inst/scripts/Cap5.R

#-*- R -*-

##########################################################
###                                                    ###
### Script tratti da `Laboratorio di statistica con R' ###
###                                                    ###
###          Stefano M. Iacus & Guido Masaratto        ###
###                                                    ###
### CAPITOLO 5                                         ###
##########################################################

require(labstatR)


### Sez 5.1.1 LA LEGGE DEI GRANDI NUMERI

n <- seq(10,10000,length=40)
 
par(mfrow=c(2,2))
for(k in 1:4){
  mn <- numeric(40)
  for(i in 1:40)
   mn[i] <-  mean(rnorm(n,mean=10,sd=2))
  plot(n,mn,type="l",ylim=c(8,12),xaxt="n")
  abline(h=10,lty=2)
  axis(1,c(100,5000,10000))
}
par(mfrow=c(1,1))

### Sez 5.1.2 TEOREMA DEL LIMITE CENTRALE
n <- c(10,50,100,1000)
p <- 0.5
 
par(mfrow=c(2,2))
 
for(k in n){
  mn <- numeric(500)
  for(i in 1:500){
   x <- rbinom(k,1,p)
   mn[i] <- mean(x)
  }
 
  z <- (mn-p)/sqrt(p*(1-p)/k)
  hist(z,freq=FALSE,ylim=c(0,pnorm(0)),
   xlim=c(-4,4),col="red",main=paste("n = ",k))
  curve(dnorm(x),-4,4,add=TRUE)
 
} 
par(mfrow=c(1,1))

### Sez 5.2.1 INTERVALLO DI CONFIDENZA PER LA MEDIA
x <- c(0.39, 0.68, 0.82, 1.35, 1.38, 1.62, 1.70, 1.71, 1.85, 2.14, 2.89, 3.69)
 
s2 <- var(x)
mx <- mean(x)
n <- length(x)
l.inf <- mx - qt(0.975,df=n-1) * sqrt(s2/n) 
l.sup <- mx + qt(0.975,df=n-1) * sqrt(s2/n) 
cat("(",l.inf,":",l.sup,")\n")

t.test(x)
t.test(x, conf.lev=0.99)

### Sez 5.2.2 INTERVALLO DI CONFIDENZA PER LE PROPORZIONI
curve(x*(1-x),0,1)
abline(v=0.5,lty=2)

pn <- 0.51
n <- 2500
l.inf <- pn - qnorm(0.975) * sqrt(pn*(1-pn)/n) 
l.sup <- pn + qnorm(0.975) * sqrt(pn*(1-pn)/n) 
cat("(",l.inf,":",l.sup,")\n")

l.inf <- pn - qnorm(0.975) * 0.5/sqrt(n) 
l.sup <- pn + qnorm(0.975) * 0.5/sqrt(n) 
cat("(",l.inf,":",l.sup,")\n")

prop.test(x=1275,n=2500,corr=FALSE)
prop.test(1275, 2500)
binom.test(1275,2500)

pnorm(0.02*sqrt(2500))
pnorm(0.02*sqrt(1000))
pnorm(0.02*sqrt(500))

pbinom(1250,2500,0.51,lower.tail=FALSE)
pbinom(500,1000,0.51,lower.tail=FALSE)
pbinom(250,500,0.51,lower.tail=FALSE)

### Sez 5.2.3 INTERVALLO DI CONFIDENZA PER LA VARIANZA
x
var(x)
ic.var(x)
ic.var(x,FALSE)


### Sez 5.3.1 TEST D'IPOTESI PER LA MEDIA
x <- c(2928, 2997, 2689, 3081, 3011, 2996, 2962, 
 3007, 3000, 2953, 2792, 2947, 3094, 2913, 3017)
mean(x)
t.test(x, mu=3010, alternative="less")

x <- c(1975, 1869, 1879, 1790, 1860, 1895, 1810, 1831,
 1759, 1585, 1553, 1774, 1640, 1761, 1946, 1915, 1894, 
 1971, 1876, 1716, 1652, 1591, 1700, 1842, 1781)
mean(x)
t.test(x,mu=1730,alternative="greater")

### Sez 5.3.2 VERIFICA DI IPOTESI SULLE PROPORZIONI
z <- (32/50 - 0.5)/sqrt(0.5*0.5/50)
z
pnorm(z,lower.tail=FALSE) # il p-value
binom.test(32,50,alternative="greater")

prop.test(32,50,alternative="greater")

### Sez 5.3.3 VERIFICA DI IPOTESI SULLA VARIANZA
x <- rnorm(100, sd=5)
var(x)

test.var(x,20)
test.var(x,30)

test.var(x,22,"less")

### Sez 5.4.1 VERIFICA DI IPOTESI PER DUE PROPORZIONI
farmaco <-  c("Placebo", "Aspirina")
esito <- c("Infartuati", "Non infartuati")
tab <- matrix( c(239,139,10795,10898), 2,2,
    dimnames = list(farmaco,esito))

tab

p1 <- 239/11034
p2 <- 139/11037
p1
p2

p <- (239+139)/22071
p

qnorm(0.99)

prop.test( c(239,139), c(11034,11037) )
prop.test( c(239,139), c(11034,11037), alternative = "greater" )
prop.test( c(239,139), c(11034,11037), alternative = "less" )

### Sez 5.4.2 CONFRONTO TRA LE MEDIE DI GRUPPI
qt(0.95,df=10)

m <- rnorm(5,mean=540,sd=299)
f <- rnorm(7,mean=300,sd=238)
t.test(m,f,alternative="greater")

### Sez 5.4.3 CONFRONTO TRA VARIANZE
x <- rnorm(10, sd=5)
y <- rnorm(15, sd=3)
var(x)
var(y)

var.test(x,y,alternative="greater")
var.test(x,y,alternative="less")

### Sez 5.5 VERIFICA DI IPOTESI DI INDIPENDENZA
qchisq(0.95,df=4)
voti <- c("S", "B", "O")
tab <- matrix( c(1,1,2,3,3,1,0,2,2), 3,3, dimnames = list(voti, voti))
tab
summary(tab)
chisq.test(tab)

farmaco <- c("Placebo", "Aspirina")
esito <- c("Infartuati", "Non infartuati")
tab <- as.table(matrix( c(239,139,10795,10898), 2,2,
         dimnames = list(farmaco, esito)))
tab
summary(tab)

### Sez 5.6 ANALISI DI REGRESSIONE IN AMBITO INFERENZIALE
x <- rnorm(30,sd=3)
y <- 30+3*x
lm(y~x) -> mod
summary(mod)
anova(mod)
str(mod)

x <- rnorm(30,sd=3)
y <- 30+sin(3*x)
plot(x,y)
lm(y~x) -> mod
summary(mod)
anova(mod)

plot(x,y)
mod <- lm(y~x)
abline(mod)
curve(30+sin(3*x),min(x),max(x),add=TRUE)
abline(h=30,lty=2)


### Sez 5.6.1 BANDE DI CONFIDENZA
data(cars)
attach(cars)
lm(dist~speed) -> mod
pc <- predict(mod,interval="c")
pc
pp <- predict(mod,interval="p")
pp
plot(speed,dist)
abline(mod)
matlines(speed,pc[,2:3],lty=3:3,col=1:1,lwd=2:2)
pp <- predict(mod,interval="p")
matlines(speed,pp[,2:3],lty=2:2,col=1:1,lwd=3:3)
detach()
 

### Sez 5.7 ESTENSIONI DEL MODELLO DI REGRESSIONE
x1 <- runif(50)
x2 <- sin(runif(50))
y <- 4 + 2*x1 - x2 + rnorm(50)
lm(y ~ x1 + x2)

z <- data.frame(y, x1, x2)
lm( y ~ . , data = z)

# specificando i regressori da includere
lm(y ~ x1 + x3 + x5, data=z)
# o specificando quelli da escludere
lm(y ~ . - x2 - x4)
lm(y ~ x1 + x2 -1)

### Sez 5.7.1 ANALISI DELLA VARIANZA (AD UNA VIA)
x <- c(rep(1,6), rep(2,6), rep(3,6), rep(4,6)) 
y <- c(64, 72, 68, 77, 56, 95, 78, 91, 97, 82, 85, 77,
       75, 93, 78, 71, 63, 76, 55, 66, 49, 64, 70, 68)
bombe <- data.frame(y,x)
lm(y~x)
anova(lm(y~x))
plot(x,y)
abline(lm(y~x))

x <- factor(x)
x
lm(y~x)
anova(lm(y~x))

### Sez 5.7.2 REGRESSIONE LOGISTICA
glm(y~x1+x2+x3, family=binomial("logit"))
glm(y~x1+x2+x3, binomial)

str(interinale,vec.len=0)
glm(avviato~., binomial, data=interinale) -> model 
model
summary(model)

attach(interinale)
table(avviato, areares)
ftable(avviato, areares)
table(lingue)
prop.table(table(avviato))

glm(avviato~. - istruzione, binomial, data=interinale) -> model1 
summary(model1)

predict(model1,type="response") -> pr
plot(density(pr),xlim=c(0,0.2),main="",xlab="probabilita' di avviamento")
# estraiamo gli indici di x che corrispondono 
# all'intervallo (0.03, 0.05)
which(den$x > 0.03 & den$x < 0.05) -> intervallo
# in corrispondenza di questi cerchiamo il minimo 
# valore di y
which(den$y[intervallo] == min(den$y[intervallo]))

# estraiamo l'indice di x cui corrisponde il minimo 
# valore di y
intervallo[5]
den$x[46] # ecco la soglia cercata!
interinale2 <- data.frame(interinale, pr = pr, avviabile = (pr > 0.034) )
str(interinale2,vec.len=0)

table(interinale2$avviato, interinale2$avviabile) -> tab
tab
prop.table(tab,1)
 
### Sez 5.8.1 IL Q-Q PLOT
# Normale
x <- rnorm(100)
qqnorm(x)
qqline(x)
# Chi Quadrato
x <- rchisq(100,df=1)
qqnorm(x)
qqline(x)

# Normale vs Normale
x <- rnorm(100)
y <- rnorm(100)
qqplot(x,y)
# Normale vs Chi Quadrato
x <- rnorm(100)
y <- rchisq(100,df=1)
qqplot(x,y)

### Sez 5.8.2 LA FUNZIONE DI RIPARTIZIONE EMPIRICA
x <- c(3,1,4)
library(stepfun)
plot(ecdf(x))

# Normale contro Normale
x <- rnorm(50)
plot(ecdf(x))
curve(pnorm(x),min(x),max(x),add=TRUE)
 
# t di Student contro Normale
x <- rt(50,df=3)
plot(ecdf(x))
curve(pnorm(x),min(x),max(x),add=TRUE)
curve(pt(x,df=3),min(x),max(x),add=TRUE,lty=3,lwd=2)

# Uniforme contro Normale
x <- runif(50)
plot(ecdf(x))
curve(pnorm(x),min(x),max(x),add=TRUE)
curve(punif(x),min(x),max(x),add=TRUE,lty=3,lwd=2)
 
# Chi-Quadrato contro Normale
x <- rchisq(50,df=3)
plot(ecdf(x))
curve(pnorm(x),min(x),max(x),add=TRUE)
curve(pchisq(x,df=3),min(x),max(x),add=TRUE,lty=3,lwd=2)

x <- rnorm(50)
y <- rchisq(50,df=1)
xmin <- min( min(x), min(y) )
xmax <- max( max(x), max(y) )
 
plot(ecdf(x),xlim=c(xmin,xmax))
plot(ecdf(y),add=TRUE)


### Sez 5.8.3 IL TEST DI KOLMOGOROV-SMIRNOV
x <- rnorm(50)
ks.test(x,"pnorm",mean=1,sd=2)
x <- rnorm(50,mean=1,sd=2)
ks.test(x,"pnorm",mean=1,sd=2)

x <- rnorm(50,mean=1,sd=2)
mean(x)
sqrt(var(x))
ks.test(x,"pnorm",mean=mean(x),sd=sqrt(var(x)))
# si "accetta" H0

x <- rchisq(50,df=1)
mean(x)
sqrt(var(x))
ks.test(x,"pnorm",mean=mean(x),sd=sqrt(var(x)))
# si rifiuta H0

x <- rnorm(50)
y <- rchisq(50,df=1)
ks.test(x,y) # Normale vs ChiQuadrato
# si rifiuta H0

x <- rnorm(50)
y <- rnorm(50)
ks.test(x,y) # Normale vs Normale
# si "accetta" H0

### Sez 5.8.4 IL TEST CHI-QUADRATO DI ADATTAMENTO
p <- rep(0,5)
p[1] <- pnorm(-3)
p[2] <- pnorm(-1) - pnorm(-3)
p[3] <- pnorm(1) - pnorm(-1)
p[4] <- pnorm(3) - pnorm(1)
p[5] <- 1 - pnorm(3)
 
y <- rnorm(50)
n <- numeric(5)
n <- table(cut(y,c(-100,-3,-1,1,3,100)))
chisq.test(x=n,p=p)

y <- rt(50,df=1)
n <- numeric(5)
n <- table(cut(y,c(-100,-3,-1,1,3,100)))
chisq.test(n,p=p)

p <- c(9,3,3,1)/16
n <- c(315, 108, 101, 32)
chisq.test(x=n, p=p)
 
# EOF Cap5.R

Try the labstatR package in your browser

Any scripts or data that you put into this service are public.

labstatR documentation built on May 31, 2017, 3:38 a.m.