SEMI-PARAMETRIC GENERALIZED ADDITIVE CONSTRAINED REGRESSION

The systematic component for the model is eta=f₁(x₁)+...+f_J(x_J)+z'b+e

where b is a parameter vector and the f_j j=1,...J

has one of the following shapes

0=constant
1=increasing
2=decreasing
3=convex
4=concave

The following code has arguments:
y = response vector n by 1
xmat = n by J matrix of predictors to be modeled non-parametrically with "shape" constraints
zmat = n by p matrix of predictors to be modeled parametrically

The code returns:
edf0 = the expected null degrees of freedom for each of the possible models
cic = the CIC values for each of the possible models

least-squares version

binary response version

count response version