| Title: | A Semi-Parametric Test for Specifying Functional Form |
|---|---|
| Description: | A central decision in a parametric regression is how to specify the relation between an dependent variable and each explanatory variable. This package provides a semi-parametric tool for comparing different transformations of an explanatory variables in a parametric regression. The functions is relevant in a situation, where you would use a box-cox or Box-Tidwell transformations. In contrast to the classic power-transformations, the methods in this package allows for theoretical driven user input and the possibility to compare with a non-parametric transformation. |
| Authors: | Toke Emil Panduro <[email protected]>, Cathrine Ulla Jensen <[email protected]> |
| Maintainer: | Cathrine Ulla Jensen <[email protected]> |
| License: | GPL (>= 2) |
| Version: | 1.6 |
| Built: | 2024-12-03 06:30:53 UTC |
| Source: | CRAN |
PanJen is built on the idea that the choice of a functional form can be extrapolated from model fit measures. The function provides a ranking of different transformations according to their Baysian Information Criterion (BIC). The BIC provides a relative goodness-of-fit measure while accounting for the complexity of the model. The function provides BIC for each transformation, a model without the variable in question and a socalled smoothing of the variable. The models are estimated a Generalized additive model (GAM).A GAM is a special case of the Generalized Linear Model (GLM), where it is possible to include one or more so called smoothing functions. A smoothing function is a non-parametric way to include a continuous independent variable in a parametric model and thus make it semi-parametric. Please see Wood (2006) for an elaboration. The printed output is a table with the transformations sorted according to their explanatory power measured by AIC. The table shows both AIC and BIC for each regression, where BIC penalize for the number of explanatory variables
choose.fform(data,base_form,variable,functionList, distribution=gaussian)choose.fform(data,base_form,variable,functionList, distribution=gaussian)
data |
A data.frame |
variable |
A character-string with the name of the variable to test |
base_form |
A formula-object with the regressions without the variable that is tested |
functionList |
A list of transformations. Please see the example for an elaboration |
distribution |
Assumed distribution, see mgcv-vignette for an elaboration. Default is gaussian |
rank.table |
The printed table. The transformations are ranked according to their explanatory power measured by AIC. The table shows both the AIC and BIC value, where the BIC value penalizes for the number of variables |
models |
A list of estimated models |
dataset |
A dataframe with the dataset |
fforms |
The formula provided by the user |
Toke Emil Panduro & Cathrine Ulla Jensen
Simon Wood, Generalized Additive Models: an introduction with R.,hapman and Hall/CRC, 2006
## Test a linear specification (x), a log (I(log(x))) and a square I(x^2)) library(PanJen) data("hvidovre") form<-formula(lprice ~brick+roof_tile+roof_cemen+rebuild90+rebuild00+y7+y8+y9) fxlist= list( linear = function(x) x, sqr = function(x) x^2, log=function(x) log(x) ) PanJenAreaC<-choose.fform(data=hvidovre,variable="area",base_form=form, functionList=fxlist)## Test a linear specification (x), a log (I(log(x))) and a square I(x^2)) library(PanJen) data("hvidovre") form<-formula(lprice ~brick+roof_tile+roof_cemen+rebuild90+rebuild00+y7+y8+y9) fxlist= list( linear = function(x) x, sqr = function(x) x^2, log=function(x) log(x) ) PanJenAreaC<-choose.fform(data=hvidovre,variable="area",base_form=form, functionList=fxlist)
PanJen is built on the idea that the choice of a functional form can be extrapolated from model fit measures. The function provides a ranking of different transformations according to their Baysian Information Criterion (BIC). The BIC provides a relative goodness-of-fit measure while accounting for the complexity of the model. The function provides BIC for a set of predefined transformations aswell as a model without the variable in question and a socalled smoothing of the variable. The models are estimated a Generalized additive model (GAM).A GAM is a special case of the Generalized Linear Model (GLM), where it is possible to include one or more so called smoothing functions. A smoothing function is a non-parametric way to include a continuous independent variable in a parametric model and thus make it semi-parametric. Please see Wood (2006) for an elaboration. The printed output is a table with the transformations sorted according to their explanatory power measured by AIC. The table shows both AIC and BIC for each regression, where BIC penalize for the number of explanatory variables
fform(data,variable,base_form, distribution=gaussian)fform(data,variable,base_form, distribution=gaussian)
data |
A data.frame |
variable |
A character-string with the name of the variable to test |
base_form |
A formula-object with the regressions without the variable that is tested |
distribution |
Assumed distribution, see mgcv-vignette for an elaboration. Default is gaussian |
rank.table |
The printed table. The transformations are ranked according to their explanatory power measured by AIC. The table shows both the AIC and BIC value, where the BIC value penalizes for the number of variables |
models |
A list of estimated models |
dataset |
A dataframe with the dataset |
Toke Emil Panduro & Cathrine Ulla Jensen
Simon Wood, Generalized Additive Models: an introduction with R.,hapman and Hall/CRC, 2006
library(PanJen) data("hvidovre") form<-formula(lprice ~brick+roof_tile+roof_cemen+rebuild70+rebuild80+rebuild90+rebuild00+y7+y8+y9) PanJenArea<-fform(data=hvidovre,variable="area",base_form=form)library(PanJen) data("hvidovre") form<-formula(lprice ~brick+roof_tile+roof_cemen+rebuild70+rebuild80+rebuild90+rebuild00+y7+y8+y9) PanJenArea<-fform(data=hvidovre,variable="area",base_form=form)
A dataset with trade price and attributes for 900 homes in a Danish munnicipality
data("hvidovre")data("hvidovre")
A data frame with 901 observations on the following 19 variables
lpricea numeric vector, price, log price in 1000 EUR
bricka numeric vector, dummy, wall made out of brick =1
roof_tilea numeric vector, roof_tile
dummy, roof made out of tiles =1
roof_cemena numeric vector, dummy, roof made out of cement
y7a numeric vector, home sold in 2007
y8a numeric vector, home sold in 2008
y9a numeric vector, home sold in 2009
rebuild70a numeric vector, home rebuild in 1970's
rebuild80a numeric vector, home rebuild in 1980's
rebuild90a numeric vector, home rebuild in 1990's
rebuild00a numeric vector,home rebuild in 2000's
areaa numeric vector,living area in square meters
agea numeric vector, build year
bathroomsa numeric vector, number of bathrooms
highwaysa numeric vector, distance to nearest highway in meters
big_roadsa numeric vector, distance to nearest large road in meters
railwaysa numeric vector, railways
distance to nearest railway in meters
nature_SLDa numeric vector, nature_SLD
distance to nearest nature area in meters
lake_SLDa numeric vector,distance to nearest lake in meters
Panduro et al (in review JAERE)
Panduro T.E., Jensen, C.U, Lundhede, T.H., von Graevenitz, K. and Thorsen, B.J., Estimating demand schedules in hedonic analysis: The case of urban parks, (in review JAERE)
data(hvidovre)data(hvidovre)
plots the output from fform() and choose.fform()
plotff(input)plotff(input)
input |
A PJ-objected produced by fform or choose.fform |
Toke Emil Panduro and Cathrine Ulla Jensen
library(PanJen) data("hvidovre") form<-formula(lprice ~brick+roof_tile+roof_cemen+rebuild90+rebuild00+y7+y8+y9) PanJenArea<-fform(data=hvidovre,variable="area",base_form=form) plotff(PanJenArea)library(PanJen) data("hvidovre") form<-formula(lprice ~brick+roof_tile+roof_cemen+rebuild90+rebuild00+y7+y8+y9) PanJenArea<-fform(data=hvidovre,variable="area",base_form=form) plotff(PanJenArea)