This function performs nonparametric graduation of mortality rates using discrete beta kernel smoothing techniques.
dbkGrad(obsq, limx, limy, exposures = NULL, transformation = c("none", "log", "logit",
"Gompertz"), bwtypex = c("FX", "VC", "EX"), bwtypey = c("FX", "VC", "EX"),
adaptx = c("a", "b", "ab"), adapty = c("a", "b", "ab"), hx = 0.002, hy = 0.002,
sx = 0.2, sy = 0.2, cvres = c("propres", "res"), cvhx = FALSE, cvhy = FALSE,
cvsx = FALSE, cvsy = FALSE, alpha = 0.05)# S3 method for dbkGrad
print(x, …)
# S3 method for dbkGrad
as.data.frame(x, row.names = x$limx[1]:x$limx[2], optional = FALSE, …)
# S3 method for dbkGrad
residuals(object, type = c("working", "proportional", "response",
"deviance", "pearson"), …)
a matrix (or an object which can be coerced to a matrix using as.matrix()) of observed mortality rates. Dimnames, if provided, should be numeric; row names should be ages and column names years.
optional vector of two integers; if provided, limx (limy) sets a lower and a upper row (column) limit. Only data within these intervals are graduated.
an optional matrix containing the exposed to the risk of death for each age and year. Dimensions of exposures should correspond to those of obsq.
an optional character string; the transformation specified is applied to the observed data before graduation. Graduated data are then back-transformed. Possible values are
"none" for no transformation, "log", "logit" and "Gompertz".
an optional character string. It specifies the type of bandwidth to be adopted by row (by column) and must be:
"FX" for a fixed bandwidth (default),
"EX" for an adaptive bandwidth based on exposures (see Mazza A, Punzo A, 2013a, for details);
"VC" for an adaptive bandwidth based on a vector of weights derived from the variation coefficients which, in turn, depends from the exposures (see Mazza A, Punzo A, in press, for details).
an optional character string. It is the type of adaptive bandwidth to be adopted by row (by column) and must be:
"a" a different bandwidth for each evaluation age x at which the rates are estimated;
"b" a different bandwidth can be attributed to each age, regardless from the evaluation point;
"ab" a different bandwidth can be selected for each evaluation point and for each age.
an optional scalar. It is the global bandwidth used for the variable on the rows (columns).
Default value is 0.002.
If cvhx=TRUE (cvhy=TRUE), then the smoothing parameter is computed by means of cross-validation using this value as an initialization.
an optional scalar.
It is the sensitive parameter used for the variable on the rows (columns).
Default value is 0.2.
If cvsx=TRUE (cvsy=TRUE), then the sensitive parameter is computed by means of cross-validation using this value as an initialization.
an optional logical; if cvsx=TRUE (cvsy=TRUE) then cross-validation is used to select the smoothing parameter.
Default value is TRUE.
Parameter hx, (hy) is the initial value used in cross-validation.
an optional logical; if TRUE then cross-validation is used to select the sensitive parameter.
Default value is FALSE.
The value of sx (sy) is used to initialize the cross-validation process.
an optional character string; if cvres="propres" (the default), then cross-validation selects the smoothing parameter and/or the sensitive parameter by minimizing the squares of the proportional differences between observed and estimated values, while if cvres="res" then the sum of square residuals is minimized.
an optional scalar.
When the exposures argument is provided, the function returns (1-alpha)*100% pointwise confidence intervals and pointwise confidence bands for fitted values.
Default value is 0.05.
a dbkGrad object
a NULL or a character vector giving the row names for the data frame. Missing values are not allowed. Default value is 0:x$limx, limy.
logical. If TRUE, setting row names and converting column names (to syntactic names: see make.names) is optional.
additional arguments to be passed to or from methods.
"working","proportional","response","deviance", "pearson"
a dbkGrad class object.
Returned from this function is an dbkGrad object which is a list with the following components:
a matrix containing the graduated values.
a matrix containing the working residuals fitted.values - obsq.
a matrix.
kernels %*% obsq returns the fitted.values
a scalar.
It is the cross-validation residual sum of squares (RSS) computed over the fitted values, using the residuals specified in cvres.
a scalar. It is the global bandwidth used for the variable on the rows (columns).
a scalar.
It is the sensitive parameter used for the variable on the rows (columns). It is returned when bandwidth = "EX" or bandwidth = "VC"
pointwise confidence interval. Returned when exposures is provided.
limits of the Bonferroni confidence bands. Returned when exposures is provided.
limits of the Sidak confidence bands. Returned when exposures is provided.
a matrix containing the observed mortality rates with dimensions set by limx, limy.
a matrix containing the exposures with dimensions set by limx, limy.
a vector with lower and upper row (column) limits. Only data within these interval are graduated.
an object of class call.
In the cross-validation routine, minimization is performed using the Levenberg-Marquardt algorithm (More 1978) in the minpack.lm package (Elzhov, Mullen, and Bolker 2010).
Elzhov TV, Mullen KM, Bolker B (2010) minpack.lm: R Interface to the Levenberg-Marquardt Nonlinear Least-Squares Algorithm Found in MINPACK. R package version 1.1-5. URL http://CRAN.R-project.org/package=minpack.lm.
Mazza A, Punzo A (2011) Discrete Beta Kernel Graduation of Age-Specific Demographic Indicators. In S Ingrassia, R Rocci, M Vichi (eds.), New Perspectives in Statistical Modeling and Data Analysis, Studies in Classification, Data Analysis and Knowledge Organization, pp. 127-134. Springer-Verlag, Berlin-Heidelberg.
Mazza A, Punzo A (2013a) Graduation by Adaptive Discrete Beta Kernels. In A Giusti, G Ritter, M Vichi (eds.), Classification and Data Mining, Studies in Classification, Data Analysis and Knowledge Organization, pp. 77-84. Springer-Verlag, Berlin-Heidelberg.
Mazza A, Punzo A (2013b) Using the Variation Coefficient for Adaptive Discrete Beta Kernel Graduation. In P Giudici, S Ingrassia, M Vichi (eds.), Advances in Statistical Modelling for Data Analysis, Studies in Classification, Data Analysis and Knowledge Organization, pp. 225-232, Springer International Publishing, Switzerland.
Mazza A, Punzo A (2014) DBKGrad: An R Package for Mortality Rates Graduation by Discrete Beta Kernel Techniques. Journal of Statistical Software, Code Snippets, 572, 1-18.
More J (1978) The Levenberg-Marquardt Algorithm: Implementation and Theory. In G Watson (ed.), Numerical Analysis, volume 630 of Lecture Notes in Mathematics, pp. 104-116. Springer- Verlag, Berlin-Heidelberg.
Punzo A (2010) Discrete Beta-type Models. In H Locarek-Junge, C Weihs (eds.), Classification as a Tool for Research, Studies in Classification, Data Analysis and Knowledge Organization, pp. 253-261. Springer-Verlag, Berlin-Heidelberg.
# NOT RUN {
data("ItalyM")
# unidimensional analysis
res1 <- dbkGrad(obsq=obsq, limx=c(6,71), limy=104, exposure=population, bwtypex="EX", adaptx="ab")
plot(res1, plottype="obsfit", CI=FALSE, CBBonf=TRUE)
plot(res1, plottype="residuals", restype="pearson")
plot(res1, plottype="checksd", restype="pearson")
residuals(res1, type="pearson")
# bidimensional analysis
res2 <- dbkGrad(obsq=obsq, limx=c(6,46), limy=c(60,80), exposure=population,
transformation="logit", bwtypex="VC", bwtypey="EX", hx=0.01, hy=0.008, adaptx="ab", adapty="b")
plot(res2, plottype="obsfit")
plot(res2, plottype="obsfit", plotstyle="persp", col="black")
# }
Run the code above in your browser using DataLab