R: Defining the constrained segmented distributed lag term
csdl
R Documentation
Defining the constrained segmented distributed lag term
Description
Function employed within the tempeff formula to specify the variable with a csdl relationship.
The function does not fit the model, it simply returns information exploited by tempeff to fit a
"modTempEff" model.
the variable, typically the temperature, having the constrained segmented distributed
lag parameterization with the response.
psi
numeric to provide the starting value for one or two breakpoints of the
constrained segmented distributed lag relationship.
L
a numerical two-length vector to specify how many lags have to be considered to assess
the effect of cold and heat.
ndx
apparent dimension (i.e. the rank) of the two B-spline bases for the DL curves
of cold and heat. Default to ndx=round(L/3).
ridge
a two-length named list of characters to specify the possible ridge penalty to be applied
to DL coefficients. This list has to be named ('cold' and 'heat') and each component has to be
in terms of "l". Use ridge=NULL whether no ridge penalty has to be employed. See examples below.
DL
logical indicating if the difference penalty should be applied to the DL coefficients
or to spline coefficients. Default to spline coefficients (DL=FALSE).
diff.varying
logical indicating if the difference penalty should be global
or depending on the lag value to penalize differences mainly at larger rather than early lags.
Details
This function has to be used within the formula in tempeff. It returns information to
fit a constrained segmented distributed lag parameterization within the Poisson regression model
via the function tempeff.
Value
A list with the temperature variable and the arguments as attributes.
Note
All the arguments of csdl may be passed via ... in the call of tempeff. This feature may
be useful when the same model has to be re-fitted (via update) by modifying only one argument
of csdl(). See tempeff.
Author(s)
Vito Muggeo
References
Eilers, P., Marx, B. D. (1996). Flexible Smoothing with B-splines and Penalties
Statistical Science11, 89–121.
Wood, S. N. (2006) Generalized Additive Models: An Introduction with R. Chapman and Hall/CRC Press.
Muggeo, V. M. R. (2008) Modeling temperature effects on mortality: multiple
segmented relationships with common break points
Biostatistics9, 613–620.
See Also
tempeff
Examples
## Not run:
# Evaluate temperature effects up to 45 lags for cold and heat and
# a single breakpoint; use 20 as starting value, a global difference
# penalty on spline coefficients and no ridge penalty
csdl(my.temperature,psi=20,L=c(45,45),ridge=NULL)
# Evaluate temperature effects up to 45 lags for cold and 15 lag for
# heat, via P-splines with a global difference penalty on DL coefficients
# and an additional quadratic ridge penalty
csdl(my.temperature,psi=20,L=c(45,15),DL=TRUE,ridge=list(cold="l^2",
heat="l^2"))
## End(Not run)