represented by squares and rectangles
- Factors are represented by ovals in path diagrams
- Relationships between variables are indicated by lines
- Lines have either one or two arrows
- The goal of the analysis might be to test a model, to test specific hypothesis about models, modify existing models etc.
Model identification
Only models that are identified can be estimated.
A model is identified if there is a uniwue numerical solution for each of the parameters in the model
