R Programming

Robin Evans
robin.evans@stats.ox.ac.uk
Michaelmas 2014

This version: November 5, 2014

Administration
The course webpage is at
http://www.stats.ox.ac.uk/~evans/teaching.htm
Lectures are at 10am on Mondays and Wednesdays, and practicals at 9am
on Tuesdays and Thursdays; in reality, there will be rather a lot of overlap
between these two formats.
Please bring your own laptop to use during all classes, and ensure that you
have R working (see below). If you dont have access to a laptop, let me
know and we will try to provide one.
I will hold office hours each week during Michaelmas term on Wednesdays
between 12pm and 1pm; my office is on the first floor of 2 SPR, room
204. Im very happy to help with any difficulties or problems you are having
with R, but please take steps to help yourselves first (see below for a
list of resources).

Software
You should install R on your own computer at the first opportunity. Visit
http://cran.r-project.org/
for details. Ensure you have the latest version (as of the start of Michaelmas
2014, this was version 3.1.1). Try to spend some time getting used to the
basics of the software, including arithmetic operations and functions. There
are many excellent online tutorials for this purpose.
1

Resources
A strength of R is its help files, which we will discuss. These are accessed
with the ? and ?? commands.
The internet has almost all the answers, and knows much more about R
than I do. If you have a problem, its extremely likely that someone will
have had the same difficulty already, and posted a question on an internet
forum.
Books are useful, though not required. Here are a some of them with brief
comments.
1. Venables, W.N. and Ripley, B.D. (2002) Modern Applied Statistics with
S. Springer-Verlag. 4th edition.
The classic text.
2. Chambers (2010) - Software for Data Analysis: Programming with R,
Springer.
One of few books with information on more advanced programming (S4,
overloading).
3. Wickham, H. (2014) Advanced R. Chapman and Hall.
A great new book on the more advanced features: a good follow up to this
class.
4. Crawley, M. (2007) The R Book. Wiley.
Very thorough.
5. Fox, J. (2002) A R and S-PLUS Companion to Applied Regression. Sage.
Does what it says.
6. Ligges, U. (2009) Programmieren mit R. Third edition. Springer.
In German(!)
7. Rizzo, M. L. (2008) Statistical Computing with R. CRC/Chapman &
Hall.
More computational different examples to the other books.
8. Braun, W. J. and Murdoch, D. J. (2007) A First Course in Statistical
Programming with R. CUP.
Detailed and well written, but at a rather low level. A bit redundant given
the above.

9. Maindonald J. and Braun, W. J. (2003) Data Analysis and Graphics

using R Second or third edition CUP.
Advanced statistical graphics
10. Spector, P. (2008) Data Manipulation with R. Springer
Especially for data manipulation.
11. Dalgaard, P. (2009) Introductory Statistics with R. Second Edition. Springer.
Probably redundant given the above.

Getting the Most out of the Class

Learning R has much in common with learning a natural language: its easy
to get going with a few simple phrases, though youll find some idiosyncrasies in the syntax, and occasional aspects are downright illogical. Once
youve mastered these few difficulties, the only barrier to fluency is the vast
vocabulary of R: even in the basic packages there are many commands which
you will never use or understand, but the more you learn the more elegantly
you will be able to express yourself. There is a smaller core of everyday language which we will focus on, and which you will be expected to understand
in exams and practical assessments.
These lecture notes are intended for reference, and will (by the end of the
course) contain sections on all the major topics we cover. Lectures will not
follow the notes exactly, so be prepared to take your own notes; the practical
classes will complement the lectures, and you can be examined on anything
we study in either.
Dont copy and paste the commands from this guide into R; you will find
it very hard to remember the details of the language and will have to look
everything up when you come to code something yourself.
Make sure you try the exercises, and understand the code involved in
each one; if something doesnt make sense, use Rs help functions, ask a
classmate, try using internet resources, or ask me for help (preferably in
that order). Some exercises are marked with an asterisk (*), which means
they are a little more advanced than is necessary for the class.
If you find any mistakes or omissions in these notes, Id be very grateful to
be informed.

1
1.1

Introduction
What R is good at

Statistics for relatively advanced users: R has thousands of packages, designed, maintained, and widely used by statisticians.
Statistical graphics: try doing some of our plots in Stata and you wont have
much fun.
Flexible code: R has a rather liberal syntax, and variables dont need to be
declared as they would in (for example) C++, which makes it very easy to
code in. This also has disadvantages in terms of how safe the code is.
Vectorization: R is designed to make it very easy to write functions which
are applied pointwise to every element of a vector. This is extremely useful
in statistics.
R is powerful: if a command doesnt exist already, you can code it yourself.

1.2

What R is not so good at

Statistics for non-statisticians: there is a steep learning curve, which puts

some people off. Try Stata, SAS or SPSS (if you must).
Numerical methods, such as solving partial differential equations; try Matlab.
Analytical methods, such as algebraically integrating a function. Try Mathematica or Maple.
Precision graphics, such as might be useful in psychology experiments. Try
Matlab.
Optimization. Though it does have some very easy to use methods built-in.
Low-level, high-speed or critical code; use C, C++, Java or similar. (However note that such code can be called from R to give the best of both
worlds.

1.3

General Properties

R makes it extremely easy to code complex mathematical or statistical procedures, though the programs may not run all that quickly. You can interface
R with other languages (C, C++, Fortran) to provide fast implementations
of subroutines, but writing this code (and making it portable) will typically
take longer. Where the advantage falls in this trade-off will depend upon

what youre doing; for most things you will encounter during your degree, R
is sufficiently fast.
R is open source and widely adopted by statisticians, biostatisticians, and
geneticists. There is a huge wealth of existing libraries so you can often
save time by using these, though it is sometimes easier to start from scratch
than to adapt someone elses function to meet your needs. Contributing new
packages to the central repository (CRAN) is easy: even your lecturer has
managed it. As a result, R packages are not build to very high standards
(but see Bioconductor).
R is portable, and works equally well on Windows, OS X and Linux.

1.4

Interfaces

For Windows and OS X, the standard R download comes with an R GUI,

which is adequate for simple tasks. You can also run R from the command
line in any operating system.
There are a number of more powerful interfaces which you may like to try.
Heres a few.
RStudio. Very popular, with a nice interface and well thought out, especially for more advanced usage: can be a bit buggy, so make sure you
update it regularly. Available on all platforms.
Emacs with ESS. (Emacs Speaks Statistics) is available on all platforms,
and is very powerful when you get used to it. Has a habit of freezing
in my experience, though.
TinnR. Alternative Windows interface.
I intend to demonstrate a few of these different approaches during class.

Basic Arithmetic and Objects

R has a command line interface, and will accept simple commands to it. This
is marked by a > symbol, called the prompt. If you type a command and
press return, R will evaluate it and print the result for you.
> 6 + 9
[1] 15
> x <- 15
> x - 1
[1] 14
The expression x <- 15 creates a variable called x and gives it the value 15.
This is called assignment; the variable on the left is assigned to the value
on the right. The left hand side must contain only contain a single variable.

> x + 4 <- 15

# doesn't work

Assignment can also be done with = (or ->).

> x = 5
> 5*x -> x
> x
[1] 25
The operators = and <- are identical, but many people prefer <- because it
is not used in any other context, but = is, so there is less room for confusion.

2.1

Vectors

The key feature which makes R very useful for statistics is that it is vectorized. This means that many operations can be performed point-wise on a
vector. The function c() is used to create vectors:

> x <- c(1, -1, 3.5, 2)

> x
[1]

1.0 -1.0

3.5

2.0

Then if we want to add 2 to everything in this vector, or to square each

entry:
> x + 2
[1] 3.0 1.0 5.5 4.0
> x^2
[1]

1.00

1.00 12.25

4.00

This is very useful in statistics:

> sum((x - mean(x))^2)
[1] 10.69
Exercise 2.1. The weights of five people before and after a diet programme
are given in the table.
Before
After

78
67

72
65

78
79

79
70

105
93

Read the before and after values into two different vectors called before
and after. Use R to evaluate the amount of weight lost for each participant.
What is the average amount of weight lost?
*Exercise 2.2. How would you write a function equivalent to sum((x - mean(x))^2)
in a language like C or Java?
Some useful vectors can be created quickly with R. The colon operator is
used to generate integer sequences
> 1:10
[1]

9 10

> -3:4
[1] -3 -2 -1

> 9:5
[1] 9 8 7 6 5
More generally, the function seq() can generate any arithmetic progression.

> seq(from=2, to=6, by=0.4)

[1] 2.0 2.4 2.8 3.2 3.6 4.0 4.4 4.8 5.2 5.6 6.0
> seq(from=-1, to=1, length=6)
[1] -1.0 -0.6 -0.2

0.2

0.6

1.0

Sometimes its necessary to have repeated values, for which we use rep()
> rep(5,3)
[1] 5 5 5
> rep(2:5,each=3)
[1] 2 2 2 3 3 3 4 4 4 5 5 5
> rep(-1:3, length.out=10)
[1] -1

3 -1

We can also use Rs vectorization to create more interesting sequences:

> 2^(0:10)
[1]

16

32

64

128

256

512 1024

> 1:3 + rep(seq(from=0,by=10,to=30), each=3)

[1]

3 11 12 13 21 22 23 31 32 33

The last example demonstrates recycling, which is also an important part of

vectorization. If we perform a binary operation (such as +) on two vectors
of different lengths, the shorter one is used over and over again until the
operation has been applied to every entry in the longer one. If the longer
length is not a multiple of the shorter length, a warning is given.
> 1:10 * c(-1,1)
[1] -1

2 -3

4 -5

6 -7

8 -9 10

> 1:7 * 1:2

Warning:
length
[1]

longer object length is not a multiple of shorter object

5 12

Exercise 2.3. Create the following vectors in R using seq() and rep().
(i) 1, 1.5, 2, 2.5, . . . , 12
(ii) 1, 8, 27, 64, . . . , 1000.
1
(iii) 1, 21 , 31 , 14 , . . . , 100
.

(iv) 1, 0, 3, 0, 5, 0, 7, . . . , 0, 49.
P
(v) 1, 3, 6, 10, 15, . . . , ni=1 i, . . . , 210 [look up ?cumsum].
(vi)

1, 2, 2, 3, 3, 3, 4, . . . , 9, 10, . . . , 10. [Hint: type ?seq, and read about

| {z }
10 times

the times argument.]

Exercise 2.4. The ith term in the Taylor expansion of log(1+x) is (1)i+1 xi /i.
Create a vector containing the first 100 terms for x = 0.5. [Write out the
first few entries by hand if that helps.]
Let
rn (x) = log(1 + x)

n
X
(1)i+1 xi
i=1

Evaluate rn (1) for n = 10, 100, 1000, . . . , 106 .

9

2.2

Subsetting

Its frequently necessary to extract some of the elements of a larger vector.

In R you can use square brackets to select an individual element or group of
elements:
> x <- c(5,9,2,14,-4)
> x[3]
[1] 2
> # note indexing starts from 1
> x[c(2,3,5)]
[1]

2 -4

> x[1:3]
[1] 5 9 2
> x[3:length(x)]
[1]

2 14 -4

There are two other methods for getting subvectors. The first is using a
logical vector (i.e. containing TRUE and FALSE) of the same length:
> x > 4
[1]

TRUE

TRUE FALSE

TRUE FALSE

> x[x > 4]

[1]

9 14

or using negative indices to specify which elements should not be selected:

10

> x[-1]
[1]

2 14 -4

> x[-c(1,4)]
[1]

2 -4

(Note that this is rather different to what other languages such as C or

Python would interpret negative indices to mean.)
Exercise 2.5. The built-in vector LETTERS contains the uppercase letters
of the alphabet. Produce a vector of (i) the first 12 letters; (ii) the odd
numbered letters; (iii) the (English) consonants.

2.3

Logical Operators

As we see above, the comparison operator > returns a logical vector indicating whether or not the left hand side is greater than the right hand side.
Here we demonstrate the other comparison operators:
> x <= 2

# less than or equal to

[1] FALSE FALSE

> x == 2

TRUE FALSE

# equal to

[1] FALSE FALSE

TRUE FALSE FALSE

> x != 2

# not equal to

[1]

TRUE FALSE

TRUE

TRUE

TRUE

TRUE

Note the double equals sign ==, to distinguish between assignment and comparison.
We may also wish to combine logical vectors. If we want the elements of x
within a range, we can use the following:

11

> (x > 0) & (x < 10)

[1]

TRUE

TRUE

# 'and'

TRUE FALSE FALSE

The & operator does a pointwise and comparison between the two sides.
Similarly, the vertical bar | does pointwise or, and the unary ! operator
performs negation.
> (x == 5) | (x > 10)
[1]

TRUE FALSE FALSE

TRUE FALSE

> !(x > 5)

[1]

TRUE FALSE

TRUE FALSE

TRUE

Exercise 2.6. The function rnorm() generates normal random variables.

For instance, rnorm(10) gives a vector of 10 i.i.d. standard normals. Generate 20 standard normals, and store them as x. Then obtain subvectors
of
(i) the entries in x which are less than 1;
(ii) the entries between 12 and 1;
(iii) the entries whose absolute value is larger than 1.5.

2.4

Character Vectors

As you might have noticed in the exercise above, vectors dont have to
contain numbers. We can equally create a character vector, in which
each entry is a string of text. Strings in R are contained within double
quotes ":
> x <- c("Hello", "how do you do", "lovely to meet you", 42)
> x
[1] "Hello"
[4] "42"

"how do you do"

12

"lovely to meet you"

Notice that you cannot mix numbers with strings: if you try to do so the
number will be converted into a string. Otherwise character vectors are
much like their numerical counterparts.
> x[2:3]
[1] "how do you do"

"lovely to meet you"

> x[-4]
[1] "Hello"

"how do you do"

"lovely to meet you"

> c(x[1:2], "goodbye")

[1] "Hello"

2.5

"how do you do" "goodbye"

Matrices

Matrices are much used in statistics, and so play an important role in R. To

create a matrix use the function matrix(), specifying elements by column
first:
> matrix(1:12, nrow=3, ncol=4)
[,1] [,2] [,3] [,4]
[1,]
1
4
7
10
[2,]
2
5
8
11
[3,]
3
6
9
12
This is called column-major order. Of course, we need only give one of
the dimensions:
> matrix(1:12, nrow=3)
unless we want vector recycling to help us:
> matrix(1:3, nrow=3, ncol=4)
[,1] [,2] [,3] [,4]
13

[1,]
[2,]
[3,]

1
2
3

1
2
3

1
2
3

1
2
3

Sometimes its useful to specify the elements by row first

> matrix(1:12, nrow=3, byrow=TRUE)
There are special functions for constructing certain matrices:
> diag(3)

[1,]
[2,]
[3,]

[,1] [,2] [,3]

1
0
0
0
1
0
0
0
1

> diag(1:3)

[1,]
[2,]
[3,]

[,1] [,2] [,3]

1
0
0
0
2
0
0
0
3

> 1:5 %o% 1:5

[1,]
[2,]
[3,]
[4,]
[5,]

[,1] [,2] [,3] [,4] [,5]

1
2
3
4
5
2
4
6
8
10
3
6
9
12
15
4
8
12
16
20
5
10
15
20
25

The last operator performs an outer product, so it creates a matrix with

(i, j)-th entry xi yj . The function outer() generalizes this to any function
f on two arguments, to create a matrix with entries f (xi , yj ). (More on
functions later.)
> outer(1:3, 1:4, "+")
[,1] [,2] [,3] [,4]
14

[1,]
[2,]
[3,]

2
3
4

3
4
5

4
5
6

5
6
7

Matrix multiplication is performed using the operator %*%, which is quite

distinct from scalar multiplication *.
> A <- matrix(c(1:8,10), 3, 3)
> x <- c(1,2,3)
> A %*% x
# matrix multiplication

[1,]
[2,]
[3,]

[,1]
30
36
45

> A*x

[1,]
[2,]
[3,]

# NOT matrix multiplication

[,1] [,2] [,3]

1
4
7
4
10
16
9
18
30

Standard functions exist for common mathematical operations on matrices.

> t(A)

[1,]
[2,]
[3,]

# transpose

[,1] [,2] [,3]

1
2
3
4
5
6
7
8
10

> det(A)

# determinant

[1] -3
> diag(A)
[1]

# diagonal

5 10
15

> solve(A)

# inverse

[,1]
[,2] [,3]
[1,] -0.6667 -0.6667
1
[2,] -1.3333 3.6667
-2
[3,] 1.0000 -2.0000
1
Exercise 2.7. Construct the matrix

1
2
3
2
6
B= 4
3 1 3
Show that B B B is a scalar multiple of the identity matrix, and find
the scalar.
Matrices can be subsetted much the same way as vectors, although of course
they have two indices. Row number comes first:
> A[2,1]
[1] 2
> A[2,2:ncol(A)]
[1] 5 8
> A[,1:2]

[1,]
[2,]
[3,]

# blank indices give everything

[,1] [,2]
1
4
2
5
3
6

> A[c(),1:2]

# empty indices give nothing!

[,1] [,2]
Notice that, where appropriate, R automatically reduces a matrix to a vector
or scalar when you subset it. You can override this using the optional drop
argument.
16

> A[2,2:ncol(A),drop=FALSE]

[1,]

# returns a matrix

[,1] [,2]
5
8

You can stitch matrices together using the rbind() and cbind() functions.
These employ vector recycling:
> cbind(A, t(A))

[1,]
[2,]
[3,]

[,1] [,2] [,3] [,4] [,5] [,6]

1
4
7
1
2
3
2
5
8
4
5
6
3
6
10
7
8
10

> rbind(A, 1, 0)

[1,]
[2,]
[3,]
[4,]
[5,]

[,1] [,2] [,3]

1
4
7
2
5
8
3
6
10
1
1
1
0
0
0

Exercise 2.8. Construct the following matrices:

(a)



1 3 5 7
2 4 6 8

(b)

1 1 1 1
1 1 1 1

..
..
..
.
.
.
1 1 1 1

(dimensions 15 10).

(c) The 5 15 matrix with three 1s in shifting positions:

1 1 1 0 0 0 0
0 0 0 1 1 0 0

(dimensions 5 15).
.. .. ..
..
. . .
.
0 0 0 0 0 1 1
17

[Hint: use column subsetting.]

(d)

1 2 3
2 3 4

3 4 5

..
..
.
.

9 10
10 11

9 10
10 11

..
.
;

..
.
17

17 18
17 18 19

[Look at the outer() function.]

(e)

1
2

..
.
9

9
1

..
.

;
..
.
6

6 7
1 6
7 8
2
3

3
4
..
.
4

[The modular arithmetic operator %% may be useful here.]

(f)


I5 1
0 I6
where Ik is the k k-identity matrix, and 1 and 0 are matrices with all
entries 1 and 0 respectively.
Exercise 2.9. Solve the following system of simultaneous equations using
matrix methods.
a + 2b + 3c + 4d + 5e = 5
2a + 3b + 4c + 5d + e = 2
3a + 4b + 5c + d + 2e = 5
4a + 5b + c + 2d + 3e = 10
5a + b + 2c + 3d + 4e = 11

Dont just create your matrix by hand!

Exercise 2.10. In this section weve seen that the behaviour of the function
diag() depends upon its inputs. Can you think of some examples where
this might cause a problem?

18

2.6

Lists

Other than vectors and matrices, the main object for holding data in R is a
list1 . These are a bit like vectors, except that each entry can be any other
R object, even another list.
> x <- list(1:3, TRUE, "Hello", list(1:2, 5))
Here x has 4 elements: a numeric vector, a logical, a string and another list.
We can select an entry of x with double square brackets:
> x[[3]]
[1] "Hello"
To get a sub-list, use single brackets:
> x[c(1,3)]
[[1]]
[1] 1 2 3
[[2]]
[1] "Hello"
Notice the difference between x[[3]] and x[3].
We can also name some or all of the entries in our list, by supplying argument names to list():
> x <- list(y=1:3, TRUE, z="Hello")
> x
$y
[1] 1 2 3
[[2]]
[1] TRUE
$z
[1] "Hello"
1

Technically speaking, lists are also a kind of vector in R, but not every object in them
has to have the same type; ordinary logical, numeric or character vectors are known as
atomic vectors.

19

Notice that the [[1]] has been replaced by $y, which gives us a clue as to
how we can recover the entries by their name. We can still use the numeric
position if we prefer:
> x$y
[1] 1 2 3
> x[[1]]
[1] 1 2 3
The function names() can be used to obtain a character vector of all the
names of objects in a list.
> names(x)
[1] "y" ""

"z"

Youve seen most standard R objects now: almost all the more complicated
ones are just lists! Well see this in the next section.

20

Data

R comes with many datasets built-in, particularly in the MASS package. A

package is a collection (or library) of functions, datasets, and other objects;
most packages are not loaded automatically, so you have to do it yourself:
> library(MASS)
You can now access various datasets from this package. Try looking at the
dataset called hills.
> head(hills)
To find out what the data represent, use the help function ?hills.

3.1

Data Frames

The object hills is something called a data frame. A data frame is a

series of records represented by rows (in this case one for each race), each
containing values in several fields (in this case dist, climb, time).
You can check that hills is a data frame by inspecting its class(es)
> class(hills)
[1] "data.frame"
or more reliably by using an is() command
> is(hills, "data.frame")
[1] TRUE
Well talk more about classes later in the course.
Data frames share many of the characteristics of matrices. We can select
rows or columns in the same way:
> hills[3,]

Craig Dunain

dist climb time

6
900 33.65
21

> hills[hills$dist >= 12,]

Bens of Jura
Lairig Ghru
Seven Hills
Two Breweries
Moffat Chase

dist climb
time
16 7500 204.62
28 2100 192.67
14 2200 98.42
18 5200 170.25
20 5000 159.83

However, they also behave like lists indexed by the columns:

> hills$time
[1] 16.08
[11] 192.67
[21] 15.95
[31] 85.58

48.35 33.65
43.05 65.00
27.90 47.63
32.38 170.25

45.60 62.27
44.13 26.93
17.93 18.68
28.10 159.83

73.22 204.62
72.25 98.42
26.22 34.43

36.37
78.65
28.57

29.75
17.42
50.50

The truth is that, like almost all complicated objects in R, data frames
are lists with some additional structure. Formally speaking, they are not
matrices, but they do behave similarly in certain circumstances.
Exercise 3.1. How do the results of the following commands differ from
what we would expect if hills were a matrix?
>
>
>
>

hills[1,]
hills[3]
hills %*% c(1,2,4)
mean(hills)

3.2

Manipulating Data using with()

We often want to use functions on the columns of a data frame, and it quickly
becomes inconvenient to repeatedly type (for example) hills$ before every
such event. For example, the command below will give a scatter plot of the
race times against climbs, amongst only those races less than 10 miles long.

> plot(hills$climb[hills$dist < 10], hills$time[hills$dist < 10])

22

39.75
32.57
20.95

The with() function allows us to refer to the names of objects in a data

frame (or, in fact, any list) without having to keep referring to the data
frame itself. For example, the command above becomes
> with(hills, plot(climb[dist < 10], time[dist < 10]))
If you just type climb or dist on their own, R wont know what object
youre referring to. Technically with() alters the scope of the expression
being evaluated (i.e. the code given in the second argument) so that it can
see the columns of the data frame as objects. Well learn a bit more about
scope when we talk about functions later on.
Exercise 3.2. Using with(), find the mean of the average speeds (in miles
per hour) for races which are between 5 and 10 miles long

3.3

Creating Data Frames

The command data.frame() is used to create a data frame, each argument

representing a column.
> books <- data.frame(author=c("Ripley", "Cox", "Snijders", "Cox"),
+
year=c(1980, 1979, 1999, 2006),
+
publisher=c("Wiley", "Chapman", "Sage", "CUP"))
> books

1
2
3
4

author
Ripley
Cox
Snijders
Cox

year publisher
1980
Wiley
1979
Chapman
1999
Sage
2006
CUP

Exercise 3.3. (a) Create a small data frame representing a database of

films. It should contain the fields title, director, year, country, and
at least three films.
(b) Create a second data frame of the same format as above, but containing
just one new film.
(c) Merge the two data frames using rbind().
(d) Try sorting the titles using sort(): what happens?

23

3.4

Factors

There are two main types of data which you will encounter this year: numerical and categorical. Weve seen how to create numerical vectors already.
Suppose we have the heights of 100 individuals, the first 50 male and the
rest female.
>
>
>
>

set.seed(1442)
# fixes the random numbers
height = round(rnorm(100, mean=rep(c(170,160),each=50), sd=10))
sex = rep(c("M", "F"), each=50)
head(sex)

[1] "M" "M" "M" "M" "M" "M"

We can tell R to treat sex as a categorical variable:
> Sex = as.factor(sex)
> head(Sex)
[1] M M M M M M
Levels: F M
Note that it is displayed slightly differently. The new variable Sex is called a
factor; a factor is a categorical variable which takes various discrete levels,
in this case M and F for male and female.
R knows to do sensible things with factors:
> plot(Sex, height)

24

200
190
180
170
160
150
140

What happens if you try to plot sex against height instead? The distinction
between categorical and non-categorical data is especially important if we
have numbered groups.
The information in a factor is stored as a vector of integers:
> as.integer(Sex)
[1] 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
[36] 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
[71] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Just as a data frame is really a list, a factor is really a vector of integers
(for levels) together with some extra information giving each level a names.
The additional information is contained within a list of attributes. You
can view this list directly.
> attributes(Sex)
$levels
[1] "F" "M"

25

$class
[1] "factor"
The attributes in this case are its class (youll see this in many objects)
and a vector of the level names. The class tells R that this object should be
treated as a factor so that, for example, it will be displayed to you in the
right way.
You may find that sometimes data are stored as a factor when you dont
want them to be (see the exercise in the previous section). You can turn a
factor back in to a character vector easily enough:
> as.character(Sex)
[1]
[18]
[35]
[52]
[69]
[86]

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

"M"
"M"
"M"
"F"
"F"
"F"

Exercise 3.4. (a) Sample the numbers {1, 2, 3} uniformly with replacement 50 times; use this to create a factor with levels Yes, No and Maybe.
(b) Create a subvector by removing the Maybe entries from the factor above.
What levels does the new factor have?
(c) Use the command droplevels() to remove the level Maybe.
Exercise 3.5. Take a look at the birthwt data from the MASS package.
How is race stored in these data? Is this sensible?
Define a factor based on race:
> Race = factor(birthwt$race)
Compare the effect of the commands summary(), plot() and mean() on
each of Race and birthwt$race. Which do you find more useful?

3.5

Row and Column Names

The labels above and to the left of the values in hills are not part of the
data itself, but can be accessed:

26

"M"
"M"
"M"
"F"
"F"

"M"
"M"
"F"
"F"
"F"

> names(hills)
[1] "dist"

"climb" "time"

> row.names(hills)
[1]
[4]
[7]
[10]
[13]
[16]
[19]
[22]
[25]
[28]
[31]
[34]

"Greenmantle"
"Ben Rha"
"Bens of Jura"
"Traprain"
"Lomonds"
"Cairngorm"
"Black Hill"
"Meall Ant-Suidhe"
"N Berwick Law"
"Largo Law"
"Ben Nevis"
"Cockleroi"

"Carnethy"
"Ben Lomond"
"Cairnpapple"
"Lairig Ghru"
"Cairn Table"
"Seven Hills"
"Creag Beag"
"Half Ben Nevis"
"Creag Dubh"
"Criffel"
"Knockfarrel"
"Moffat Chase"

"Craig Dunain"
"Goatfell"
"Scolty"
"Dollar"
"Eildon Two"
"Knock Hill"
"Kildcon Hill"
"Cow Hill"
"Burnswark"
"Acmony"
"Two Breweries"

As we saw above, in a data frame the column names can be used for indexing
(e.g. hills$time); the row names cannot be used in this way.
This additional information is stored as attributes, which are in a separate
list2 attached to the object hills:
> attributes(hills)
We could add an attribute to hills if we wanted:
> attributes(hills) <- c(attributes(hills), list(type="races"))
> attributes(hills)
Note that type(hills) doesnt access hills$type, the most important
attributes such as names and class happen to have functions named after
them, which can be used to extract relevant information.

3.6

Reading in Data

Whilst R provides many interesting datasets, it is often necessary to load

data externally. The main commands for doing this are read.table() and
read.csv().
2

Actually theyre not stored as a list (see ?attributes), but they behave very similarly.

27

Here is an example using the smoking.dat dataset, available on the course

website.
> dat <- read.table("smoking.dat", header=TRUE)
> head(dat)

1
2
3
4
5
6

STATE
AL
AZ
AR
CA
CT
DE

CIG
18.20
25.82
18.24
28.60
31.10
33.60

BLAD
2.90
3.52
2.99
4.46
5.11
4.78

LUNG
17.05
19.80
15.98
22.07
22.83
24.55

KID
1.59
2.75
2.02
2.66
3.35
3.36

LEUK
6.15
6.61
6.94
7.06
7.20
6.45

> class(dat)
[1] "data.frame"
What happens if header=TRUE is omitted?
When you specify the file name, be sure to use the double quotes (") around
it. You also need to give the correct path to the file. R will automatically
look for the file in its working directory. You can check what this is:
> getwd()
[1] "/data/redcrest/evans/Dropbox/Teaching/R Programming/2014"
Then if your file is in a subfolder called files, you need to write (for example)
> dat <- read.table("files/smoking.dat", header=TRUE)
In some systems you can use file.choose() to get the full path to a file.
In particular this works well on R GUI for Windows or OS X. For example:

> dat <- read.table(file.choose(), header=TRUE)

will bring up a window for you to choose the file from.
*Exercise 3.6. Look at the documentation for read.table(). Use the
function to read in only lines 11 to 20 (Indiana up to Minnesota).

28

Functions

Everything which is done in R is done by functions. A function in a programming langauge is much like its mathematical equivalent: it has some
inputs called arguments, and an output called the return value. In R a
function can only return a single object. If you type a functions name at
the console, you can see its structure:
> setdiff
function (x, y)
{
x <- as.vector(x)
y <- as.vector(y)
unique(if (length(x) || length(y))
x[match(x, y, 0L) == 0L]
else x)
}
<bytecode: 0x2a49e98>
<environment: namespace:base>
There are two important parts to the function: the signature, which in
this case is function(x, y), and the body, which is the code between the
curly brackets. Broadly speaking, when a function is called, it takes the
information in the arguments, applies the code in the body to them, and
then spits out the final expression in the function. In this case thats the
complex looking expression unique( ).

4.1

Arguments

You can look at the arguments for a function by typing args:

> args(setdiff)
function (x, y)
NULL
Arguments are a little complicated in R. Youll notice that they have a
name: the arguments of setdiff() are called x and y. However, you dont
usually have to specify an argument by name, because arguments also have
a position:

29

> a <- c(1,4,5,7)

> b <- c(1,2,5,9)
> setdiff(a, b)
# everything in 'a' and not in 'b'
[1] 4 7
It assumes that the first argument supplied should be x, and the second y.
You can override this by specifying the name, and then the order doesnt
matter:
> setdiff(y=b, x=a)
[1] 4 7
> setdiff(b, a)
[1] 2 9
If you specify some of the argument names but not all, then it will use the
ordering to deduce the others.
> setdiff(y=b, a)

# 'a' must be the argument 'x'

[1] 4 7
Most functions dont require all of their arguments to be specified.
> x <- rnorm(10)
> y <- x + rnorm(10)
> lm(y ~ x)

Call:
lm(formula = y ~ x)
Coefficients:
(Intercept)
0.0212

x
0.7415

> args(lm)
30

function (formula, data, subset, weights, na.action, method = "qr",

model = TRUE, x = FALSE, y = FALSE, qr = TRUE, singular.ok = TRUE,
contrasts = NULL, offset, ...)
NULL
The function lm() (which fits a linear model) only requires the single argument formula to run; the other arguments are optional: some of them have
default values which are shown in the signature (such as model = TRUE),
whereas others simply alter the behaviour of the function when specified
(such as data).
R also uses partial matching for arguments, so as long as you give enough
of the arguments name to make in unambiguous which one you mean, it
will work:
> ## these two will do the same thing
> lm(form = y ~ x)
> lm(formula = y ~ x)

4.2

Writing Functions

To define your own function you just have to construct something in the
same format as above:
> square = function(x) {
+
x^2
+ }
> square(4)
[1] 16
Objects which are created inside a function do not exist outside it:
> mean2 <- function(x) {
+
n <- length(x)
+
sum(x)/n
+ }
>
> mean2(1:10)
[1] 5.5
31

> n
Error:

object n not found

Clearly an object called n was used inside the function above, but it was
only inside the functions namespace. Most functions in R do not have side
effects: they return a value, but do not change any of the objects which
you can reach at the console. In order to use a function, you usually have
to assign its output to something.
> x <- mean2(1:10)
> x
[1] 5.5
Exercise 4.1. The logit function is defined as


x
logit(x) = log
,
0 < x < 1.
1x
Write an R function in one argument to implement this. How does your
function behave for values of x such as 0, 1, or 2?
Exercise 4.2. Recall that the Taylor expansion of log(1 + x) is
log(1 + x) =

X
i=1

(1)i+1

xi
i

Write a function with arguments x and n, which calculates the Taylor approximation to log(1 + x) using n terms.
How many terms do you need to get within 106 of the correct solution
when x = 0.99?
Exercise 4.3. Given real vectors x, y of length n, the least squares slope
(, )T is given by
P
(xi x
)(yi y)
= iP
)2
i (xi x
= y x
.
Write a function which takes two arguments, x and y, and returns a vector
of length 2 containing and . Verify that your function gives the correct
answer using Rs built-in function lm() [the syntax is lm(y~x)].
32

4.3

for() Loops

The most common way to execute a block of code multiple times is with a
for() loop. Whats going on in the code below?
> factorial2 = function(n) {
+
out = 1
+
+
+

for (i in 1:n) {
out = out*i
}

+
out
+ }
> factorial2(10)
[1] 3628800
You may have seen for() loops in other languages. The syntax in R is for
(i in x) for some vector (or list) x, where i will take each value in x. Most
commonly, x is a vector of the first n natural numbers.
i is a dummy variable, and can be called whatever you like, though it retains
its value outside the loop.
> for (sillyname in 1:4) print(sillyname)
[1]
[1]
[1]
[1]

1
2
3
4

> sillyname
[1] 4
Exercise 4.4. Write a function to perform matrix-vector multiplication. It
should take a matrix A and a vector b as arguments, and return the vector
Ab. Use two loops to do this, rather than %*% or any vectorization.
I generally recommend using seq_len() or seq_along() in for() loops,
because it always behaves the way you want (and runs quicker than seq()):

33

> n = 0
> 1:n # not a sequence of length n=0
[1] 1 0
> seq(n)

# ditto

[1] 1 0
> seq_len(n)

# better!

integer(0)
> for (i in seq_len(n)) print(i)
> for (i in seq(n)) print(i)
[1] 1
[1] 0

4.4

Conditional Code

Its extremely common to need code to do different things depending upon

the number given to it. Lets write a short function to find the absolute
value of a number.
> abs2 = function(x) {
+
if (x < 0) out = -x
+
else out = x
+
out
+ }
> abs2(-4)
[1] 4
The if() function will only execute the code which follows if the expression
in parentheses evaluates to TRUE 3 . When the expression is FALSE the code
3

Actually, any non-zero number will act the same way as TRUE, but its safer to only
use logicals.

34

following the else statement will be used instead. There is no need to

include an else, in which case the program will do nothing if the condition
in FALSE.
Take care not to allow the logical expression following the if() to be a
vector, or R will spit out a warning.
> abs2(c(1,-3))

# won't work properly.

Warning: the condition has length > 1 and only the first element
will be used
[1]

1 -3

If you only need to assign a single value (whether a number, logical, or

string) based on a condition, you can use ifelse():
> ifelse(TRUE, 94, "hello")
[1] 94
> ifelse(FALSE, 94, "hello")
[1] "hello"

4.5

while() loops

Here is a short function to check whether an integer is prime.

> isPrime = function(n) {
+
i = 2
+
if (n < 2) return(FALSE)
+
while (i < sqrt(n)) {
+
if (n %% i == 0) return(FALSE)
+
i = i+1
+
}
+
return(TRUE)
+ }
> isPrime(10)

35

[1] FALSE
> isPrime(37)
[1] TRUE
This illustrates several points. First, we dont need to wait until reaching
the end of a function to return a value; we can use the return keyword
instead.
The other feature is the while() loop. This will keep running until the
expression in the parenthesis becomes false.

4.6

Writing for Speed

Loops are slow in R! It is usually much better to try to vectorize a

function rather than calling it lots of times with a loop. Of course, for
squaring this is already built in to R.
> system.time(for (i in 1:1e6) i^2)
user
0.177

system elapsed
0.012
0.196

> system.time(seq_len(1e6)^2)
user
0.025

system elapsed
0.001
0.026

We can write a second function to do matrix-vector multiplication (see Exercise 4.4), but this time replacing the inner loop by a vectorized function
to take dot products.
> mult2 = function(A, b) {
+
n1 = nrow(A)
+
n2 = ncol(A)
+

out = numeric(n1)

+
+

for (i in 1:n1) {
out[i] = sum(A[i,]*b)
36

+
}
+
out
+ }
Now suppose we create a large matrix, and look at the difference in timing.

> A = matrix(rnorm(1e6), 1e3, 1e3)

> b = rnorm(1e3)
> system.time(mult(A,b))
user
1.498

system elapsed
0.039
1.572

> system.time(mult2(A,b))
user
0.013

system elapsed
0.005
0.018

> system.time(colSums(t(A)*b))
user
0.012

# can you see why this works?

system elapsed
0.000
0.012

> system.time(A %*% b)

user
0.004

system elapsed
0.000
0.004

The difference is dramatic. The moral of this is that its usually better to
use a built-in function, and almost always better to vectorize. The reason
%*% is so fast is that R calls underlying FORTRAN routines which have been
optimized over decades.

4.7

Recursion

Functions can recurse, which means they call themselves; here is a function
which calculates the entry Fn in the Fibonacci sequence with F0 = F1 = 1,
and Fk = Fk1 + Fk2 for k 2:

37

> fib = function(n) {

+
if (n < 2) return(1)
+
else return(fib(n-1) + fib(n-2))
+ }
> fib(10)
[1] 89
Recursion can be very slow though, so try to avoid it if possible. You can
also use Recall() instead of writing the functions name in order to recurse.
Exercise 4.5. The number of moves required to complete a Towers of Hanoi
puzzle with k pieces is Hk = 2Hk1 + 1 if k > 1, with H1 = 1. Write a
recursive function to evaluate Hk .
Exercise 4.6. Write a function to calculate a Fibonacci sequence using a
loop instead of a recursion. Compare the execution time to fib() (above)
for calculating the 30th term in the sequence using system.time().

4.8

Scope

When a function is called, the code inside it is run in a separate environment to the code you run directly at the command line. This means its
possible for a variable inside a function to have the same name as something
at the command line without causing any problems:
>
>
+
+
+
>

x <- 3
f = function(y) {
x <- 5
x + y
}
f(4)

[1] 9
> x

# still the same value as before

[1] 3
However, if a function fails to find a variable withing its own environment,
then it will look to the parent environment for such a value: this is either
the function which called the current function, or the global environment
(i.e. the one you use at the command line).
38

>
>
+
+
>

x <- 3
g = function(y) {
x + y
}
g(4)

[1] 7
Whilst this sort of behaviour can sometimes seem helpful, it is much better
to avoid writing confusing code like this. You are strongly recommended to
write functions which only require the information in their own arguments
to run.
This is the same principle used by the functions with() and subset():
they create an environment for the data frame (or list) you give as their
first argument; if any names supplied dont match columns within the data
frame, R searches in the global environment:
> conv = 1.609
> with(hills, mean(dist/conv))

# where do conv and dist come from?

[1] 4.679
Exercise 4.7. What will happen if I create an object called dist before
running the commands above?

39

Graphics

Graphics and data visualization are an integral part of statistics, and R

makes it easy to produce common plots quickly, as well as giving a powerful interface for more esoteric output. The basic command is the generic
function plot(). This will try to do the most sensible thing for the kind of
data you provide.
We have already seen that plot(x) and plot(x,y) will produce different
plots depending upon the class of the inputs. This is very partially summarized in the following table:

x
numeric
factor

(missing)
series plot
bar chart

y
numeric
scatter plot
box plots

factor
spine plot
spine plot

In fact there are many more plotting methods, most of which you will rarely
use.

5.1

One dimensional plots

For graphical summaries of one dimensional data we have already seen boxplots and (in the practical) a time series for random walks. Among the most
useful is the histogram:
> hist(nlschools$lang, breaks=25, col=2)

40

100
0

50

Frequency

150

Histogram of nlschools$lang

10

20

30

40

50

60

nlschools$lang

Note that the optional argument breaks chooses (approximately) how many
bins the histogram should have, and col alters the colour of the bars. Of
course, all plots should have properly labelled axes and a title, which can
be easily added.
> hist(nlschools$lang, breaks=25, col=2, xlab="Score",
+
main="Language test scores of Dutch 8th grade pupils")
Even the simple plot command for a single numeric vector comes with a
large range of options.
> x = cumsum(rnorm(250))
> plot(x, type="l", col=3)

41

0
5
10
20

15

50

100

150

200

250

Index

Try this with type="b" or type="h" and see what happens. You can only
find out about a few of the graphics options with the documentation for
plot(). Try looking at ?par to find the real detail.

5.2

Adding to Plots

Consider the following simple scatter plot, augmented with the line y = x.

>
>
>
>

x = rnorm(300)
y = x + rnorm(300)
plot(x,y, pch=20, col=4, cex=0.5)
abline(a=0, b=1, lty=4, lwd=1.5)

42

4
0
4

y=x
line of best fit

What does cex=0.5 do?

The function abline() plots lines of the form y = a + bx where a and b are
specified. As you can see, the width and appearance of the line is adjusted
with the options lty (line type), lwd (line width) and col (colour).
You can also use abline() in conjunction with output from a simple linear
model. Adding the following gives the red line:
> abline(lm(y ~ x), col=2)

5.3

Legends

The legend() command can be used to provide additional information

about your plots. The basic syntax is
> legend(x=-4, y=4, legend=c("y=x","line of best fit"),
+
lty=c(4,1), lwd=c(1.5,1), col=1:2)
where (x, y) is the top-left hand corner of the box, legend is a character
vector of annotations, and the other options are used to describe what to
display. Many other options are available via the help file.
43

5.4

Formulae and Boxplots

R contains a special object type called a formula which can be used to

represent statistical models compactly. Try typing some algebraic expression
at the console separated by the tilde operator ~ (this is on the # key on a
UK keyboard, and left of 1 on a US keyboard).
> x ~ a + b*c
x ~ a + b * c
The formula object can be used to express relationships between variables
under the convention that the left-hand side is modelled by the right-hand
side.
This can be used, for example, when producing plots:
> data(genotype)
> head(genotype)

1
2
3
4
5
6

Litter Mother
Wt
A
A 61.5
A
A 68.2
A
A 64.0
A
A 65.0
A
A 59.7
A
B 55.0

> boxplot(Wt ~ Litter, data=genotype)

44

70
65
60
55
50
45
40
35

The function interprets the formula as requiring that the left-hand side be
summarized in a way which is broken down by the right. Note that Wt and
Litter are contained within genotype, and are not recognized at the console4 , but the argument data=genotype ensures that the boxplot() function knows where to look for genotype$Wt.

5.5

Lattice Graphics

The plots above are all found in the base package of R, which is to say that
they are all preloaded functions. A very popular and powerful extension
to Rs graphics capabilities is made using the package lattice. The range
of plots which can be produced even using lattices default methods is
staggering, and we will show only a few small examples here.
The basic command is xyplot(), whose first argument is usually a formula.

> library(lattice)
> head(crabs)
sp sex index
4

FL

RW

CL

CW

BD

That is, they are not in the global environment

45

1
2
3
4
5
6

B
B
B
B
B
B

M
M
M
M
M
M

1 8.1
2 8.8
3 9.2
4 9.6
5 9.8
6 10.8

6.7
7.7
7.8
7.9
8.0
9.0

16.1
18.1
19.0
20.1
20.3
23.0

19.0
20.8
22.4
23.1
23.0
26.5

7.0
7.4
7.7
8.2
8.2
9.8

> form <- log(FL) ~ log(RW) | sp*sex

> form
log(FL) ~ log(RW) | sp * sex
> xyplot(form, data=crabs)

2.0 2.2 2.4 2.6 2.8 3.0

M
B

M
O
3.0
2.8
2.6
2.4
2.2

log(FL)

2.0

F
B

F
O

3.0
2.8
2.6
2.4
2.2
2.0
2.0 2.2 2.4 2.6 2.8 3.0

log(RW)

The formula form in this case has three parts. The left-hand side log(FL)
is to be plotted against the right log(FL); since both these variables are
continuous, we will obtain a scatter plot. The conditioning bar | indicates
that we wish the information to be broken down by the third term, sp*sex
(i.e. by species and by sex). Hence lattice produces four separate scatter
plots, each with the same axes.

46

The most common use of the lattice package is to produce these trellis
plots for representing multivariate data. A few more examples you might
find useful:
> histogram(~ height | voice.part, data=singer)

60

65

70

75

Soprano 2

Soprano 1

Tenor 1

Alto 2

Percent of Total

40
30
20
10
0

Alto 1
40
30
20
10
0

Bass 2

Bass 1

Tenor 2

40
30
20
10
0
60

65

70

75

60

65

70

75

height

Notice that the command is histogram(), not hist(), and the plotting
options are different.
> library(MASS)
> densityplot(galaxies)

47

0.00015

Density

0.00010

0.00005

0.00000
10000

20000

30000

galaxies

5.6

Function Plots

When applied to a vectorized function of one argument, plot() will produce

a graph in the specified range.
> plot(sin, -2*pi, 2*pi)

48

1.0
0.5
0.0
1.0

0.5

sin

For non-vectorized functions, this has to be done manually:

>
+
+
+
>
>
>
>
+
+
+
+

func = function(x, n=10) {

idx = 1:n
return(sum((-1)^(idx+1)*x^idx/idx))
}
plot(function(x) log(1+x), -0.5, 1.2)
for (i in 2:10) {
x <- seq(from=-0.5, to=1.2, length=1000)
y <- sapply(x, func, n=i)
points(x, y, type="l")
}

49

0.5
0.0
0.5

function(x) log(1 + x)

0.5

0.0

0.5

1.0

Using the lattice package you can also produce surface plots for functions
with two arguments using wireframe plots.
>
>
>
>

func = function(x,y) (x^3-x)*sin(x+y)

xs = ys = seq(-5, 5, length.out=100)
out = outer(xs, ys, func)
wireframe(out)

50

out

column

row

See also contourplot() and levelplot().

5.7

Customized Plots

To draw a plot from scratch, use the plot.new() command:

>
>
>
>
>
>
>
>
>
>

set.seed(1328)
x = rnorm(100)

# to get the same values as me

# generate data

plot.new()
plot.window(xlim=c(-3,3), ylim=c(-0.1,0.5))
axis(side=1, pos=-0.1)
hist(x, breaks=15, add=TRUE, freq=FALSE, col=2)
plot(dnorm, -3, 3, add=TRUE)
points(x, rep(-0.05,100), pch="|")
title(main="Normal random variables")

plot.window() is used to control the range of the plot, the axis() function
draws on the axes, and title() is used to annotate with text. Try each of
the above commands in turn to see what they do.

51

5.8

Exporting Plots

You will likely need to use R plots in LaTeX documents for your practicals
and projects. If you are using a GUI such as the default R interface on
OS X or Windows, then select the window and go to File > Save As. I
recommend saving plots in PDF format, as this makes it easiest to integrate
with a LaTeX document. Other interfaces such as RStudio make it similarly
easy to create plots.
You can also save plots from the command line. The way to do this is to tell
R to send your plot commands to a file, instead of to the screen. This means
you wont be able to see your plot whilst you produce it (but presumably
youll have already checked what it looks like!) Type pdf("yourfilename.pdf")
to start, then run your chosen plot commands. Then finish with dev.off()
to go back to the default state. For example:
> pdf("plotfile.pdf")
> plot(hills)
> dev.off()

52

The apply Family of Functions

Much coding involves the repeated application of the same function to several different pieces of data in a vector or list. For this reason, R has a series
of functions for performing such tasks, which results in much simpler and
easier to understand code.

6.1

apply()

If we want to perform a function on every row or column in a matrix (or on

an array, see next section), we can use the apply() function. The syntax
is apply(x, d, f), which will apply the function f to each of the dth dimension of x. If d=1 this corresponds to rows, and d=2 to the columns of a
matrix.
> A = cbind(1:10, (1:10)^2, (1:10)^3)
> apply(A, 2, sum)
[1]

55

385 3025

It will also work for matrix-like objects, such as data frames (although see
also sapply() below).
> library(MASS)
> apply(hills, 2, mean)
dist
climb
7.529 1815.314

time
57.876

> apply(hills, 2, sd)

dist
climb
5.524 1619.151

time
50.041

Exercise 6.1. Using apply(), write a function which, given an (I J)matrix X = (xij ) computes the magnitude of each row, that is
q
x2i1 + x2i2 + + x2iJ ,
for each i = 1, . . . , I
and returns the results as a vector.

53

Exercise 6.2. Write a function which, given an (I J)-matrix X, returns

a vector of length I with entries
yi =

x
i+
si

where x
i+ and si are respectively the sample mean and sample standard
deviation of entries in the ith row of X. [The mean divided by the standard
deviation is sometimes called the coefficient of variation.]
What happens if you use apply() with a function like range(), which returns more than one value?
Exercise 6.3. Take a look at the data set EuStockMarkets (this is in the
datasets package, which should be already loaded). Find the mean absolute
change in returns from one day to the next for each stock (that is, the average
of |xi+1 xi | over all days i). [Hint: recall the diff() function.]
Bonus*: Think of a more sensible measure of the volatility than this and
implement it [hint: one that doesnt depend upon the scale].
Note that apply() does not run substantially faster than writing a loop to
do the same thing, it is simply easier to code up and to read.
For the particular task of sums or means of rows or columns in a matrix, R
contains special functions rowSums(), colSums(), rowMeans(), colMeans().
These are all much faster than the equivalent apply() commands.
> # 2000 x 2000 random matrix
> x = matrix(rnorm(4e6), 2000, 2000)
> system.time(apply(x,1,sum))
user
0.121

system elapsed
0.004
0.127

> system.time(rowSums(x))
user
0.01

system elapsed
0.00
0.01

Exercise 6.4. Write a function to renormalize the columns of a matrix so

that they sum to 1.
Exercise 6.5. Write a function to perform the same task as in Exercise 6.1,
but this time using rowSums(). Compare the speed of these two methods
using system.time().
54

6.2

sapply() and lapply()

If we want to apply a function to every entry in a list or vector, we can

use lapply(). The syntax is just lapply(x, f) for a vector or list x and a
function f. It returns a list of the same length as x containing the results.
The following example uses lapply() twice, first on a vector, and then on
the resulting list.
> mu = c(-2,-1,0,1,2)
> out = lapply(mu, function(x) rnorm(100, mean=x))
> lapply(out, mean)
[[1]]
[1] -2.169
[[2]]
[1] -1.141
[[3]]
[1] 0.1753
[[4]]
[1] 1.105
[[5]]
[1] 2.008
Note that in the previous example, we didnt want the final answer as a list,
so we might use the unlist() command to turn the results into a vector.
The sapply() function does this automatically when appropriate, but is
otherwise the same as lapply().
> sapply(out, mean)
[1] -2.1688 -1.1414

0.1753

1.1050

2.0080

If the function being applied returns something more complicated than a

single number, you should use lapply() instead.

6.3

replicate()

Sometimes we wish to repeat exactly the same operation multiple times,

without having a different input (this is typically used in conjunction with
55

the generation of random numbers).

> out = replicate(20, rnorm(100), simplify=FALSE)
Now out is a list of 20 independent data sets, each consisting of 100 standard
normal random variables.
Exercise 6.6. Suppose we wish to investigate the distribution of the maximum of 10 Poisson random variables with parameter = 5. Generate 1000
independent data sets consisting of such Poisson random variables (see the
command rpois()), find the maximum of each, and plot as a histogram.

6.4

tapply()

In the same vein as the previous examples, we may wish to evaluate a

function on data in a vector or list which are grouped according to the
levels of some other factor. For example,
> library(MASS)
> head(genotype)

1
2
3
4
5
6

Litter Mother
Wt
A
A 61.5
A
A 68.2
A
A 64.0
A
A 65.0
A
A 59.7
A
B 55.0

> with(genotype, tapply(Wt, Mother, mean))

A
B
I
J
55.40 58.70 53.36 48.68
returns a vector of means. If the function provided gives more than a single
value, then tapply() will adapt accordingly:
> with(genotype, tapply(Wt, Mother, summary))
$A
Min. 1st Qu.
36.3
49.0

Median
58.2

Mean 3rd Qu.

55.4
62.1
56

Max.
68.2

$B
Min. 1st Qu.
42.0
55.2

Median
59.8

Mean 3rd Qu.

58.7
63.5

Max.
69.8

Min. 1st Qu.

39.7
48.9

Median
54.2

Mean 3rd Qu.

53.4
57.5

Max.
61.8

Min. 1st Qu.

39.6
42.9

Median
50.0

Mean 3rd Qu.

48.7
53.5

Max.
61.0

$I

$J

It is also possible to provide more than one grouping in the form of a list or
data frame, in which case the data are broken down by both:
> tapply(genotype$Wt, genotype[,1:2], mean)
Mother
Litter
A
A 63.68
B 52.33
I 47.10
J 54.35

B
52.40
60.64
64.37
56.10

I
54.12
53.92
51.60
54.53

J
48.96
45.90
49.43
49.06

Exercise 6.7. Find the heaviest rats born to each mother in the genotype()
data.

6.5

mapply()

Sometimes it may be useful to apply a function of several arguments repeatedly, where more than one argument can change.
> mapply(seq, from=c(1,4,-3), to=c(2,9,0), by=0.5)
[[1]]
[1] 1.0 1.5 2.0
[[2]]
[1] 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0
[[3]]
[1] -3.0 -2.5 -2.0 -1.5 -1.0 -0.5
57

0.0

Shorter arguments are recycled as necessary.

Arrays and Tables

The data sets we have seen so far were best expressed as data frames, with
a single row corresponding to each observation. For discrete data sets, this
is not always the most useful or efficient way to represent the information,
and it may be more useful to use a contingency table.
Consider the data set occupationalStatus in the datasets package. This
contains information on the occupational status of 3,498 father and son
pairs, grouped from 1 (highest status) to 8 (lowest). One way to present
these data would be a data frame with 2 columns and 3,498 rows; however,
since there are only 64 possible distinct entries in the data frame, we can
represent the data rather more compactly as a matrix:
> occupationalStatus
destination
origin
1
2
3
4
1 50 19 26
8
2 16 40 34 18
3 12 35 65 66
4 11 20 58 110
5
2
8 12 23
6 12 28 102 162
7
0
6 19 40
8
0
3 14 32

5
7
11
35
40
25
90
21
15

6
7
8
11
6
2
20
8
3
88 23 21
183 64 32
46 28 12
554 230 177
158 143 71
126 91 106

Here each row represents the occupational status of the father, and each
column that of the son.
Exercise 7.1. (a) What is the probability of a son having the same occupational status as his father? [Hint: investigate what diag(x) does if x
is a matrix.]
(b) Renormalize the data so that each row sums to 1. In the new data set the
ith row represents the conditional distribution of a sons occupational
status given that his father has occupational status i.
(c) What is the probability that a son has occupational status between 1
and 3, given that his father has status 1?
What if the father has occupational status 8?
58

(d) Calculate the vector y where yi is the probability that the son has occupational status between 1 and 3, given that his father has occupational
status i.

7.1

Higher Dimensions

Of course, if we have a data set consisting of more than two pieces of categorical information about each subject, then a matrix is not sufficient. The
generalization of matrices to higher dimensions is the array. Arrays are
defined much like matrices, with a call to the array() command. Here is a
2 3 3 array:
> arr = array(1:18, dim=c(2,3,3))
> arr
, , 1

[1,]
[2,]

[,1] [,2] [,3]

1
3
5
2
4
6

, , 2

[1,]
[2,]

[,1] [,2] [,3]

7
9
11
8
10
12

, , 3
[,1] [,2] [,3]
[1,]
13
15
17
[2,]
14
16
18
Each 2-dimensional slice defined by the last co-ordinate of the array is shown
as a 2 3 matrix. Note that we no longer specify the number of rows and
columns separately, but use a single vector dim whose length is the number
of dimensions. You can recover this vector with the dim() function.
> dim(arr)
[1] 2 3 3

59

Note that a 2-dimensional array is identical to a matrix. Arrays can be

subsetted and modified in exactly the same way as a matrix, only using the
appropriate number of co-ordinates:
> arr[1,2,3]
[1] 15
> arr[,2,]
[,1] [,2] [,3]
[1,]
3
9
15
[2,]
4
10
16
> arr[1,1,] = c(0,-1,-2)
> arr[,,1,drop=FALSE]

# change some values

, , 1

[1,]
[2,]

[,1] [,2] [,3]

0
3
5
2
4
6

Notice that dimensions are dropped by default.

I have placed some data on housing in a 4-way array on the course website.
To load the data navigate to the appropriate directory and run the command
> house = dget("housing.dat")
The whole table is quite large and difficult to understand, but if we focus
on some of the 2-way marginal distributions it could help.
> margin.table(house, 1:2)
Infl
Sat
Low Medium High
Low
282
206
79
Medium 170
189
87
High
175
264 229
> margin.table(house, c(1,4))
60

Cont
Sat
Low High
Low
262 305
Medium 178 268
High
273 395
This is equivalent to
> apply(house, 1:2, sum)
Infl
Sat
Low Medium High
Low
282
206
79
Medium 170
189
87
High
175
264 229
Exercise 7.2. Create a new data set in which the levels Atrium and Terrace
are combined in to a single level called Other.

7.2

Tables

Given a large data set of discrete items, a useful summary is just to count
the number of items in each category; this can be done using the table()
command.
> library(MASS)
> head(cabbages)

1
2
3
4
5
6

Cult Date HeadWt VitC

c39 d16
2.5
51
c39 d16
2.2
55
c39 d16
3.1
45
c39 d16
4.3
42
c39 d16
2.5
53
c39 d16
4.3
50

> table(cabbages$Date)

d16 d20 d21

20 20 20
61

> table(cabbages[,1:2])
Date
Cult d16 d20 d21
c39 10 10 10
c52 10 10 10
For univariate data this gives a vector of counts, and for multi-way data a
contingency table in the form of an array. Note that the vector is named,
which makes it look slightly more complex than it actually is, but it behaves
just like a normal vector.
> tab = table(cabbages$Date)
> names(tab)
[1] "d16" "d20" "d21"

7.3

Binning Data

If we have continuous data it may be useful to summarize them using a

discretization; that is, we group together some of the data into intervals.

> head(Nile)
[1] 1120 1160

963 1210 1160 1160

> bins = c(0,seq(from=700,to=1300,by=100),Inf)

> bins
[1]

700

800

900 1000 1100 1200 1300

Inf

> disNile = cut(Nile, bins)

> head(disNile)
[1] (1.1e+03,1.2e+03] (1.1e+03,1.2e+03] (900,1e+03]
(1.2e+03,1.3e+03]
[5] (1.1e+03,1.2e+03] (1.1e+03,1.2e+03]
8 Levels: (0,700] (700,800] (800,900] (900,1e+03] ... (1.3e+03,Inf]
> table(disNile)
62

disNile
(0,700]
(700,800]
(800,900]
6
20
25
(1e+03,1.1e+03] (1.1e+03,1.2e+03] (1.2e+03,1.3e+03]
12
11
6

(900,1e+03]
19
(1.3e+03,Inf]
1

The cut() command turns the data into a factor defined by the edges of
the bins you provide. The default labels are rather unwieldy, but this can
be changed:
> disNile = cut(Nile, bins, labels=bins[-9])
> table(disNile)
disNile
0 700
6
20

800
25

900 1000 1100 1200 1300

19
12
11
6
1

63

Strings

A string is a piece of text enclosed in double quotes ("), such as "Hello!".

It may include spaces, numbers and punctuation. You can send a string to
the screen using the cat() function.
> cat("Hello")
Hello
Note that this is quite different from the print() function, which is discussed
in Section 9. So if we write a function and wish to provide information to
the person using it, this is useful:
> myFunc = function(x) {
+
cat("Squaring the number\n")
+
x^2
+ }
>
> myFunc(3)
Squaring the number
[1] 9
Note the \n at the end of our string: this tells the command to start a new
line, otherwise it looks rather ugly (try it). The symbol \ is an escape
character, which allows us to perform special tasks with it. For example,
we cant use the double quotes symbol ", because R will interpret this to
mean that we have finished our string. Instead we write \" to represent a
double quote, and similarly we have to use \\ to represent \.
For example:
> cat("She said \"Hello\" to the man.\n")
She said "Hello" to the man.
> cat("This is a backslash: \\ - how does it make you feel?\n")
This is a backslash: \ - how does it make you feel?
Note that the function cat() also works with numbers and can have several
arguments:
64

> cat(57, "clouds", "\n")

57 clouds
By default cat() separates the arguments with a space, but you can change
this:
> cat(57, "clouds", "\n", sep="")
57clouds
> cat(57, "clouds", "\n", sep=".")
57.clouds.
Exercise 8.1. Write a function which takes a positive integer n and write
the numbers 1, 2, . . . , n on the screen, separated by a comma and a space
and finishing with a full stop. Like this:
1, 2, 3, 4, 5, 6, 7, 8, 9, 10.
Exercise 8.2. Write a function withBox() with a single string argument
str, which prints str inside a box of asterisks like this:
> withBox("Hello")
*********
* Hello *
*********
The function nchar() is useful for giving the length of a string.

8.1

The paste() function

Its often useful to be able to stick strings together, for which purpose we
have the function paste().
> paste("Hello", "there")
[1] "Hello there"
65

If given a vector paste() uses vector recycling, which can be rather useful:

> paste("Plan", LETTERS[1:5])

[1] "Plan A" "Plan B" "Plan C" "Plan D" "Plan E"
Like cat() we also have a sep= option and can work with numbers:
> paste("x", 1:10, sep="")
[1] "x1"

"x2"

"x3"

"x4"

"x5"

"x6"

"x7"

"x8"

"x9"

"x10"

You can also make paste() concatenate all its output into a single string
with the collapse= optional argument:
> paste(LETTERS[1:10])
[1] "A" "B" "C" "D" "E" "F" "G" "H" "I" "J"
> paste(LETTERS[1:10], collapse=" ")
[1] "A B C D E F G H I J"
Exercise 8.3. Write a function which, given a positive integer n, produces
a string of the form 1 < 2 < ... < n. So, for example,
> f(5)
[1] "1 < 2 < 3 < 4 < 5"
[Bonus: do it with a function using only one line of code.]
Exercise 8.4. Write a function which, given a vector x of positive integers,
returns a list of the same length as x, and the ith entry of the list is a
character vector of length x[i]. The entries in the 1st element of the list
should be "a1", "a2", and so on, and in the 2nd should be "b1", "b2" for
i = 2, etc.
For example:

66

> listfunc(c(1,4,2))
[[1]]
[1] "a1"
[[2]]
[1] "b1" "b2" "b3" "b4"
[[3]]
[1] "c1" "c2"
[You can do this in one line of code using mapply()!]

8.2

Other Manipulation

Other potentially useful functions are:

> nchar("How long is this string?")
[1] 24
> strsplit("separate words are fun", split=" ")
[[1]]
[1] "separate" "words"

"are"

"fun"

> substr("I don't want all of this string", 14, 24)

[1] "all of this"

67

Classes and Methods

You may by now have noticed that certain functions, such as plot() or
summary(), appear to behave very differently when applied to different types
of object.
>
>
>
>

x <- rnorm(100)
y <- x + rnorm(100)
mod1 <- lm(y ~ x)
summary(x)

Min. 1st Qu.

-2.8400 -0.6890

Median
0.0079

Mean 3rd Qu.

0.0766 0.8910

Max.
2.5900

> summary(mod1)

Call:
lm(formula = y ~ x)
Residuals:
Min
1Q
-2.2756 -0.7760

Median
0.0506

3Q
0.7570

Max
2.1564

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.00477
0.09791
0.05
0.96
x
1.22301
0.09248
13.23
<2e-16 ***
--Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.977 on 98 degrees of freedom
Multiple R-squared: 0.641,Adjusted R-squared: 0.637
F-statistic: 175 on 1 and 98 DF, p-value: <2e-16
You might be forgiven for wondering whether these functions have endless
lines of code telling them what to do in dozens of different circumstances.
In fact these are generic functions.
> summary
function (object, ...)
68

UseMethod("summary")
<bytecode: 0x7fbdc5b5bce0>
<environment: namespace:base>
When a generic function is called, it looks for a suitable method to use on
the arguments passed to it. You can look at the different methods available
using methods()
> methods(summary)
Which method is chosen depends upon the objects class; in other words,
what type of object it is.
> class(mod1)
[1] "lm"
When you call summary with an object of class lm, it looks for a method
with the name summary.lm(). So if you call
> summary.lm(mod1)
the result is the same as above.

9.1

Classes and Attributes

The class of an object is one of its attributes. Attributes are a named

list of bits of information about an object that are stored alongside it. The
only difference between the list of attributes and any other list is that each
entry must have a name, and the name must be unique. You can see the
attributes of an object using the attributes() command:
> attributes(hills)
$names
[1] "dist"

"climb" "time"

$class
[1] "data.frame"

69

$row.names
[1] "Greenmantle"
[4] "Ben Rha"
[7] "Bens of Jura"
[10] "Traprain"
[13] "Lomonds"
[16] "Cairngorm"
[19] "Black Hill"
[22] "Meall Ant-Suidhe"
[25] "N Berwick Law"
[28] "Largo Law"
[31] "Ben Nevis"
[34] "Cockleroi"

"Carnethy"
"Ben Lomond"
"Cairnpapple"
"Lairig Ghru"
"Cairn Table"
"Seven Hills"
"Creag Beag"
"Half Ben Nevis"
"Creag Dubh"
"Criffel"
"Knockfarrel"
"Moffat Chase"

"Craig Dunain"
"Goatfell"
"Scolty"
"Dollar"
"Eildon Two"
"Knock Hill"
"Kildcon Hill"
"Cow Hill"
"Burnswark"
"Acmony"
"Two Breweries"

$type
[1] "races"
You can also access and set specific attributes by name using attr():
> attr(hills, "class")
[1] "data.frame"
> attr(hills, "row.names")[4] = "Big Hill"
> row.names(hills)
[1]
[4]
[7]
[10]
[13]
[16]
[19]
[22]
[25]
[28]
[31]
[34]

"Greenmantle"
"Big Hill"
"Bens of Jura"
"Traprain"
"Lomonds"
"Cairngorm"
"Black Hill"
"Meall Ant-Suidhe"
"N Berwick Law"
"Largo Law"
"Ben Nevis"
"Cockleroi"

"Carnethy"
"Ben Lomond"
"Cairnpapple"
"Lairig Ghru"
"Cairn Table"
"Seven Hills"
"Creag Beag"
"Half Ben Nevis"
"Creag Dubh"
"Criffel"
"Knockfarrel"
"Moffat Chase"

"Craig Dunain"
"Goatfell"
"Scolty"
"Dollar"
"Eildon Two"
"Knock Hill"
"Kildcon Hill"
"Cow Hill"
"Burnswark"
"Acmony"
"Two Breweries"

This is in addition to the special functions we have already seen: dimnames(),

dim(), names(), rownames(), etc. You can also create entirely new attributes:
70

> attr(hills, "quality") = "Great!"

> attributes(hills)[4]
$type
[1] "races"

9.2

Setting Classes and Methods

Class is an attribute, so you can set it just as easily as any other:

> x <- list(temp=19.5, wind=12, wind.dir="SSW", rain=20, summ="showers")
> class(x) <- "weather"
> attr(x, "class")
[1] "weather"
When you type an objects name at the console, R searches for a print()
method in order to decide how to display it to you. Since there is no such
method for objects of class weather, it just uses print.default(); we can
create a new method by just naming a function print.weather().
> print.weather = function(object) {
+
cat("Weather report\n")
+
cat("Temperature ", object$temp, "C\n", sep="")
+
cat("Wind ", object$wind, " kts, from ", object$wind.dir, "\n", sep="")
+
cat("Rain ", object$rain, " mm: ", object$summ, "\n", sep="")
+
object
+ }
Now:
> x
Weather report
Temperature 19.5C
Wind 12 kts, from SSW
Rain 20 mm: showers
Similarly we can write functions summary.weather(), plot.weather(), or
whatever we think is appropriate.
Exercise 9.1. Write a new summary method for factors, which prints
"Factor Summary" to the screen, before giving a table of counts.
71

10

Debugging

When constructing a complicated program containing several functions, it

is inevitable that we make mistakes. Sometimes these errors will result in
code which does not run at all, resulting in an error message from R.
> "a" + 1
Error:

non-numeric argument to binary operator

On other occasions, code may run but fail to give the correct answer. This
is more dangerous, since we may not realise, but it also makes it harder to
find the source of the problem. Getting warning messages from R is a sign
that something is not working as you intended: do not ignore these!
> 1:3 + 1:2
Warning:
length

longer object length is not a multiple of shorter object

[1] 2 4 4
In any of these cases, you will need to discover what is wrong, and figure
out how to fix it.

10.1

Prevention

There are few steps you can take in your programming style to reduce the
likelihood of making mistakes in the first place. These are:
Make your code modular, and dont repeat yourself. Try to
construct your code in the spirit of R: you should use self contained
functions to perform specific tasks, and call those functions when necessary. If your program needs to perform the same task more than
once, then have it call the same code each time: this way its easier to
detect and cure any mistakes.
Write simple code when possible. R has lots of useful built in
features which allow complicated code to be written very succinctly
and clearly, such as by using the apply() family. Try to make use
of such features when possible. In the examination for this course,
you will lose marks for failing to use a function like lapply() in a
straightforward situation.
72

 Comment your code. Youll have noticed by now that anything

followed by a # symbol in R is ignored by the computer. This is to allow
you to put comments in the code, so that someone reading it later can
understand what it is intended to do (often this is just you sometime
later, after youve forgotten what you were doing when you first wrote
the code). Its also helpful to give the variables and functions you use
sensible and descriptive names.
>
>
>
>
>

## generate random numbers

set.seed(241)
rndSteps <- rnorm(N)
## generate cumulative sum for random walk
rndWalk <- cumsum(rndSteps)

In an exam context, comments are a useful way to indicate to the

examiner what you are trying to do.
Think about special cases. The most common errors arise from
failing to think about what will happen when (for example) your input
is of length 1 or 0; one particularly common example is in the indexing
of arrays:
>
>
+
+
>
>

# function to sum over subset of columns of a matrix

row_sums <- function(x, cols) {
apply(x[,cols], 1, sum)
}
x <- matrix(1:9, 3, 3)
row_sums(x, 2:3)

[1] 11 13 15
> row_sums(x, 2)
Error:

dim(X) must have a positive length

Of course, this is easily prevented by adding drop=TRUE to the subset

command.

10.2

Modularity

R is fundamentally a functional language, which means that all operations

are performed by functions, which are generally self-contained. Its good
practice to keep your code modular, in the sense of being made up of
73

distinct functions which perform small tasks, and that you can test (and
debug) individually.
Most fundamentally, if you have a complex instruction which isnt working,
try breaking it down into pieces to see whether each piece is doing what you
intend.
> if(any(x > 3) && y != 2) {
+
print("Hello")
+ }
Error:

missing value where TRUE/FALSE needed

> any(x > 3)

[1] TRUE
> y != 2
logical(0)
> y
NULL
In more complicated code, it is important to write separate functions which
are responsible for individual parts of your program. These can then be
tested for errors separately, reducing the complexity of the problem. This
also makes it much easier to reuse bits of code in other circumstances.
>
+
+
+
>
>
+
+
+
>
>
+

g = function(y) {
if (y < 0) warning("Some warning")
return(y)
}
h = function(z) {
stop("Some error message")
z
}
f = function(x) {
# this is a function which calls some other functions
74

+
out = g(x) + h(x)
+ }
It is also easy to trace errors down to the particular function in which they
occur, but harder to search within that code for problems. If an error
occurs, immediately calling the traceback() function will show the series
of functions which were called in the run up to the problem (the call stack).

> f(2)
Error:

Some error message

> traceback()
3: stop("Some error message") at #3
2: h(x) at #3
1: f(2)
This shows that the error occurred in the function h(). We can fix the code
and call the function again.
> h = function(z) {
+
z
+ }
> f(-2)
Warning:

Some warning

We now obtain a warning. We cannot use traceback() with a warning, but

we can tell R to turn any warnings into errors for the purposes of debugging:

Warning:

Some warning

No traceback available
> options(warn=2)
> f(-2)
Error in g(x) :

(converted from warning) Some warning

75

> traceback()
7: doWithOneRestart(return(expr), restart)
6: withOneRestart(expr, restarts[[1L]])
5: withRestarts({
.Internal(.signalCondition(simpleWarning(msg, call), msg,
call))
.Internal(.dfltWarn(msg, call))
}, muffleWarning = function() NULL)
4: .signalSimpleWarning("Some warning",
quote(g(x)))
3: warning("Some warning") at #2
2: g(x) at #3
1: f(-2)
It is recommended that you add in your own stop() commands to deal with
problems which you think are likely to arise (such as bad user input), and
to provide an informative error message. This is much better than having
the code fail at a lower level in some internal R function, leaving the reason
for failure as a potential mystery.

10.3

Error Recovery

An extremely useful feature of R is error recovery, which is built in to the

interface. Set it up using:
> options(error=recover)
Now if any function calls an error, R will pause its execution and show
you the call stack. You then have the option to look inside any of those
functions, and inspect the objects within them. This allows you to see if
theyre doing what you expect.
After choosing a function to work with (by typing the relevant number), R
brings up a browser, which looks like the ordinary command line, but with
a slightly different prompt. Typing commands at this prompt allows you to
inspect objects within the particular function you selected. Use the escape
character Q to exit5 .
To revert to normal working, use:
5

If you have an object called Q youll need to use print(Q) to look at it.

76

> options(error=NULL)

10.4

Other Functions

The function debug() allows you to inspect what a function does line by
line: this is particularly useful if your code is not doing what you expect,
but does not actually result in an error. Try calling:
> debug(g)
> f(2)
R will bring up the browser as soon as g() is called (in this case from within
f(). The browser will be inside g()s environment, and you can inspect
the objects within it on that basis. In addition, pressing Enter without
typing any commands will step through one line of g()s code at a time,
allowing you to check what is happening as the function progresses. The
special command Q can, again, be used to break out of this.
To turn off this feature use the function undebug(). If you only wish the
console to appear this way on one occasion, you can also use debugonce()
instead of debug().
You can also temporarily edit a function so as to see what it is doing in
a more hands-off fashion. Use the function trace() with the edit=TRUE
option.
> trace(g, edit=TRUE)
This brings up an editor containing a version of your function. You can
do whatever you like to this and it will not permanently alter your code;
inserting print() statements is a good way to use this. You can revert to
the original version of your function using untrace().

77

11

Arithmetic Subtleties*

Computers store most numbers in the form l2k , for some number 1 l < 2,
and an integer k {52, 51, . . . , 52}. This can lead to rounding errors, as
the following illustrates6 .
> 0.3 - 0.1 - 0.2
[1] -2.776e-17
This can cause problems when making comparisons, so the command all.equal()
can be used to avoid this problem.
> x = 0.3 - 0.1 - 0.2
> x == 0
[1] FALSE
> all.equal(x, 0)
[1] TRUE
If things are different it returns an explanation, so use isTRUE(all.equal(...))
in a conditional statement:
> if (isTRUE(all.equal(x, 0))) print("We've got nothing!")
[1] "We've got nothing!"
Very small or large numbers may not respond as you expect in ordinary
arithmetic.
> 2^-1074
[1] 4.941e-324
> 2^-1075
[1] 0
6

See http://www.smbc-comics.com/index.php?db=comics&id=2999

78

> (2^-1074)/1.5
[1] 4.941e-324
> 2^2000
[1] Inf
Large numbers may be replaced with Inf, which behaves as you might expect
infinity to.
> Inf/2
[1] Inf
> (-1)*Inf
[1] -Inf
> 0*Inf
[1] NaN
> Inf - Inf
[1] NaN
> exp(-Inf)
[1] 0

11.1

Integers and Doubles

R does distinguish between integers and ordinary numbers (doubles), but it

is not easy to tell the difference, and integers are often coerced into doubles
without the user noticing. This can cause problems

79

> i1 = 0L
> i2 = 0
> (i1 == i2)
[1] TRUE
> identical(i1, i2)
[1] FALSE
> i3 = 1e40 + 1
> i3-1e40
[1] 0

80

12

List of the Most Useful Functions in R

# libraries
library(), require(), install.packages()
# basic objects
`=`, `<-`, `<<-`, is(), as()
c(), list(), unlist(), matrix(), array(), data.frame(), `~`
cbind(), rbind()
head(), tail(), rev()
# attributes
attributes(), attr()
length(), dim(), ncol(), nrow()
names(), dimnames(), rownames(), colnames()
cat(), print(), print.default(), paste(),
# logic
`!`, `&`, `|`, `&&`, `||`, `==`
any(), all(), ifelse()
apply(), lapply(), sapply(), replicate(), tapply(), mapply()
seq(), seq_along(), seq_len(), `:`
rep(), rep.int()
unique(), duplicated()
match(), which(), `%in%`, which.max(), which.min()
sort(), order(), rank()
factor(), is.factor(), as.factor(),
levels(), nlevels(), droplevels()
table(), cut()
t(), `%*%`, det(), solve(), diag()
sum(), prod(), max(), min()
pmax(), pmin()
cumsum(), cumprod(), diff()
colSums(), rowSums(), colMeans(), rowMeans()
# accessing objects
setwd(), getwd(), ls(), rm()
with(), attach(), detach(),
dput(), dget()
scan(), read.table(), read.csv()

81

# loops
if(), for(), while(), next, break, switch(),
# functions
args(), missing(), Recall(), return()
warning(), stop()
# performance and debugging
system.time(),
debug(), debugonce(), undebug(), trace(), untrace()
# statistics
mean(), sd(), var(), cor(), cov()
runif(), rnorm(), rgamma(), rpois(), rbinom(), rexp(),
sample(), set.seed()
# graphics
plot.new(), plot.window(), axis(), legend()
points(), lines(), abline(),
plot(), hist(), boxplot(), pairs()
# arithmetic
abs(), ceiling(), floor(), round(), signif(),
exp(), log(),
`+`, `-`, `*`, `/`, `%%`,
# numerical methods
integrate(), nlm(), optim()
eigen(), svd(), qr()

82