Fits the constrained segmented distributed lag log-linear regression model to daily
time series data of mortality and temperature and additional confounding factors.
the model formula such as ‘response ~ parametric terms + csdl(temperature) + seas(day)’,
see details.
data
the dataset where the variables are stored.
subset
an optional vector specifying a subset of observations to be used.
na.action
a function to indicate how to handle NA observations.
fcontrol
a list with components returned by fit.control().
etastart
possible starting values on the scale of the linear predictor.
drop.L
integer, specifying whether the first 'drop.L' observations should be removed
before fitting. This is useful for model comparison purposes, see notes.
...
additional arguments to be passed to csdl() in the formula, see details.
Details
The function fits a log-linear regression model to assess the effects of temperature on mortality using a
‘constrained segmented distributed lag parameterization’ (csdl). It is assumed that the data are daily time series
of mortality (or perhaps morbidity) and temperature.
The left hand side of the formula includes the response (daily counts), and the right hand side may include one
or more of the following
linear confounders (such as influenza epidemics or day-of-week);
nonparametric long term trend, via the function seas;
the csdl effect of temperature via the function csdl.
All the arguments of csdl() may be passed via ... directly in the call of tempeff. This may
facilitate the user when different models have to be fitted by changing only some of (and not all) the arguments
of csdl(). See the example below.
Value
The function returns an object of class "modTempEff". It is the list returned by gam.fit of
package mgcv with the additional components
psi
The estimated breakpoint with corresponding standard error (bayesian and frequentist).
betaCold
The estimated DL coefficients for the cold effect.
SE.c
The frequentist standard errors of the cold DL estimates.
SE.c.bayes
The bayesian standard errors of the cold DL estimates.
ToTcold
Estimate and frequentist standard error of the total (net) effect of cold.
ToTcold.bayes
Estimate and bayesian standard error of the total (net) effect of cold.
edf.cold
The df associated at each spline coefficient of the DL curve of cold.
rank.cold
The apparent dimension of the B-spline basis of the DL for cold.
betaHeat
The estimated DL coefficients for the heat effect.
SE.h
The frequentist standard errors of the heat DL estimates.
SE.h.bayes
The bayesian standard errors of the heat DL estimates.
ToTheat
Estimate and frequentist standard error of the total (net) effect of heat.
ToTheat.bayes
Estimate and bayesian standard error of the total (net) effect of heat.
edf.heat
The df associated at each spline coefficient of the DL curve of heat.
rank.heat
The apparent dimension of the B-spline basis of the DL for heat.
rank.seas
When ndx.seas>0, the apparent dimension of the B-spline basis for seasonality.
edf.seas
When ndx.seas>0, the df associated at spline coefficients of seasonality.
fit.seas
When ndx.seas>0, the fitted long-term trend (on the log scale).
Note
When a csdl term is included in the formula, the first max(L) observations are discarded
before model fitting. When a csdl term is not included, the argument drop.L may be used to discard
the first drop.L observations anyway. Fitting models with the same number of observations may be useful
to compare them via likelihood-based criteria (via anova.modTempEff, say). tempeff() returns objects of class "modTempEff", so proper methods may be employed. The returned
object has class "modTempEff" even if tempeff() is called without csdl() in the formula,
or even if the model is fitted with fixed (not estimated) breakpoints (via tempeff(..,fcontrol=fit.control(it.max=0))).
Muggeo, V.M.R. (2008) Modeling temperature effects on mortality: multiple
segmented relationships with common break points
Biostatistics9, 613–620.
Muggeo, V.M.R. (2009) Analyzing Temperature Effects on Mortality Within the R
Environment: The Constrained Segmented Distributed Lag Parameterization
Journal of Statistical Software, 32 12, 1–17.
See Also
modTempEff-package, plot.modTempEff, summary.modTempEff,
gam.fit in package mgcv
Examples
## Not run:
library(modTempEff)
data(dataDeathTemp)
o1<-tempeff(dec1~day+factor(dweek)+factor(year)+factor(month)+
csdl(mtemp,L=c(60,60),psi=20),
data=dataDeathTemp, fcontrol = fit.control(display=TRUE))
#add a ridge penalty: note how you *can* specify ridge!
#you do NOT need to use csdl(..,ridge=..)
o2<-update(o1, ridge=list(cold="l^2", heat="l^2"))
#a model without temperature effects (the first drop.L obs are dropped)
o3<-tempeff(dec1~day+factor(dweek)+factor(year)+factor(month),
data=dataset,drop.L=60)
#see ?anova.modTempEff for model comparisons
## End(Not run)