data.salarios<-read_csv("salarios.csv",
locale = locale(encoding = "latin1"))
data.salarios %>%
ggplot(aes(x=lnhwage)) +
geom_histogram(aes(y=..density..),
bins=20,
fill="#69b3a2",
color="#e9ecef",
alpha=0.9) +
ggtitle("Histograma del log del salario por horar") +
theme(plot.title = element_text(size=15))
El parámetro bw en geom_density especifica la mitad del ancho de ventana
Debemos de especificar la mitad de \(h\) que acabamos de calcular
data.salarios %>%
ggplot(aes(x=lnhwage)) +
geom_histogram(aes(y=..density..),
bins=20, fill="#69b3a2", color="#e9ecef", alpha=0.9)+
geom_density(aes(x=lnhwage, color='Óptimo'), kernel="epanechnikov", bw=h/2, adjust=1)+
geom_density(aes(x=lnhwage, color='1/2 óptimo'), kernel="epanechnikov", bw=h/2, adjust=0.5)+
geom_density(aes(x=lnhwage, color='2 óptimo'), kernel="epanechnikov", bw=h/2, adjust=2)+
theme(legend.position = 'right')+
scale_color_manual("h",values = c('Óptimo' = 'black', '1/2 óptimo' = 'red', '2 óptimo'='blue'))
delta.gausiano <- 0.7764
delta.uniforme <- 1.3510
delta.cuartico <- 2.0362
h.gausiano <- constante*delta.gausiano*ajuste*w.N^(-0.2)
h.uniforme <- constante*delta.uniforme*ajuste*w.N^(-0.2)
h.cuartico <- constante*delta.cuartico*ajuste*w.N^(-0.2)
data.salarios %>%
ggplot(aes(x=lnhwage)) +
geom_histogram(aes(y=..density..), bins=20, fill="#69b3a2", color="#e9ecef", alpha=0.9)+
geom_density(aes(x=lnhwage, color='epanechnikov'), kernel="epanechnikov", bw=h/2, adjust=1)+
geom_density(aes(x=lnhwage, color='gausiano'), kernel="gaussian", bw=h.gausiano/2, adjust=1)+
geom_density(aes(x=lnhwage, color='uniforme'), kernel="rectangular", bw=h.uniforme/2, adjust=1)+
geom_density(aes(x=lnhwage, color='cuártico'), kernel="biweight", bw=h.cuartico/2, adjust=1)+
theme(legend.position = 'right')+
scale_color_manual("Kernel", values = c('epanechnikov' = 'black', 'gausiano' = 'red', 'uniforme'='blue', 'cuártico'='green'))
data.sim %>%
ggplot(aes(x=x,y=y))+
geom_point()+
geom_smooth(aes(x=x, y=y, color='MCO'),
method=lm,
se=FALSE)+
geom_line(aes(x=x, y=y5, color='kNN, k=5'))+
geom_line(aes(x=x, y=y25, color='kNN, k=25'))+
theme(legend.position = 'right')+
scale_color_manual("Método",
values = c('kNN, k=5' = 'blue', 'kNN, k=25'='green', 'MCO'='black'))
fit.lowess <- lowess(data.sim$x, y=data.sim$y, f=25/100)
data.sim <- data.sim %>%
mutate(y.lowess=fit.lowess$y)
data.sim %>%
ggplot(aes(x=x, y=y))+
geom_point()+
geom_smooth(aes(x=x, y=y, color='MCO cúbico'),
method=lm,
formula= y ~ x+I(x^2)+I(x^3),
se=FALSE)+
geom_line(aes(x=x, y=y.lowess, color='Lowess k=25'))+
theme(legend.position = 'right')+
scale_color_manual("Método",
values = c('MCO cúbico' = 'blue', 'Lowess k=25' = 'red'))
$plr1
Partially Linear Model
Regression data: 526 training points, and 50 evaluation points, in 3 variable(s)
With 2 linear parametric regressor(s), 1 nonparametric regressor(s)
y(z)
Bandwidth(s): 3.176922
x(z)
Bandwidth(s): 3.945161
1.044574
V1 V2
Coefficient(s): 0.08373704 0.02174876
Kernel Regression Estimator: Local-Linear
Bandwidth Type: Fixed
Continuous Kernel Type: Second-Order Gaussian
No. Continuous Explanatory Vars.: 1
$plr2
Partially Linear Model
Regression data: 526 training points, and 50 evaluation points, in 3 variable(s)
With 2 linear parametric regressor(s), 1 nonparametric regressor(s)
y(z)
Bandwidth(s): 3.176922
x(z)
Bandwidth(s): 3.945161
1.044574
V1 V2
Coefficient(s): 0.08373704 0.02174876
Kernel Regression Estimator: Local-Linear
Bandwidth Type: Fixed
Continuous Kernel Type: Second-Order Gaussian
No. Continuous Explanatory Vars.: 1
$plr3
Partially Linear Model
Regression data: 526 training points, and 50 evaluation points, in 3 variable(s)
With 2 linear parametric regressor(s), 1 nonparametric regressor(s)
y(z)
Bandwidth(s): 3.176922
x(z)
Bandwidth(s): 3.945161
1.044574
V1 V2
Coefficient(s): 0.08373704 0.02174876
Kernel Regression Estimator: Local-Linear
Bandwidth Type: Fixed
Continuous Kernel Type: Second-Order Gaussian
No. Continuous Explanatory Vars.: 1Podemos especificar otras formas de estimar los intervalos de confianza
Recuperamos los objetos estimados guardados en g.robinson para hacer un gráfico con mejor edición
g <- fitted(g.robinson$plr3)
se <- g.robinson[["plr3"]][["merr"]]
lci <- g - se[,1]
uci <- g + se[,2]
#Este objeto nos dicen dónde fueron evaluados
exp.eval <- g.robinson[["plr3"]][["evalz"]][["V1"]]
fitted <- data.frame(exp.eval, g,lci,uci)
ggplot() +
geom_point(data=wage1, aes(exper,lwage), color='black', alpha=0.5) +
geom_line(data=fitted, aes(exp.eval, g), linetype='solid')+
geom_line(data=fitted, aes(exp.eval, uci), linetype='dashed')+
geom_line(data=fitted, aes(exp.eval, lci), linetype='dashed')