Title: Composites and Nonlinear Modeling
1Composites and Nonlinear Modeling
data from Grace and Keeley (2006) Ecological
Applications
2Bivariate relationship between total cover and
richness
How do we model the fit to the nonlinear line?
3A Simple Nonlinear Model
4A Simple Nonlinear Model - Results
Note that these are standardized coefficients.
What do they mean?
Are the two main path coefficients significant
contributors to the model?
5Amos Output
p-values imply both paths contribute significantly
.
we can also test to see if the model fit is poor
with paths omitted.
6Model Evaluation
Chi-square increased by more than 3.84 when a
single path was omitted, thus missing path deemed
important to include.
7Using Composite to Summarize Nonlinear Effect
fixed path
zero variance
latent variable
A composite is created using a latent
variable having a zero variance. Identification
achieved by fixing one incoming path to 1.0.
8Nonlinear Model Results
standardized coefficient
R-square consistent with path coef.
Test for nonlinearity performed prior to
introduction of composite.
9What about when the nonlinear effect is
endogenous?
10Endogenous Nonlinear Model
Note that the polynomial term is treated as an
exogenous variable.
11Endogenous Nonlinear Model Results
12Endogenous Nonlinear Model with Composite
13We can also use composites for other purposes.
Step 1 - test that all paths should be in model.
soil prop1
soil prop2
earthworm density
e1
soil prop3
flooding frequency
Step 2 - estimate collective effect using
composite.
soil prop1
0
soil effects
soil prop2
earthworm density
e1
soil prop3
Now we can compare the collective effects of soil
properties to the effects of flooding frequency
on earthworm density.
flooding frequency