As a running example, this tutorial relies on an imaginary data set that measures the relationship between latent intelligence and a latent sense of humor. The data can be downloaded from here as an SPSS file.
The measurement model for intelligence is based on scores from a standardized test with four components: reading, writing, math, and analysis.
The circles in the path diagram represent latent (unobserved) variables, and the rectangles represent manifest (observed) variables. The lines running from the latent variable to the observed variables represent factor loadings. In the full SEM (see below), arrows pointing from one latent variable to another represent structural relationships between the unobserved variables. The remaining arrows in the diagram represent variable-specific error not captured by the model. These are also sometimes represented by two-headed arrows beginning and ending with the same variable.
The measurement model for humor is based on three survey questions asking subjects to rate how much they enjoy The Simpsons, Family Guy, and American Dad.
The prediction is that higher intelligence leads to a better sense of humor. Thus, the full structural model is the following: