Description Usage Arguments Value Author(s) References See Also
A function to robustly estimate both the mean and the dispersion function using B-splines bases. The robustification is done in a similar fashion to the robustbase::glmrob function.
1 2 3 4 |
formulaM |
the mean formula - typically of the type y ~ bsp(x1) + bsp(x2). Parametric models or mixtures of parametric and non parametric bases can be specified |
formulaG |
the dispersion formula - as for formulaM, but no response variable to be specified.
|
family |
as in glm, can be "gaussian", "poisson" or "Binomial" |
data |
dataset where variables are stored (optional) |
startM |
starting value for the coefficients needed to estimate the mean function |
startG |
starting value for the coefficients needed to estimate the dispersion function |
selection |
the method used to select the smoothing parameter.
can be "RGCV" (the default), "RAIC", "GCV", "AIC" or "none", in which case no selection is done and a smoothing parameter value should be provided in the |
weights |
as in glm - to be used with care, since the data get always weigthed by the estimated variance function |
control |
a list to control some behaviours of the estimation procedure, see DoubleRobGamControl |
trace |
should a trace to follow the convergence be printed? Default is FALSE |
scale |
similar to weigth, to be used if the dispersion function should be kept fixed |
weights.on.x |
mutated from robustbase::glmrob, a vector of weights for the x-variables. Used to accomodated for leverage points. |
method |
can be "quasi" (the default) or "pseudo", according to whether deviance or pearson's residuals should be used as response variables when fitting the dispersion function |
An object of the gamMD
class.
If both the mean and the dispersion function are estimated the object will contain the following elements:
converged
, convVec
, data
, fitG
, fitM
, iter
, relE
.
fitM
contains information on the mean estimation procedure;
fitG
contains information on the dispersion estimation procedure and has the same elements as fitM
.
If only the mean function is estimated all the values of the fitM
object are given in gamMD
object.
The elements of fitM
and fitG
are:
coefficients |
the estimated regression coefficients |
fitted.values |
the fitted values of the model |
desMat |
the full design matrix of the model |
dims |
the dimension of the design matrix associated to each component |
family |
family used for fitting the model - set to |
linear.predictors |
estimated linear predictors for the model |
deviance |
deviance residuals - in |
residuals |
Pearson residuals |
s.resid |
standardised Pearson residuals |
y, yd |
response variable used in, respectively, the mean and diseprsion estimation procedure |
converged |
logic indicator on whether the last inner iteration has converged |
sm.p |
smoothing parameters used in the estimation procedure - either fixed or chosen |
GCV |
Generalised Cross Validation value |
AIC |
Akaike Information criterion value |
RGCV |
Roubust Generalised Cross Validation value |
RAIC |
Roubust Akaike Information criterion value |
w.x |
the weights on the covariates used to correct for the effect of leverage points |
estRphi |
roubustly estimated dispersion parameter |
tcc |
tuning constant c used in the robust estimation |
yt |
the variable response variable transformed according to the link function and then centered |
df |
overall equivalent degrees of freedom for the fit |
vecdf |
a vector giving the degrees of freedom used by each covariate in the model |
cov.coef |
covariance matrix of the coefficients |
formula |
formula used |
dimsP |
size of the parametric part of the model |
Ilaria Prosdocimi (ilapro@ceh.ac.uk)
Croux, C., Gijbels, I. and Prosdocimi, I. (2012), Robust Estimation of Mean and Dispersion Functions in Extended Generalized Additive Models. Biometrics, 68: 31-44. doi: 10.1111/j.1541-0420.2011.01630.x
DoubleRobGamControl
, DoubleGam
, gamMD
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.