Chapter 6: QQ plots

Paul Northrop

This vignette produces some of the QQ plots that appear at the end of Chapter 6 of the notes.

> library(stat0002)

Normal QQ plots

We create plots that are similar to those in Figure 6.24 of the notes. They are slightly different from the ones in the notes because R’s qqnorm function uses a method of calculating the theoretical quantiles that is designed specifically for the normal distribution, whereas in Section 6.11.1 we used an approach that ties in with the way that we had calculated sample quantiles earlier in STAT0002. The dashed lines in the plots are drawn through the (sample and theoretical) lower and upper quartiles.

> # Normal
> normal <- rnorm(100)  
> qqnorm(normal, pch = 16, xlab = "Theoretical N(0, 1) Quantiles", main = "normal")
> qqline(normal, lwd = 3, lty = 2)
> # Create an outlier
> outlier <- normal
> outlier[100] <- 6
> qqnorm(outlier, pch = 16, xlab = "Theoretical N(0, 1) Quantiles", main = "outlier")
> qqline(outlier, lwd = 3, lty = 2)

> # Heavy-tailed (Student's t distribution, with 2 degrees of freedom)
> StudentsT <- rt(100, 2)
> qqnorm(StudentsT, pch = 16, xlab = "Theoretical N(0, 1) Quantiles", main = "heavy-tailed")
> qqline(StudentsT, lwd = 3, lty = 2)
> # Light-tailed (Uniform(-3/4, 3/4))
> uniform <- runif(100, -0.75, 0.75)
> qqnorm(uniform, pch = 16, xlab = "Theoretical N(0, 1) Quantiles", main = "light-tailed")
> qqline(uniform, lwd = 3, lty = 2)

> # Positively skewed (gamma(2, 1) distribution)
> gam <- rgamma(100, shape = 2)
> qqnorm(gam, pch = 16, xlab = "Theoretical N(0, 1) Quantiles", main = "positive skew")
> qqline(gam, lwd = 3, lty = 2)
> # Negatively skewed (10 - gamma(2, 1) distribution)
> neggam <- 10 - rgamma(100, 2)
> qqnorm(neggam, pch = 16, xlab = "Theoretical N(0, 1) Quantiles", main = "negative skew")
> qqline(neggam, lwd = 3, lty = 2)

Exponential QQ plots

We produce an exponential QQ plot based on the waiting times between births in the Australian birth times dataset. The rate $\lambda$ of the assumed Poisson process of births is estimated using the reciprocal of the sample mean, as detailed in the notes. The plot is equivalent to the figure in the notes, but the simulation envelopes are different because, unless we fix the random number seed, the simulated datasets on which the envelopes are based will be different each time we call qexp.

> # Calculate the waiting times until each birth
> waits <- diff(c(0, aussie_births[, "time"]))
> # Produce the QQ plot
> lambdahat <- qqexp(waits, envelopes = 19)

> # Estimate of lambda: rate of births per hour
> lambdahat * 60
estimate of lambda 
          1.839721 

The qqexp function offers the option to estimate $\lambda$ using $\ln 2 / m$, where $m$ is the sample median of the waiting times. Can you see why this makes sense? The following code implements this and shows how we can alter the appearance of the plot.

> lambdahat2 <- qqexp(waits, statistic = "median", envelopes = 19, pch = 16,
+                     line = list(lty = 2, lwd = 2, col= "blue"))

> # Estimate of lambda calculated from the sample median
> lambdahat2 * 60
estimate of lambda 
           1.56939