R for reproducible scientific analysis

Data structures

Learning Objectives

  • To be aware of the different types of data
  • To begin exploring the data.frame, and understand how it’s related to vectors, factors and lists
  • To be able to ask questions from R about the type, class, and structure of an object.

One of R’s most powerful features is its ability to deal with tabular data - like what you might already have in a spreadsheet or a CSV. Let’s start by making a toy dataset in your data/ directory, called feline-data.csv:


We can load this into R via the following:

cats <- read.csv(file = "data/feline-data.csv")
    coat weight likes_string
1 calico    2.1         TRUE
2  black    5.0        FALSE
3  tabby    3.2         TRUE

The read.csv function is used for reading in tabular data stored in a text file where the columns of data are delimited by commas (csv = comma separated values). Tabs are also commonly used to separated colums - if your data are in this format you can use the function read.delim. If the columns in your data are delimited by a character other than commas or tabs, you can use the more general and flexible read.table function.

We can begin exploring our dataset right away, pulling out columns by specifying them using the $ operator:

[1] 2.1 5.0 3.2
[1] calico black  tabby 
Levels: black calico tabby

We can do other operations on the columns:

## Say we discovered that the scale weighs one Kg light:
cats$weight + 2
[1] 4.1 7.0 5.2
paste("My cat is", cats$coat)
[1] "My cat is calico" "My cat is black"  "My cat is tabby" 

But what about

cats$weight + cats$coat
Warning in Ops.factor(cats$weight, cats$coat): '+' not meaningful for
[1] NA NA NA

Understanding what happened here is key to successfully analyzing data in R.

Data Types

If you guessed that the last command will return an error because 2.1 plus "black" is nonsense, you’re right - and you already have some intuition for an important concept in programming called data types. We can ask what type of data something is:

[1] "double"

There are 5 main types: double, integer, complex, logical and character.

[1] "double"
[1] "integer"
[1] "complex"
[1] "logical"
[1] "character"

Note the L suffix to insist that a number is an integer. No matter how complicated our analyses become, all data in R is interpreted as one of these basic data types. This strictness has some really important consequences.

A user has added details of another cat. This information is in the file data/feline-data_v2.csv.

tabby,2.3 or 2.4,TRUE

Load the new cats data like before, and check what type of data we find in the weight column:

cats <- read.csv(file="data/feline-data_v2.csv")
[1] "integer"

Oh no, our weights aren’t the double type anymore! If we try to do the same math we did on them before, we run into trouble:

cats$weight + 2
Warning in Ops.factor(cats$weight, 2): '+' not meaningful for factors

What happened? When R reads a csv into one of these tables, it insists that everything in a column be the same basic type; if it can’t understand everything in the column as a double, then nobody in the column gets to be a double. The table that R loaded our cats data into is something called a data.frame, and it is our first example of something called a data structure - that is, a structure which R knows how to build out of the basic data types.

We can see that it is a data.frame by calling the class function on it:

[1] "data.frame"

In order to successfully use our data in R, we need to understand what the basic data structures are, and how they behave. For now, let’s remove that extra line from our cats data and reload it, while we investigate this behavior further:



And back in RStudio:

cats <- read.csv(file="data/feline-data.csv")

Vectors & Type Coercion

To better understand this behavior, let’s meet another of the data structures: the vector.

my_vector <- vector(length = 3)

A vector in R is essentially an ordered list of things, with the special condition that everything in the vector must be the same basic data type. If you don’t choose the datatype, it’ll default to logical; or, you can declare an empty vector of whatever type you like.

another_vector <- vector(mode='character', length=3)
[1] "" "" ""

You can check if something is a vector:

 chr [1:3] "" "" ""

The somewhat cryptic output from this command indicates the basic data type found in this vector - in this case chr, character; an indication of the number of things in the vector - actually, the indexes of the vector, in this case [1:3]; and a few examples of what’s actually in the vector - in this case empty character strings. If we similarly do

 num [1:3] 2.1 5 3.2

we see that that’s a vector, too - the columns of data we load into R data.frames are all vectors, and that’s the root of why R forces everything in a column to be the same basic data type.

Discussion 1

Why is R so opinionated about what we put in our columns of data? How does this help us?

You can also make vectors with explicit contents with the concatenate function:

concat_vector <- c(2,6,3)
[1] 2 6 3

Given what we’ve learned so far, what do you think the following will produce?

quiz_vector <- c(2,6,'3')

This is something called type coercion, and it is the source of many surprises and the reason why we need to be aware of the basic data types and how R will interpret them. When R encounters a mix of types (here numeric and character) to be combined into a single vector, it will force them all to be the same type. Consider:

coercion_vector <- c('a', TRUE)
[1] "a"    "TRUE"
another_coercion_vector <- c(0, TRUE)
[1] 0 1

The coercion rules go: logical -> integer -> numeric -> complex -> character, where -> can be read as are transformed into. You can try to force coercion against this flow using the as. functions:

character_vector_example <- c('0','2','4')
[1] "0" "2" "4"
character_coerced_to_numeric <- as.numeric(character_vector_example)
[1] 0 2 4
numeric_coerced_to_logical <- as.logical(character_coerced_to_numeric)

As you can see, some surprising things can happen when R forces one basic data type into another! Nitty-gritty of type coercion aside, the point is: if your data doesn’t look like what you thought it was going to look like, type coercion may well be to blame; make sure everything is the same type in your vectors and your columns of data.frames, or you will get nasty surprises!

But coercion can also be very useful! For example, in our cats data likes_string is numeric, but we know that the 1s and 0s actually represent TRUE and FALSE (a common way of representing them). We should use the logical datatype here, which has two states: TRUE or FALSE, which is exactly what our data represents. We can ‘coerce’ this column to be logical by using the as.logical function:

[1] 1 0 1
cats$likes_string <- as.logical(cats$likes_string)

Concatenate will also append things to an existing vector:

ab_vector <- c('a', 'b')
[1] "a" "b"
concat_example <- c(ab_vector, 'SWC')
[1] "a"   "b"   "SWC"

You can also make series of numbers:

mySeries <- 1:10
 [1]  1  2  3  4  5  6  7  8  9 10
 [1]  1  2  3  4  5  6  7  8  9 10
seq(1,10, by=0.1)
 [1]  1.0  1.1  1.2  1.3  1.4  1.5  1.6  1.7  1.8  1.9  2.0  2.1  2.2  2.3
[15]  2.4  2.5  2.6  2.7  2.8  2.9  3.0  3.1  3.2  3.3  3.4  3.5  3.6  3.7
[29]  3.8  3.9  4.0  4.1  4.2  4.3  4.4  4.5  4.6  4.7  4.8  4.9  5.0  5.1
[43]  5.2  5.3  5.4  5.5  5.6  5.7  5.8  5.9  6.0  6.1  6.2  6.3  6.4  6.5
[57]  6.6  6.7  6.8  6.9  7.0  7.1  7.2  7.3  7.4  7.5  7.6  7.7  7.8  7.9
[71]  8.0  8.1  8.2  8.3  8.4  8.5  8.6  8.7  8.8  8.9  9.0  9.1  9.2  9.3
[85]  9.4  9.5  9.6  9.7  9.8  9.9 10.0

We can ask a few questions about vectors:

sequence_example <- seq(10)
head(sequence_example, n=2)
[1] 1 2
tail(sequence_example, n=4)
[1]  7  8  9 10
[1] 10
[1] "integer"
[1] "integer"

Finally, you can give names to elements in your vector:

names_example <- 5:8
names(names_example) <- c("a", "b", "c", "d")
a b c d 
5 6 7 8 
[1] "a" "b" "c" "d"

Challenge 1

Start by making a vector with the numbers 1 through 26. Multiply the vector by 2, and give the resulting vector names A through Z (hint: there is a built in vector called LETTERS)

Data Frames

We said that columns in data.frames were vectors:

 num [1:3] 2.1 5 3.2
 logi [1:3] TRUE FALSE TRUE

These make sense. But what about

 Factor w/ 3 levels "black","calico",..: 2 1 3


Another important data structure is called a factor. Factors usually look like character data, but are typically used to represent categorical information. For example, let’s make a vector of strings labeling cat colorations for all the cats in our study:

coats <- c('tabby', 'tortoiseshell', 'tortoiseshell', 'black', 'tabby')
[1] "tabby"         "tortoiseshell" "tortoiseshell" "black"        
[5] "tabby"        
 chr [1:5] "tabby" "tortoiseshell" "tortoiseshell" "black" ...

We can turn a vector into a factor like so:

CATegories <- factor(coats)
[1] "factor"
 Factor w/ 3 levels "black","tabby",..: 2 3 3 1 2

Now R has noticed that there are three possible categories in our data - but it also did something surprising; instead of printing out the strings we gave it, we got a bunch of numbers instead. R has replaced our human-readable categories with numbered indices under the hood:

[1] "character"
[1] "integer"

Challenge 2

Is there a factor in our cats data.frame? what is its name? Try using ?read.csv to figure out how to keep text columns as character vectors instead of factors; then write a command or two to show that the factor in cats is actually a character vector when loaded in this way.

In modeling functions, it’s important to know what the baseline levels are. This is assumed to be the first factor, but by default factors are labeled in alphabetical order. You can change this by specifying the levels:

mydata <- c("case", "control", "control", "case")
factor_ordering_example <- factor(mydata, levels = c("control", "case"))
 Factor w/ 2 levels "control","case": 2 1 1 2

In this case, we’ve explicitly told R that “control” should represented by 1, and “case” by 2. This designation can be very important for interpreting the results of statistical models!


Another data structure you’ll want in your bag of tricks is the list. A list is simpler in some ways than the other types, because you can put anything you want in it:

list_example <- list(1, "a", TRUE, 1+4i)
[1] 1

[1] "a"

[1] TRUE

[1] 1+4i
another_list <- list(title = "Research Bazaar", numbers = 1:10, data = TRUE )
[1] "Research Bazaar"

 [1]  1  2  3  4  5  6  7  8  9 10

[1] TRUE

We can now understand something a bit surprising in our data.frame; what happens if we run:

[1] "list"

We see that data.frames look like lists ‘under the hood’ - this is because a data.frame is really a list of vectors and factors, as they have to be - in order to hold those columns that are a mix of vectors and factors, the data.frame needs something a bit more flexible than a vector to put all the columns together into a familiar table. In other words, a data.frame is a special list in which all the vectors must have the same length.

In our cats example, we have an integer, a double and a logical variable. As we have seen already, each column of data.frame is a vector.

[1] calico black  tabby 
Levels: black calico tabby
[1] calico black  tabby 
Levels: black calico tabby
[1] "integer"
 Factor w/ 3 levels "black","calico",..: 2 1 3

Each row is an observation of different variables, itself a data.frame, and thus can be composed of element of different types.

    coat weight likes_string
1 calico    2.1         TRUE
[1] "list"
'data.frame':   1 obs. of  3 variables:
 $ coat        : Factor w/ 3 levels "black","calico",..: 2
 $ weight      : num 2.1
 $ likes_string: logi TRUE

Notice the subtle differences between different ways of calling variables, observations and elements from data.frames:

1 calico
2  black
3  tabby
[1] calico black  tabby 
Levels: black calico tabby
[1] calico black  tabby 
Levels: black calico tabby
1 calico
2  black
3  tabby
[1] calico black  tabby 
Levels: black calico tabby
    coat weight likes_string
1 calico    2.1         TRUE
[1] calico
Levels: black calico tabby


Last but not least is the matrix. We can declare a matrix full of zeros:

matrix_example <- matrix(0, ncol=6, nrow=3)
     [,1] [,2] [,3] [,4] [,5] [,6]
[1,]    0    0    0    0    0    0
[2,]    0    0    0    0    0    0
[3,]    0    0    0    0    0    0

And similar to other data structures, we can ask things about our matrix:

[1] "matrix"
[1] "double"
 num [1:3, 1:6] 0 0 0 0 0 0 0 0 0 0 ...
[1] 3 6
[1] 3
[1] 6

Challenge 3

What do you think will be the result of length(matrix_example)? Try it. Were you right? Why / why not?

Challenge 4

Make another matrix, this time containing the numbers 1:50, with 5 columns and 10 rows. Did the matrix function fill your matrix by column, or by row, as its default behaviour? See if you can figure out how to change this. (hint: read the documentation for matrix!)

Challenge 5

Create a list of length two containing a character vector for each of the sections in this part of the workshop:

  • Data types
  • Data structures

Populate each character vector with the names of the data types and data structures we’ve seen so far.

Challenge 6

Consider the R output of the matrix below:

     [,1] [,2]
[1,]    4    1
[2,]    9    5
[3,]   10    7

What was the correct command used to write this matrix? Examine each command and try to figure out the correct one before typing them. Think about what matrices the other commands will produce.

  1. matrix(c(4, 1, 9, 5, 10, 7), nrow = 3)
  2. matrix(c(4, 9, 10, 1, 5, 7), ncol = 2, byrow = TRUE)
  3. matrix(c(4, 9, 10, 1, 5, 7), nrow = 2)
  4. matrix(c(4, 1, 9, 5, 10, 7), ncol = 2, byrow = TRUE)

Challenge solutions

Discussion 1

By keeping everything in a column the same, we allow ourselves to make simple assumptions about our data; if you can interpret one entry in the column as a number, then you can interpret all of them as numbers, so we don’t have to check every time. This consistency, like consistently using the same separator in our data files, is what people mean when they talk about clean data; in the long run, strict consistency goes a long way to making our lives easier in R.

Solution to Challenge 1

x <- 1:26
x <- x * 2
names(x) <- LETTERS

Solution to Challenge 2

cats <- read.csv(file="data/feline-data.csv", stringsAsFactors=FALSE)

Note: new students find the help files difficult to understand; make sure to let them know that this is typical, and encourage them to take their best guess based on semantic meaning, even if they aren’t sure.

Solution to challenge 3

What do you think will be the result of length(matrtix_example)?

matrix_example <- matrix(0, ncol=6, nrow=3)
[1] 18

Because a matrix is a vector with added dimension attributes, length gives you the total number of elements in the matrix.

Solution to challenge 4

Make another matrix, this time containing the numbers 1:50, with 5 columns and 10 rows. Did the matrix function fill your matrix by column, or by row, as its default behaviour? See if you can figure out how to change this. (hint: read the documentation for matrix!)

x <- matrix(1:50, ncol=5, nrow=10)
x <- matrix(1:50, ncol=5, nrow=10, byrow = TRUE) # to fill by row

Solution to Challenge 5

dataTypes <- c('double', 'complex', 'integer', 'character', 'logical')
dataStructures <- c('data.frame', 'vector', 'factor', 'list', 'matrix')
answer <- list(dataTypes, dataStructures)

Note: it’s nice to make a list in big writing on the board or taped to the wall listing all of these types and structures - leave it up for the rest of the workshop to remind people of the importance of these basics.

Solution to challenge 6

Consider the R output of the matrix below:

     [,1] [,2]
[1,]    4    1
[2,]    9    5
[3,]   10    7

What was the correct command used to write this matrix? Examine each command and try to figure out the correct one before typing them. Think about what matrices the other commands will produce.

matrix(c(4, 1, 9, 5, 10, 7), ncol = 2, byrow = TRUE)