STAT1006 Regression and Nonparametric Inference

Computer Lab 2

Learning Check
When you complete this session, you will be able to:
1. revise paired t-test and two-sample t-test;
2. find the rank transformation for a set of paired data and two independent samples;
3. state the null and alternative hypotheses that are used for the analysis of data using
the Wilcoxon signed rank test for paired data and the Wilcoxon rank sum test for
two independent samples;
4. find the mean and the standard deviation of the sampling distribution of the W
under the null hypothesis for each test;
5. find the P-values for the Wilcoxon signed rank test for paired and the Wilcoxon rank
sum test for two independent samples using the Normal approximation with the
continuity correction;
6. Use computer output to determine the results of the Wilcoxon signed rank test for
paired data and the Wilcoxon rank sum test for two independent samples, for both
exact (psignrank OR pwilcox or wilcox.test (exact=TRUE)) and approximate methods
(wilcox.test(exact=FALSE)).
Before you start, load the stored datasets provided for the exercises in this week below by
running the following chunk (assuming the file “Workshop2.Rdata” is saved in the same
directory as the .Rmd worksheet).
print(load("Workshop2.Rdata"))

## [1] "Ex16.07" "Ex16.05" "Ex16.10" "Battery" "Ex16.17" "Ex16.18"

Overview of Signed-rank Test: Paired data

The Wilcoxon signed-rank (or sometimes just sign-rank) test calculates the ranks of the
absolute values of these differences. Under H 0, the sum of the ranks associated with the
positive differences (w +¿¿) should be roughly equal to the sum of the ranks of the negative
differences (w −). These sums of signed-ranks can be considered to be random variables,
and they too have a distribution that depends on the sample size n.
In R, the probability mass function, the distribution function, and the quantile function are
given by dsignrank, psignrank, and qsignrank, respectively.
Not surprisingly, there is an inbuilt function, wilcox.test, that allows us to perform the
signed-rank test (and the 2-sample rank-sum test below) directly. The syntax is similar to
that for t.test, so with the signed-rank test as the analogue of the single sample t-test, the
funciton requires a single input vector, `x’.
Overview of Wilcoxon Rank Sum test
To undertake a Wilcoxon rank sum test by explicitly doing the calculations we can work
through the following steps. Suppose the two samples to be compared are Sample A and
Sample B.
1. Rank the combined observations from the two sample, keeping track of which ranks
are associated with observations from each dataset.

2. Calculate the sum of the ranks of observations from Sample A (w A) and Sample B (
w B ). Then, calculate the statistics u A and u B as u A =w A −n A ( n A +1 ) /2 and
u B=w B − nB ( n B +1 ) /2, where n A and n B denote the respective sample sizes.

3. Calculate the probability, under the null hypothesis, of observing a value as small as,
or smaller than, the value of m in ( u A ,u B ) that we observed . This is done by using the
distribution of the Wilcoxon rank-sum statistics, which is given in R by the functions
d/p/q/rwilcox. If the probability is small, we would reject H 0 in favour of H A .

The inbuilt R function to do this test directly is again wilcox.test, but as it’s a two sample
comparison it requires two input vectors, x and y.
Question 1: Walpole et al. (2016) Ex 16.7
The data in Ex16.07 give the systolic blood pressure of 16 joggers before and after an 8-
kilometer run. You are to conduct a Wilcoxon signed-rank test at the 0.05 level of
significance to test the null hypothesis that jogging 8 kilometers increases the median
systolic blood pressure by 8 points against the alternative that the increase in the median is
less than 8 points.
a. Familiarise yourself with the data by using the function head to view the first 6
rows, and boxplot. Interpret the boxplot.
b. Is it appropriate to use a paired t-test in this scenario? Provide your arguments.
c. Use appropriate R functions such as sign and rank to calculate the test statistic for
the signed-rank test.
d. For a sample size of n, the signed-rank statistic can take values between 0 and
n(n+1)/2. The distribution of the possible values for a sample of size 8 can be
plotted with the following code. Uncomment it to run, and identify the p-value
associated with your calculated test statistic.
#nn<-8
#plot(0:(nn*(nn+1)/2), dsignrank(0:(nn*(nn+1)/2), nn), type = "h", xlab =
"w", ylab = "probability", main = paste0("Signed-rank distribution (n =
",nn,")"))

e. Calculate the p-value using the function psignrank. This is an exact method.
f. Use the function wilcox.test to check your result. This is an approximate method.
g. Summarise the results using the 6 steps hypothesis testing.
Question 2: Walpole et al. (2016) Ex 16.10.
The weights of 5 people before they stopped smoking and 5 weeks after they stopped
smoking, in kilograms, are given in Ex16.10. Use the signed-rank test for paired
observations to test the hypothesis, at the 0.05 level of significance, that giving up smoking
has no effect on a person’s weight against the alternative that one’s weight increases if he
or she quits smoking.
Summarise the results using the 6 steps hypothesis testing.
Question 3: Walpole et al. (2016) Ex 16.15.
A cigarette manufacturer claims that the tar content of brand B cigarettes is lower than that
of brand A. To test this claim, the following determinations of tar content, in milligrams,
were recorded:
Brand A: 11,12,9,13,11,14
Brand B: 8,10,7
Use the rank-sum test with α =0.05 to test whether the claim is valid.
Summarise the results using the 6 steps hypothesis testing.
Question 4
The object Battery contains the operating life (in hours) of batteries from two different
brands of replacement mobile phone batteries. The operating life was defined as the time
the level of a battery reached 10% after continuous use of the phone. Test the hypothesis
that Brand A yields higher operating times than Brand B.
a. Is this a paired comparison or not?
b. Plot the data in a useful way and comment.
c. Carry out a t -test using t.test. Summarise the results using the 6 step hypothesis
testing.
d. Carry out a rank-sum test using R and state your conclusions: extract the operating
life data for each battery, and then combine then using the function c(). Once you
have done so, you can then rank them easily and calculate the rank-sums using R.
e. Check that you have done the calculations correctly by using the function
wilcox.test.
f. Identify the p-value on the following plot. (Uncomment to run.)
g. Summarise the results for nonparametric using the 6 steps hypothesis testing.
#U <- 0:(9*9)
#Udensity <- dwilcox(U, 9, 9)
#plot(U, Udensity, xlab = "U", ylab = "probability", main = "distribution of
Wilcoxon rank-sum statistic", type = "h")
Question 5: Walpole et al. (2016) Ex 16.18.
A fishing line is being manufactured by two processes. To determine if there is a difference
in the mean breaking strength of the lines, 10 pieces by each process are selected and then
tested for breaking strength. The results are given in the dataset Ex16.18.
Use the rank-sum test with α =0.1 to determine if there is a difference between the mean
breaking strengths of the lines manufactured by the two processes.
Summarise the results using the 6 steps hypothesis testing.