Structural Equation Modelling Flashcards
SEM
A very flexible technique that uses latent variables in order to answer complex questions. SEM develops a model then determines if the data could have been generated using the created model.
Latent Variable
Assumed to exist from shared variance between observable indicators. Assumed to be error free because it is the shared variance between all measured variables.
Underidentified Model
Too many unknowns in the model. More parameters than data points.
Parameter
Set of values that indicate the nature of the relationship between variables. Can either be fixed (not from data) or free (from data)
Just Identified Model
Number of parameters equals the number of data points. Uses all information so cannot test model.
Over Identified Model
Number of parameters exceeds number of data points. Not all information is used so model can be tested and compared.
Saturated Model
Contains all possible pathways. Perfect fit but very complex.
Independence Model
All pathways are assumed to be zero. Very simple but basically useless.
Default Model
Hypothesised model.
Fit Indices
Range of statistics. Look at multiple indices to determine if the model is a good fit. Typically use CMIN and RMSEA (want below .08)
Assumptions of SEM
Requires large sample, normality, linearity, no outliers, must be based on theory (confirmatory)
Structural Model
Defines the relationship between latent and measured variables.
Measurement Model
Defines the latent variables (what makes up the latent variables)
Endogenous Variables
Receive and exert directional influence on other variables in the model
Exogenous Variables
Don’t receive or exert influence on other variables.