4 Introduction to Data Carpentry

Data carpentry gives us the tools to work with large data sets. During this chapter you will learn basic concepts and skills to help you import, tidy and manage your data.


4.1 Data Types

Before learning how to work with data in R, we should know about some of the different types of data there are.

Here is a figure that lays out the hierarchical levels of data types. We use the term ‘variable’ to refer here to data.


4.1.0.1 Qualitative vs. Quantitative Data

As you can see at the highest level, there are two data types: quantitative and qualitative. The easiest way to distinguish these data types part is to think of qualitative data as data we that are types . In practical terms, you can think of quantitative data as measures such as values or counts that can be expressed as numbers and can be compared on a numeric scale - either whole numbers or numbers with decimal places. Qualitative data are measures or observations of qualities, types or characteristics. They are often cases not numbers but in text form (although they can sometimes be numbers).


4.1.1 Categorical Data

Under qualitative data, we have two different data types: Nominal and Ordinal. Both of these data types are known as categorical data, which means they represent characteristics such as a person’s gender, education level, language, ethnicity etc.

Categorical data can also be represented in a dataset by numbers. For example, if we are studying smokers and non-smokers in a study, the variable “smoking” would be a categorical variable. We could represent that data in text format using “smoker” or “non-smoker” as our data entry, or we could represent smokers as a “0” and non-smokers as a “1”. However, it is important to remember that these numbers don’t have any mathematical meaning here - it’s just a convenient way of representing the data.


4.1.1.1 Nominal Data

Nominal data, sometimes referred to as unordered data are values which represent discrete units and are used to label a variable that have no quantitative value. The key thing about nominal data is that they do not have any natural value or rank, therefore if you change the order of the data, the meaning of the data does not change.

An example of nominal data would be the results to: “What languages do you speak at home?”. If in our study we got the answers: French, English, Spanish & Hindi from different subjects, then the data type is unordered. There is no sense in that French, English, Spanish, or Hindi are higher in rank than any other - they are simply unordered categories.


4.1.1.2 Ordinal Data

Like nominal data, ordinal data describes discrete units or categories. However, unlike nominal data, ordinal data does have a natural order or ranking to it.

For example, in a study we may ask subjects, “What is the highest level of education you achieved?” They may be able to choose from the following options:

1- Elementary 2- High School 3- Undergraduate 4- Graduate

Here, each level of the data is a category, but it is ordered. Graduate is higher than Undergraduate which is higher than High School which is higher than Elementary. Therefore this data variable (education) is ordinal data.


4.1.2 Numerical Data (Discrete vs. Continuous)

Under quantitative data, there is numerical data consist of numbers with real mathematical meaning. There are two types of numerical data.


  1. Discrete data is data that is separated such that this data can only take on certain value. Most often, these are things that can only be counted in whole units. e.g. If you asked subjects in a study - “how many unique people did you text this week?”, then the answer has to be an integer (a whole number), e.g. 0,1,2,3,etc. You cannot text 2.5 people. Think about discrete data as count data since it can only be counted and not measured.


  1. Continuous data is the opposite of discrete data, it can only be measured not counted. Continuous quantitative data is most often measured using some instrument. For example, if we collected the body weight or height of each subject in a study, then we would get values that would be measured using a scale or a measuring tape and could include decimal places. e.g. you could be 170.5 lbs heavy and you can be 180.2 cm tall.

Additionally, quantitative data can be described in terms of interval and ratio data.


4.1.2.1 Interval Data

Interval data represents values that are ordered on a scale and the differences between any two values has meaning.

In interval data there is no true zero, which means if you have zero in your data, then zero is just another number. For example, zero on the Fahrenheirt temperature scale means it is zero degrees outside, but doesn’t mean the scale stops - 0 actually refers to a temperature. We can also have negative numbers in interval data which are meaningful.


4.1.2.2 Ratio data

For a variable to be considered ratio data, it holds all the properties of an interval variable - it relates to numbers (including those with decimal places) that can be ordered on a scale. However, ratio data also has a true meaning of zero. Whenever a ratio variable is zero, it means there is none of that variable.

For example, in a study, if a subject reports that they texted zero people in the last week, then here 0 really means the absence of something - it relates to none. Another example of ratio scale data is temperature in Kelvin, where zero degrees Kelvin means there is no heat being emitted.



4.2 Importing Data

There are different options for importing data. It’s possible to import data of all different formats into RStudio. We will primarily use spreadsheet type files that have been saved with the .csv suffix. These are called ‘comma separated files’.

Option 1. Import Tab You can click on the “Import Dataset” tab in the top right of RStudio - it’s located just above your global environment.

Depending on your RStudio version, you will be asked to select from a dropdown menu. When importing .csv files, you want to select From CSV... if your menu looks like this:

If your menu looks like the underneath, then you’ll want to select From Text (readr):


Option 2. Writing code. This is the option that we will use in our scripts for this course. You may notice that all the datasets that we wish to use are in a folder called “data”. To read in any of these datasets, what we need to do is use the read_csv() function. This comes from a package contained with the tidyverse package, so we must have imported that library first. We then tell it which dataset to find within the ‘data’ folder. We use the notation “data/…” to tell it to look inside the data folder. For instance, if we wished to load in the bmi.csv dataset, and assign it the name bmi, we would do it like this - make sure to put quotes around the whole file name and location:

library(tidyverse)  #load package

bmi <- read_csv("data/bmi.csv") # bring in data

head(bmi) # first six rows
## # A tibble: 6 x 9
##      id   age   bmi  chol insulin   dbp   sbp smoke educ 
##   <dbl> <dbl> <dbl> <dbl>   <dbl> <dbl> <dbl> <dbl> <chr>
## 1     1  18.3  15.4  47.2    2.78  29.5  49.7     1 D    
## 2     2  19.4  14.5  52.1    3.47  31.3  49.0     2 C    
## 3     3  21.1  16.0  52.2    4.06  32.4  51.7     1 B    
## 4     4  21.3  19.5  53.2    4.48  32.0  NA       2 A    
## 5     5  21.1  18.6  55.4    5.34  33.7  53.8     1 D    
## 6     6  23.9  19.5  54.3    6.29  35.0  56.0     2 C
tail(bmi) # last six rows
## # A tibble: 6 x 9
##      id   age   bmi  chol insulin   dbp   sbp smoke educ 
##   <dbl> <dbl> <dbl> <dbl>   <dbl> <dbl> <dbl> <dbl> <chr>
## 1    15  34.8  30.4  63.6    15.3  42.7  64.8     1 B    
## 2    16  33.6  29.9  65.5    NA    46.4  65.6     2 A    
## 3    17  33.4  NA    65.2    18.2  46.2  64.6     1 D    
## 4    18  35.1  32.5  68.6    19.7  45.8  66.1     2 C    
## 5    19  35.0  33.4  NA      21.1  NA    68.4     1 B    
## 6    20  37.5  34.1  68.0    22.1  49.3  68.6     2 A

If you want to see more on what the data looks like the following functions can help.

nrow(bmi) # how many rows in dataset
## [1] 20
ncol(bmi) # how many columns in dataset
## [1] 9
colnames(bmi) # column names
## [1] "id"      "age"     "bmi"     "chol"    "insulin" "dbp"     "sbp"    
## [8] "smoke"   "educ"

If you just want to view your entire dataset, there is a function for that.

View(bmi)

4.3 Introduction to Dataframes

A dataframe, in R is like a spreadsheet that contains your data. Each column contains values or characters for one variable. In R, columns are called variables. Rows are called observations.

In R dataframes need to contain the following:

  1. Columns should all be the same length, with unique column names. These column names should not start with numbers or punctuation, and have no spaces.

  2. The data stored in dataframes can hold many different data types. The most common are numbers. R describes columns with numbers as being numeric, although a column containing only whole numbers (e.g. 1, 5, 342, 1034) may be called integers. Columns containing any value with a decimal place (e.g. 0.01, 4.4, -7.39494) will be called double. Both double and integer are types of numeric data.

The second most common data type for a column is character. This is the case if the data in the column have any text or punctuation.

A related column type is factor. This occurs when the character type is grouped, e.g. if you had a column with different days of the week, you may wish for R to recognize that each is a separate category.

Another column type that comes up semi-regularly, although not much in this course, is logical. This is when the column contains either TRUE or FALSE values.

Let’s dive into all the features of a dataframe.

First import data, using read_csv().

df <- read_csv("data/cheese.csv")

4.3.1 Dataframe basics

  1. Look at top few rows using head(). The default for head() is to look at the top 6 rows. However, if you put a comma, and then a number, you can get that number of rows instead. Here we get the default 6, and then we get 4 and 8 rows.
head(df)
## # A tibble: 6 x 9
##   type      sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>       <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 blue         18.7       0.8          7.78    21.4  2.34    75     0   353
## 2 brick        18.8       0.784        8.60    23.2  2.79    94     0   371
## 3 brie         17.4       0.826        8.01    20.8  0.45   100     0   334
## 4 camembert    15.3       0.724        7.02    19.8  0.46    72     0   300
## 5 caraway      18.6       0.83         8.28    25.2  3.06    93     0   376
## 6 cheddar      21.1       0.942        9.39    24.9  1.28   105     0   403
head(df, 4)
## # A tibble: 4 x 9
##   type      sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>       <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 blue         18.7       0.8          7.78    21.4  2.34    75     0   353
## 2 brick        18.8       0.784        8.60    23.2  2.79    94     0   371
## 3 brie         17.4       0.826        8.01    20.8  0.45   100     0   334
## 4 camembert    15.3       0.724        7.02    19.8  0.46    72     0   300
head(df, 8)
## # A tibble: 8 x 9
##   type      sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>       <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 blue         18.7       0.8          7.78    21.4  2.34    75     0   353
## 2 brick        18.8       0.784        8.60    23.2  2.79    94     0   371
## 3 brie         17.4       0.826        8.01    20.8  0.45   100     0   334
## 4 camembert    15.3       0.724        7.02    19.8  0.46    72     0   300
## 5 caraway      18.6       0.83         8.28    25.2  3.06    93     0   376
## 6 cheddar      21.1       0.942        9.39    24.9  1.28   105     0   403
## 7 cheshire     19.5       0.87         8.67    23.4  4.78   103     0   387
## 8 colby        20.2       0.953        9.28    23.8  2.57    95     0   394

You can do the same thing with tail() to look at the bottom rows. The default again is 6, but you can tell it how many rows you wish to look at:

tail(df)
## # A tibble: 6 x 9
##   type           sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>            <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 past process,…   3.30        0.18        1.35     25.5   4.3    35     0   170
## 2 cottage,lowfa…   0.619       0.039       0.282    10.9   3       3     0    67
## 3 past process,…  19.7         0.988       8.93     22.2   1.6    94     0   375
## 4 swiss,low sod…  17.7         0.968       7.26     28.4   3.4    92     0   376
## 5 swiss,low fat    3.30        0.18        1.35     28.4   3.4    35     0   179
## 6 mozzarella,lo…  10.9         0.509       4.84     27.5   3.1    54     0   280
tail(df, 10)
## # A tibble: 10 x 9
##    type          sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##    <chr>           <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
##  1 past process…   4.41        0.222       2.00     24.6   3.5    35     0   180
##  2 american che…   8.79        0.409       4.10     16.7  11.6    36     0   239
##  3 parmesan,lo …  19.1         0.659       8.72     41.6   3.7    79     0   456
##  4 cottage,lowf…   0.632       0.031       0.284    12.4   2.7     4     0    72
##  5 past process…   3.30        0.18        1.35     25.5   4.3    35     0   170
##  6 cottage,lowf…   0.619       0.039       0.282    10.9   3       3     0    67
##  7 past process…  19.7         0.988       8.93     22.2   1.6    94     0   375
##  8 swiss,low so…  17.7         0.968       7.26     28.4   3.4    92     0   376
##  9 swiss,low fat   3.30        0.18        1.35     28.4   3.4    35     0   179
## 10 mozzarella,l…  10.9         0.509       4.84     27.5   3.1    54     0   280

We can also get various dimensions of a dataframe. nrow() gets the number of rows (also termed observations), ncol() gets the number of columns (also called variables), dim() returns both the number of rows and the number of columns, and length() is another way to get the number of columns.

nrow(df)  
## [1] 73
ncol(df)  
## [1] 9
dim(df) 
## [1] 73  9
length(df) 
## [1] 9

As we showed earlier, colnames() is a very useful function for retrieving the names of columns in a dataframe. This is particularly useful when you can’t fit all of the columns onto your screen which is common with big datasets.

colnames(df)
## [1] "type"        "sat_fat"     "polysat_fat" "monosat_fat" "protein"    
## [6] "carb"        "chol"        "fiber"       "kcal"

Sometimes you may wish to change column names. This is common when you’ve imported some messy data from elsewhere. Hopefully if it is your own data you named your columns well in the first place.

The way to change column names is to assign the new name to the appropriate column, by indicating which column with square brackets []:

colnames(df)[6] <- "carbo"

colnames(df)
## [1] "type"        "sat_fat"     "polysat_fat" "monosat_fat" "protein"    
## [6] "carbo"       "chol"        "fiber"       "kcal"

In the above example, we changed the name of the 6th column to carbo. You can change all column names by doing the following:

colnames(df) <- c("type1", "sat_fat1", "polysat_fat1", "monosat_fat1", "protein1", "carb1", "chol1", "fiber1", "kcal1")

head(df)
## # A tibble: 6 x 9
##   type1     sat_fat1 polysat_fat1 monosat_fat1 protein1 carb1 chol1 fiber1 kcal1
##   <chr>        <dbl>        <dbl>        <dbl>    <dbl> <dbl> <dbl>  <dbl> <dbl>
## 1 blue          18.7        0.8           7.78     21.4  2.34    75      0   353
## 2 brick         18.8        0.784         8.60     23.2  2.79    94      0   371
## 3 brie          17.4        0.826         8.01     20.8  0.45   100      0   334
## 4 camembert     15.3        0.724         7.02     19.8  0.46    72      0   300
## 5 caraway       18.6        0.83          8.28     25.2  3.06    93      0   376
## 6 cheddar       21.1        0.942         9.39     24.9  1.28   105      0   403

Let’s return them back to the original values:

colnames(df) <- c("type", "sat_fat", "polysat_fat", "monosat_fat", "protein", "carb", "chol", "fiber", "kcal")

head(df)
## # A tibble: 6 x 9
##   type      sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>       <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 blue         18.7       0.8          7.78    21.4  2.34    75     0   353
## 2 brick        18.8       0.784        8.60    23.2  2.79    94     0   371
## 3 brie         17.4       0.826        8.01    20.8  0.45   100     0   334
## 4 camembert    15.3       0.724        7.02    19.8  0.46    72     0   300
## 5 caraway      18.6       0.83         8.28    25.2  3.06    93     0   376
## 6 cheddar      21.1       0.942        9.39    24.9  1.28   105     0   403



4.3.2 Indexing dataframes.

There are two indexing methods we need to learn. One is the $ sign, which indicates we want to get data from a specific column. The other is the square brackets [].

The $ indicates which column to call. For instance, if we wish to get all the data from the chol column of the dataset df, then we would type df$chol and it returns all the data in that columns:

df$chol
##  [1]  75  94 100  72  93 105 103  95  17  13   7  10   4 110  89  89 116  94 114
## [20] 110  90  89  79  89  64  54  96  74  88  68 123  69  51  31 104  90  92 102
## [39]  94  94  85  72  94  85 105  79  46 105 105 105  21 100  12  88  55  62  65
## [58]  11   4  63  18  14  54  35  36  79   4  35   3  94  92  35  54

What if you just want the first 10 rows of kcal? Then we could get all the data from that column with df$kcal but then use the square brackets to tell it to get the values in positions 1 to 10 with [1:10]:

df$kcal[1:10]
##  [1] 353 371 334 300 376 403 387 394  98  97

This works, because you can think of all the data values in each column as essentially its own vector. The square brackets work just as they would with any other vector indexing (see section 3.3).

Square brackets can be used on the whole dataframe to call just certain rows and columns. Importantly row numbers need to be written before the comma, and columns after comma. If you leave anything blank, then it will assume that you mean “all” of the rows or columns.

Technically, running df, and df[,] all return the whole dataframe. What you’re saying is return all of the rows and all of the columns from df.

df
## # A tibble: 73 x 9
##    type          sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##    <chr>           <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
##  1 blue            18.7        0.8         7.78     21.4  2.34    75 0       353
##  2 brick           18.8        0.784       8.60     23.2  2.79    94 0       371
##  3 brie            17.4        0.826       8.01     20.8  0.45   100 0       334
##  4 camembert       15.3        0.724       7.02     19.8  0.46    72 0       300
##  5 caraway         18.6        0.83        8.28     25.2  3.06    93 0       376
##  6 cheddar         21.1        0.942       9.39     24.9  1.28   105 0       403
##  7 cheshire        19.5        0.87        8.67     23.4  4.78   103 0       387
##  8 colby           20.2        0.953       9.28     23.8  2.57    95 0       394
##  9 cottage,crmd…    1.72       0.123       0.778    11.1  3.38    17 0        98
## 10 cottage,crmd…    2.31       0.124       1.04     10.7  4.61    13 0.200    97
## # … with 63 more rows
df[,] #return all rows and all columns
## # A tibble: 73 x 9
##    type          sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##    <chr>           <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
##  1 blue            18.7        0.8         7.78     21.4  2.34    75 0       353
##  2 brick           18.8        0.784       8.60     23.2  2.79    94 0       371
##  3 brie            17.4        0.826       8.01     20.8  0.45   100 0       334
##  4 camembert       15.3        0.724       7.02     19.8  0.46    72 0       300
##  5 caraway         18.6        0.83        8.28     25.2  3.06    93 0       376
##  6 cheddar         21.1        0.942       9.39     24.9  1.28   105 0       403
##  7 cheshire        19.5        0.87        8.67     23.4  4.78   103 0       387
##  8 colby           20.2        0.953       9.28     23.8  2.57    95 0       394
##  9 cottage,crmd…    1.72       0.123       0.778    11.1  3.38    17 0        98
## 10 cottage,crmd…    2.31       0.124       1.04     10.7  4.61    13 0.200    97
## # … with 63 more rows

To just get the 7th row, you put a 7 before the comma, and leave after the comma blank:

df[7,]
## # A tibble: 1 x 9
##   type     sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>      <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 cheshire    19.5        0.87        8.67    23.4  4.78   103     0   387

To get the 10th to 14th rows you can put 10:14 before the comma and leave after the comma blank:

df[10:14,]
## # A tibble: 5 x 9
##   type           sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>            <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 cottage,crmd,…   2.31        0.124       1.04    10.7   4.61    13 0.200    97
## 2 cottage,nonfa…   0.169       0.003       0.079   10.3   6.66     7 0        72
## 3 cottage,lowfa…   0.979       0.07        0.443   11.8   3.66    10 0        86
## 4 cottage,lowfa…   0.645       0.031       0.291   12.4   2.72     4 0        72
## 5 cream           19.3         1.44        8.62     5.93  4.07   110 0       342

To get the 3rd column you can leave before the comma blank, and put a 3 after the comma:

df[,3]
## # A tibble: 73 x 1
##    polysat_fat
##          <dbl>
##  1       0.8  
##  2       0.784
##  3       0.826
##  4       0.724
##  5       0.83 
##  6       0.942
##  7       0.87 
##  8       0.953
##  9       0.123
## 10       0.124
## # … with 63 more rows

To get the first and second columns, leave before the comma blank and put a 1:2 after the comma:

df[,1:2]
## # A tibble: 73 x 2
##    type                         sat_fat
##    <chr>                          <dbl>
##  1 blue                           18.7 
##  2 brick                          18.8 
##  3 brie                           17.4 
##  4 camembert                      15.3 
##  5 caraway                        18.6 
##  6 cheddar                        21.1 
##  7 cheshire                       19.5 
##  8 colby                          20.2 
##  9 cottage,crmd,lrg or sml curd    1.72
## 10 cottage,crmd,w/fruit            2.31
## # … with 63 more rows

If you want to get non-consecutive columns, you need to use c() with your numbers. For example, to get the 3rd, 5th, and 9th columns you leave before the comma blank and put c(3,5,9) after the comma:

df[,c(3,5,9)]
## # A tibble: 73 x 3
##    polysat_fat protein  kcal
##          <dbl>   <dbl> <dbl>
##  1       0.8      21.4   353
##  2       0.784    23.2   371
##  3       0.826    20.8   334
##  4       0.724    19.8   300
##  5       0.83     25.2   376
##  6       0.942    24.9   403
##  7       0.87     23.4   387
##  8       0.953    23.8   394
##  9       0.123    11.1    98
## 10       0.124    10.7    97
## # … with 63 more rows

You can also combine these. So, to get the 20th to 22nd row, and the 1st, 5th and 9th column, you put 20:22 before the comma, and c(1,5,9) after the comma:

df[20:22,c(1,5,9)]
## # A tibble: 3 x 3
##   type      protein  kcal
##   <chr>       <dbl> <dbl>
## 1 gruyere      29.8   413
## 2 limburger    20.0   327
## 3 monterey     24.5   373

4.3.3 Adding and removing columns

As we will see below in the tidyverse section, there is a way to creating new columns using the function mutate() which we recommend. However, there is also a quick way to create and delete new columns which is worth knowing about.

To create a new column, you can simply type a new name after writing your dataframe name and the dollar sign, and then assign something to it. So, if you wished to create a column called food_type you’d write df$food_type <- and then put whatever you wanted to put into that column. For example, if you wanted it to contain the word ‘cheese’ you’d do the following:

df$food_type <- "cheese"

head(df)
## # A tibble: 6 x 10
##   type  sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>   <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 blue     18.7       0.8          7.78    21.4  2.34    75     0   353
## 2 brick    18.8       0.784        8.60    23.2  2.79    94     0   371
## 3 brie     17.4       0.826        8.01    20.8  0.45   100     0   334
## 4 came…    15.3       0.724        7.02    19.8  0.46    72     0   300
## 5 cara…    18.6       0.83         8.28    25.2  3.06    93     0   376
## 6 ched…    21.1       0.942        9.39    24.9  1.28   105     0   403
## # … with 1 more variable: food_type <chr>

Now every observation (row) has the entry cheese in the column food_type.

If you wanted to put in different data for each observation, e.g. the country of origin of each cheese, then you’d need a vector that was the same length as the number of rows of the dataset. See the section on manually creating dataframes just below for some examples of this.

To delete a column from a dataframe, you just assign the word NULL, which is a special term in R, to that column, and then it will disappear:

df$food_type <- NULL

head(df)
## # A tibble: 6 x 9
##   type      sat_fat polysat_fat monosat_fat protein  carb  chol fiber  kcal
##   <chr>       <dbl>       <dbl>       <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 blue         18.7       0.8          7.78    21.4  2.34    75     0   353
## 2 brick        18.8       0.784        8.60    23.2  2.79    94     0   371
## 3 brie         17.4       0.826        8.01    20.8  0.45   100     0   334
## 4 camembert    15.3       0.724        7.02    19.8  0.46    72     0   300
## 5 caraway      18.6       0.83         8.28    25.2  3.06    93     0   376
## 6 cheddar      21.1       0.942        9.39    24.9  1.28   105     0   403


4.3.4 Structure of Datasets

This little subsection is a bit more about the inner workings of dataframes. It won’t be coding that we use during this course, but if you ever do any extra R things by yourself, you may run into issues that this section could help you resolve.

You can see the struture of your data, whether each variable is a number, character, factor, logical, etc. This can be useful when trying to graph and analysis different types of data.

str(bmi)
## tibble [20 × 9] (S3: spec_tbl_df/tbl_df/tbl/data.frame)
##  $ id     : num [1:20] 1 2 3 4 5 6 7 8 9 10 ...
##  $ age    : num [1:20] 18.3 19.4 21.1 21.3 21.1 ...
##  $ bmi    : num [1:20] 15.4 14.5 16 19.5 18.6 ...
##  $ chol   : num [1:20] 47.2 52.1 52.2 53.2 55.4 ...
##  $ insulin: num [1:20] 2.78 3.47 4.06 4.48 5.34 ...
##  $ dbp    : num [1:20] 29.5 31.3 32.4 32 33.7 ...
##  $ sbp    : num [1:20] 49.7 49 51.7 NA 53.8 ...
##  $ smoke  : num [1:20] 1 2 1 2 1 2 1 2 1 2 ...
##  $ educ   : chr [1:20] "D" "C" "B" "A" ...
##  - attr(*, "spec")=
##   .. cols(
##   ..   id = col_double(),
##   ..   age = col_double(),
##   ..   bmi = col_double(),
##   ..   chol = col_double(),
##   ..   insulin = col_double(),
##   ..   dbp = col_double(),
##   ..   sbp = col_double(),
##   ..   smoke = col_double(),
##   ..   educ = col_character()
##   .. )

Here you can see that all variables are numeric, except for education with is a character.

Another common variable type is factor. This is similar to a character but it has levels. This means, that R knows that there are groups in that variable. You might notice that the variable smoke is currently a numerical variable with values being either a 1 (non smoker) or 2 (smoker). If we wanted to make a graph and plot non-smoker or smoker on the x-axis, then this would appear as 1 or 2, which isn’t helpful. In these situations, it can be helpful to convert our numeric variable into a factor. The way to do that is as follows:

bmi$smoke <- as.factor(bmi$smoke)

The $ basically allows you to call certain columns in a dataframe. The code bmi$smoke is allowing us to only change that column, or variable to a factor without changing anything else in the dataframe.

You can also change variables to characters with as.character() and to numbers with as.numeric().


4.4 Manually creating a Dataframe

Often in R, we don’t want to import a dataset, but rather create a dataset of our own manually in the script. We can do that using the function data.frame().

Let’s just do a small example. Say, we want to create a dataframe with four columns. We will have the names of 6 different pets in column 1 and call that column name. In the second column, we want the ages of those pets and we’ll call that column age. In the third column we will have the type of animal that pet is, and we’ll call that column animal. In the fourth column, we’ll have whether it’s a male or female pet and we’ll call that column sex. In the fifth column, we’ll have the main color of the pet, and we’ll call that column color.

Here are our 6 pets:

  1. Steve, an orange male goldfish who is 5.
  2. Hannah, a female blue parrot who is 12.
  3. Colin, a brown male cat who is 15.
  4. Archibald, a grouchy green male terrapin who is 3.
  5. Missy, a female yellow labrador dog who is 2.
  6. Bob, a black male spider who is 10.

We need to make sure we put each of the column information in the right order in our dataframe. This is how we manually enter the data:

petdf <-
  data.frame(
    name   = c("Steve", "Hannah", "Colin", "Archibald", "Missy", "Bob"),
    age    = c(5, 12, 15, 3, 2, 10),
    animal = c("goldfish", "parrot", "cat", "terrapin", "dog", "spider"),
    sex    = c("M", "F", "M", "M", "F", "M"),
    color  = c("orange", "blue", "brown", "green", "yellow", "black")
  )

petdf
##        name age   animal sex  color
## 1     Steve   5 goldfish   M orange
## 2    Hannah  12   parrot   F   blue
## 3     Colin  15      cat   M  brown
## 4 Archibald   3 terrapin   M  green
## 5     Missy   2      dog   F yellow
## 6       Bob  10   spider   M  black

Inside dataframe() we write each column name. Then we put an = sign, and then write the vector of characters or numbers. After each column’s data is entered, we write a comma , to indicate we’re going to the next column. The exception to this is after the last column is entered, color in this case, where we don’t write a comma after we’re done entering all the color information. This indicates that we’re done. Also note that every column has the same number of pieces of information (6). If we had unequal bits of information, R would generally not allow us to make the dataframe. There is one gotcha here though - if you enter less than the number of rows of your new dataframe, e.g. if you’d only entered two values in the color column say, R will repeat those two colors through the remaining empty rows. You need to be careful! If you’re not sure what you’re doing, the best is to ensure that you have the exact number of values as you want rows inside each vector.





4.5 tidyverse

The package tidyverse is what we will be using for most of our data carpentry. tidyverse is a larger package which includes several packages useful for managing, exploring and visualizing data such as tidyr, dplyr, ggplot2, and more.

In this section, we will learn the following functions and syntax:

filter() - subsetting data

select() - selecting columns

arrange() - sorting a column

mutate() - adding a new column

count() - counts the number of values in a column

%>% means “and next do this”

== means “is equal to”

!= means “is not equal to”

| means “or”

First, make sure that you have the tidyverse installed. If you run the following code and do not get an error message saying “tidyverse not found”, then you are good:

library(tidyverse)

If you did get the error message, then you’ll need to install the library. You only need to do this once, but you will have to load a library on every script you plan to use it.

To install the package:

install.package(tidyverse) #install

To load the package to use each time you need it:

library(tidyverse)

From the tidyverse package readr we can read in our data. We are going to work with a dataset called bloodwork that we will shorten to bw. It contains health information on several subjects:

library(tidyverse) #load
bw <- read_csv("data/bloodwork.csv")
head(bw) 
## # A tibble: 6 x 11
##   ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##   <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
## 1 JEC      19 fema… NJ           1 no        63    101     0.126      0.993
## 2 GBH      20 male  NY           1 yes       73    120     0.169      1.18 
## 3 EDH      21 fema… NJ           0 no        65    100     0.281      4.34 
## 4 AAA      21 fema… CT           3 no        66    109     0.244      2.56 
## 5 AJF      24 fema… NJ           0 no        67    108     0.092      6.45 
## 6 FJC      25 fema… NY           1 yes       80    118     0.014      3.97 
## # … with 1 more variable: immuncount2 <dbl>
tail(bw) 
## # A tibble: 6 x 11
##   ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##   <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
## 1 JIB      66 fema… NY           0 no        62    121     0.097      1.39 
## 2 HBB      67 fema… NJ           1 yes       74    147     0.288      2.27 
## 3 HDG      68 fema… NY           3 yes       65    129     0.11       2.65 
## 4 ECD      68 fema… NJ           2 yes       77    129     0.404      2.02 
## 5 HHJ      69 male  CT           2 no        71    121     0.475      0.463
## 6 CCG      70 male  CT           0 yes       80    132     0.078      1.06 
## # … with 1 more variable: immuncount2 <dbl>

As you can see, we have variables such as ids, age, the state people live in, their heart rate, how many children they have etc. etc.

4.5.1 table()

This function is not a tidyverse function but is a quick summary function that is useful to know - table(). It is good for quickly summarizing categorical or discrete numerical variables.

For example, to look at the number of smokers and non-smokers, we can do the following:

table(bw$smoker)
## 
##  no yes 
##  15  15

We have 15 non-smokers and 15 smokers in the dataset.

To look at the frequency count of how many subjects have 0, 1, 2, 3 children we can do:

table(bw$children)
## 
##  0  1  2  3 
## 13 10  5  2

We have thirteen people with 0 children ten with 1 child, five with 2 children and two with 3 children.

You can also compare two categories at once. For instance, to look at the smokers and non-smokers that have different numbers of children, we can include both in the table() function and separate by a comma:

table(bw$smoker, bw$children)
##      
##       0 1 2 3
##   no  7 4 3 1
##   yes 6 6 2 1

The tidyverse version of table() is to use a function called count(). It does come in useful sometimes, but most of the time you’ll find using table() to be easier. However, here is an example of counting how many individual by each state there are in the data. You first take your dataset, then chain with the pipe %>% the next command which is to count() the column state.

bw %>%
  count(state)
## # A tibble: 3 x 2
##   state     n
##   <chr> <int>
## 1 CT       12
## 2 NJ       10
## 3 NY        8

We can also get frequency counts for more than one column e.g. to see how many children non-smokers and smokers have, though again, the output from table() is nicer to look at:

bw %>%
  count(smoker, children)
## # A tibble: 8 x 3
##   smoker children     n
##   <chr>     <dbl> <int>
## 1 no            0     7
## 2 no            1     4
## 3 no            2     3
## 4 no            3     1
## 5 yes           0     6
## 6 yes           1     6
## 7 yes           2     2
## 8 yes           3     1



4.5.2 filter() - Subsetting Data

filter() is a way to subset data. In other words, it is a way to only keep certain rows in your dataset. These are rows that must fulfil specific criteria. For example, let’s say we wish to only keep rows that have values in the hrate column that are over 70.

The first step is to take the dataframe bw() and then tell R that you are about to do something else with the %>% syntax. Then use filter() with hrate > 70 inside it. This mean, keep the rows with hrate>70. See sections (chaining-syntax) and 4.5.6 for a bit more on chaining using %>%.

bw %>% 
  filter(hrate > 70)
## # A tibble: 16 x 11
##    ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##    <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
##  1 GBH      20 male  NY           1 yes       73    120     0.169      1.18 
##  2 FJC      25 fema… NY           1 yes       80    118     0.014      3.97 
##  3 IEE      26 fema… NY           2 no        71    118     0.093      5.41 
##  4 IEA      29 fema… CT           0 yes       74    117     0.429      5.15 
##  5 FJG      40 fema… NJ           0 yes       80    109     0.253      5.63 
##  6 AGC      43 male  NY           1 yes       77    108     0.072      1.44 
##  7 CAE      47 male  CT           1 yes       73    122     0.176      0.127
##  8 FGD      50 male  NY           0 yes       77    147     0.428      0.037
##  9 CGC      51 male  CT           0 yes       71    119     0.045      1.73 
## 10 FHA      53 fema… CT           2 no        77    125     0.099      0.034
## 11 JHC      55 fema… CT           0 no        73    121     0.093      1.32 
## 12 CFC      65 male  CT           2 yes       77    129     0.151      2.40 
## 13 HBB      67 fema… NJ           1 yes       74    147     0.288      2.27 
## 14 ECD      68 fema… NJ           2 yes       77    129     0.404      2.02 
## 15 HHJ      69 male  CT           2 no        71    121     0.475      0.463
## 16 CCG      70 male  CT           0 yes       80    132     0.078      1.06 
## # … with 1 more variable: immuncount2 <dbl>

To only keep values that are precisely equal to some value (be it a number or a character), we use filter() with == to mean “is equal to”:

For example, to only keep rows where the state is equal to NJ.

bw %>% 
  filter(state == "NJ")
## # A tibble: 10 x 11
##    ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##    <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
##  1 JEC      19 fema… NJ           1 no        63    101     0.126      0.993
##  2 EDH      21 fema… NJ           0 no        65    100     0.281      4.34 
##  3 AJF      24 fema… NJ           0 no        67    108     0.092      6.45 
##  4 BFB      28 fema… NJ           0 no        68    118     0.197      0.724
##  5 ACC      33 male  NJ           1 no        63    131     0.065      4.66 
##  6 CDC      38 fema… NJ           0 no        66    133     0.038      8.00 
##  7 EEB      39 male  NJ           1 no        68    104     0.594      3.06 
##  8 FJG      40 fema… NJ           0 yes       80    109     0.253      5.63 
##  9 HBB      67 fema… NJ           1 yes       74    147     0.288      2.27 
## 10 ECD      68 fema… NJ           2 yes       77    129     0.404      2.02 
## # … with 1 more variable: immuncount2 <dbl>

If you want to include all subjects that had 3 children, you’d do the following - note that the number does not go in quote marks:

bw %>% 
  filter(children == 3)
## # A tibble: 2 x 11
##   ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##   <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
## 1 AAA      21 fema… CT           3 no        66    109     0.244       2.56
## 2 HDG      68 fema… NY           3 yes       65    129     0.11        2.65
## # … with 1 more variable: immuncount2 <dbl>

If you want to include rows that are equal to one of two values, you can use | to mean “OR”. The following will include rows where the state column is equal to either “NJ” or “NY” in the bloodwork dataset.

bw %>% 
  filter(state == "NJ"  |  state == "NY")
## # A tibble: 18 x 11
##    ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##    <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
##  1 JEC      19 fema… NJ           1 no        63    101     0.126      0.993
##  2 GBH      20 male  NY           1 yes       73    120     0.169      1.18 
##  3 EDH      21 fema… NJ           0 no        65    100     0.281      4.34 
##  4 AJF      24 fema… NJ           0 no        67    108     0.092      6.45 
##  5 FJC      25 fema… NY           1 yes       80    118     0.014      3.97 
##  6 IEE      26 fema… NY           2 no        71    118     0.093      5.41 
##  7 BFB      28 fema… NJ           0 no        68    118     0.197      0.724
##  8 ACC      33 male  NJ           1 no        63    131     0.065      4.66 
##  9 CDC      38 fema… NJ           0 no        66    133     0.038      8.00 
## 10 EEB      39 male  NJ           1 no        68    104     0.594      3.06 
## 11 FJG      40 fema… NJ           0 yes       80    109     0.253      5.63 
## 12 AGC      43 male  NY           1 yes       77    108     0.072      1.44 
## 13 FGD      50 male  NY           0 yes       77    147     0.428      0.037
## 14 JCI      52 male  NY           1 yes       61    115     0.131      0.233
## 15 JIB      66 fema… NY           0 no        62    121     0.097      1.39 
## 16 HBB      67 fema… NJ           1 yes       74    147     0.288      2.27 
## 17 HDG      68 fema… NY           3 yes       65    129     0.11       2.65 
## 18 ECD      68 fema… NJ           2 yes       77    129     0.404      2.02 
## # … with 1 more variable: immuncount2 <dbl>

You can also keep rows that are not equal to some value using the syntax != which means “not equal to”. For example, to keep rows where the children column is not equal to 0:

bw %>% 
  filter(children != "0" )
## # A tibble: 17 x 11
##    ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##    <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
##  1 JEC      19 fema… NJ           1 no        63    101     0.126      0.993
##  2 GBH      20 male  NY           1 yes       73    120     0.169      1.18 
##  3 AAA      21 fema… CT           3 no        66    109     0.244      2.56 
##  4 FJC      25 fema… NY           1 yes       80    118     0.014      3.97 
##  5 IEE      26 fema… NY           2 no        71    118     0.093      5.41 
##  6 BHE      30 male  CT           1 no        63    120     0.215      1.19 
##  7 ACC      33 male  NJ           1 no        63    131     0.065      4.66 
##  8 EEB      39 male  NJ           1 no        68    104     0.594      3.06 
##  9 AGC      43 male  NY           1 yes       77    108     0.072      1.44 
## 10 CAE      47 male  CT           1 yes       73    122     0.176      0.127
## 11 JCI      52 male  NY           1 yes       61    115     0.131      0.233
## 12 FHA      53 fema… CT           2 no        77    125     0.099      0.034
## 13 CFC      65 male  CT           2 yes       77    129     0.151      2.40 
## 14 HBB      67 fema… NJ           1 yes       74    147     0.288      2.27 
## 15 HDG      68 fema… NY           3 yes       65    129     0.11       2.65 
## 16 ECD      68 fema… NJ           2 yes       77    129     0.404      2.02 
## 17 HHJ      69 male  CT           2 no        71    121     0.475      0.463
## # … with 1 more variable: immuncount2 <dbl>

You can also filter several variables at one time. Here, we are keeping all individuals from New York, with a heart rate of over 70, and who have more than 0 children:

bw %>% 
  filter( state == "NY", hrate > 70, children != "0" )
## # A tibble: 4 x 11
##   ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##   <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
## 1 GBH      20 male  NY           1 yes       73    120     0.169       1.18
## 2 FJC      25 fema… NY           1 yes       80    118     0.014       3.97
## 3 IEE      26 fema… NY           2 no        71    118     0.093       5.41
## 4 AGC      43 male  NY           1 yes       77    108     0.072       1.44
## # … with 1 more variable: immuncount2 <dbl>

You can create new datasets from filtered data by creating a new object. Here, we create a new dataset called ny1 that is based on filtering out rows from our bw dataset.

ny1 <- bw %>% 
  filter( state == "NY", hrate > 70, children != "0" )

ny1
## # A tibble: 4 x 11
##   ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##   <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
## 1 GBH      20 male  NY           1 yes       73    120     0.169       1.18
## 2 FJC      25 fema… NY           1 yes       80    118     0.014       3.97
## 3 IEE      26 fema… NY           2 no        71    118     0.093       5.41
## 4 AGC      43 male  NY           1 yes       77    108     0.072       1.44
## # … with 1 more variable: immuncount2 <dbl>



4.5.3 select() - Selecting specific columns

Sometimes datasets are too big and unwieldy to look at all at once. Often, we are just interested in certain columns and it makes more sense just to keep the ones we want to focus on. We can use select() to only keep certain columns:

For example, to only keep the columsn ids, smoker, hrate and children, we’d do the following:

bw %>% 
  select(ids, smoker,hrate,children)
## # A tibble: 30 x 4
##    ids   smoker hrate children
##    <chr> <chr>  <dbl>    <dbl>
##  1 JEC   no        63        1
##  2 GBH   yes       73        1
##  3 EDH   no        65        0
##  4 AAA   no        66        3
##  5 AJF   no        67        0
##  6 FJC   yes       80        1
##  7 IEE   no        71        2
##  8 BED   no        62        0
##  9 BFB   no        68        0
## 10 IEA   yes       74        0
## # … with 20 more rows

Occasionally, you may just want to get rid of certain columns from your data. To get rid of one column you can use select(-column_name). To get rid of the children column, we’d do the following:

bw %>% 
  select(-children)
## # A tibble: 30 x 10
##    ids     age sex    state smoker hrate bpsyst cellcount immuncount immuncount2
##    <chr> <dbl> <chr>  <chr> <chr>  <dbl>  <dbl>     <dbl>      <dbl>       <dbl>
##  1 JEC      19 female NJ    no        63    101     0.126      0.993       0.921
##  2 GBH      20 male   NY    yes       73    120     0.169      1.18        1.19 
##  3 EDH      21 female NJ    no        65    100     0.281      4.34        3.21 
##  4 AAA      21 female CT    no        66    109     0.244      2.56        4.01 
##  5 AJF      24 female NJ    no        67    108     0.092      6.45        9.13 
##  6 FJC      25 female NY    yes       80    118     0.014      3.97        2.85 
##  7 IEE      26 female NY    no        71    118     0.093      5.41       10.2  
##  8 BED      28 female CT    no        62    104     0.082      1.18        0.788
##  9 BFB      28 female NJ    no        68    118     0.197      0.724       0.848
## 10 IEA      29 female CT    yes       74    117     0.429      5.15        4.97 
## # … with 20 more rows

To get rid of the children, bpsyst and smoker columns, we’d do the following:

bw %>% 
  select(-children, -bpsyst, -smoker)
## # A tibble: 30 x 8
##    ids     age sex    state hrate cellcount immuncount immuncount2
##    <chr> <dbl> <chr>  <chr> <dbl>     <dbl>      <dbl>       <dbl>
##  1 JEC      19 female NJ       63     0.126      0.993       0.921
##  2 GBH      20 male   NY       73     0.169      1.18        1.19 
##  3 EDH      21 female NJ       65     0.281      4.34        3.21 
##  4 AAA      21 female CT       66     0.244      2.56        4.01 
##  5 AJF      24 female NJ       67     0.092      6.45        9.13 
##  6 FJC      25 female NY       80     0.014      3.97        2.85 
##  7 IEE      26 female NY       71     0.093      5.41       10.2  
##  8 BED      28 female CT       62     0.082      1.18        0.788
##  9 BFB      28 female NJ       68     0.197      0.724       0.848
## 10 IEA      29 female CT       74     0.429      5.15        4.97 
## # … with 20 more rows

You can also select and rename columns as you go. Here, we are selecting the columns ids, sex, smoker, hrate and children. We are renaming sex to be gender and ids to be subject. You rename by just typing the new name and then putting an = sign in front of the old name:

bw %>% 
  select(subject = ids, gender = sex, smoker, hrate, children)
## # A tibble: 30 x 5
##    subject gender smoker hrate children
##    <chr>   <chr>  <chr>  <dbl>    <dbl>
##  1 JEC     female no        63        1
##  2 GBH     male   yes       73        1
##  3 EDH     female no        65        0
##  4 AAA     female no        66        3
##  5 AJF     female no        67        0
##  6 FJC     female yes       80        1
##  7 IEE     female no        71        2
##  8 BED     female no        62        0
##  9 BFB     female no        68        0
## 10 IEA     female yes       74        0
## # … with 20 more rows

If you want these selections to be permanent then you need to rewrite selections in new dataframe. Here we call our new datafame bw1. You can see the difference between bw and bw1:

bw1 <- bw %>%  
  select(subject = ids, gender = sex, smoker,hrate,children)

head(bw)
## # A tibble: 6 x 11
##   ids     age sex   state children smoker hrate bpsyst cellcount immuncount
##   <chr> <dbl> <chr> <chr>    <dbl> <chr>  <dbl>  <dbl>     <dbl>      <dbl>
## 1 JEC      19 fema… NJ           1 no        63    101     0.126      0.993
## 2 GBH      20 male  NY           1 yes       73    120     0.169      1.18 
## 3 EDH      21 fema… NJ           0 no        65    100     0.281      4.34 
## 4 AAA      21 fema… CT           3 no        66    109     0.244      2.56 
## 5 AJF      24 fema… NJ           0 no        67    108     0.092      6.45 
## 6 FJC      25 fema… NY           1 yes       80    118     0.014      3.97 
## # … with 1 more variable: immuncount2 <dbl>
head(bw1)
## # A tibble: 6 x 5
##   subject gender smoker hrate children
##   <chr>   <chr>  <chr>  <dbl>    <dbl>
## 1 JEC     female no        63        1
## 2 GBH     male   yes       73        1
## 3 EDH     female no        65        0
## 4 AAA     female no        66        3
## 5 AJF     female no        67        0
## 6 FJC     female yes       80        1

Instead of typing out the column name each time you use select(), you can also use the column number.

The code below selects for columns 1-2, 8, 10, but does not save the information as a new object. Notice that you don’t need to put these numbers inside of c() for this to work:

bw %>% select(1:2,8,10)
## # A tibble: 30 x 4
##    ids     age bpsyst immuncount
##    <chr> <dbl>  <dbl>      <dbl>
##  1 JEC      19    101      0.993
##  2 GBH      20    120      1.18 
##  3 EDH      21    100      4.34 
##  4 AAA      21    109      2.56 
##  5 AJF      24    108      6.45 
##  6 FJC      25    118      3.97 
##  7 IEE      26    118      5.41 
##  8 BED      28    104      1.18 
##  9 BFB      28    118      0.724
## 10 IEA      29    117      5.15 
## # … with 20 more rows



4.5.4 mutate() - Creating new columns

We use mutate() to add new columns to our dataset. Say we wanted to create a new column called totalimmune that is the sum of the two columns immuncount and immuncount2. Basically, we want to add each subject’s immuncount and immuncount2 value together to create a new value called totalimmune that we’ll put into the new column.

The first thing you put inside mutate() is the name we want of the new column. immuncount + immuncount means to add these two columns.

To make these easier to see, we’ll just select three columns and call the new dataset bw2:

bw2 <- bw %>%
  select(ids, immuncount, immuncount2)

bw2
## # A tibble: 30 x 3
##    ids   immuncount immuncount2
##    <chr>      <dbl>       <dbl>
##  1 JEC        0.993       0.921
##  2 GBH        1.18        1.19 
##  3 EDH        4.34        3.21 
##  4 AAA        2.56        4.01 
##  5 AJF        6.45        9.13 
##  6 FJC        3.97        2.85 
##  7 IEE        5.41       10.2  
##  8 BED        1.18        0.788
##  9 BFB        0.724       0.848
## 10 IEA        5.15        4.97 
## # … with 20 more rows
bw2 %>% 
  mutate(totalimmune = immuncount + immuncount2)
## # A tibble: 30 x 4
##    ids   immuncount immuncount2 totalimmune
##    <chr>      <dbl>       <dbl>       <dbl>
##  1 JEC        0.993       0.921        1.91
##  2 GBH        1.18        1.19         2.37
##  3 EDH        4.34        3.21         7.55
##  4 AAA        2.56        4.01         6.57
##  5 AJF        6.45        9.13        15.6 
##  6 FJC        3.97        2.85         6.82
##  7 IEE        5.41       10.2         15.6 
##  8 BED        1.18        0.788        1.97
##  9 BFB        0.724       0.848        1.57
## 10 IEA        5.15        4.97        10.1 
## # … with 20 more rows

You can also create new columns in other ways. For example, if you want to create a new column called year and put 2020 into each row, we’d do the following:

bw2 %>% 
  mutate(year = 2020)
## # A tibble: 30 x 4
##    ids   immuncount immuncount2  year
##    <chr>      <dbl>       <dbl> <dbl>
##  1 JEC        0.993       0.921  2020
##  2 GBH        1.18        1.19   2020
##  3 EDH        4.34        3.21   2020
##  4 AAA        2.56        4.01   2020
##  5 AJF        6.45        9.13   2020
##  6 FJC        3.97        2.85   2020
##  7 IEE        5.41       10.2    2020
##  8 BED        1.18        0.788  2020
##  9 BFB        0.724       0.848  2020
## 10 IEA        5.15        4.97   2020
## # … with 20 more rows



4.5.5 arrange() - Sort Data Columns

Often it’s easier to see data if the columns are sorted in ascending or descending order. We can do this using arrange().

Lets try another example! We’ll use the pga.csv dataset that has historical golf data summary statistics in it. Let’s load in that dataset:

pga <- read_csv("data/pga.csv")

We will select the columns name, year, total.holes, total.putts, and score.avg and save as pga1.

pga1 <- pga %>% select(name, year, total.holes, total.putts, score.avg)  

head(pga1)
## # A tibble: 6 x 5
##   name            year total.holes total.putts score.avg
##   <chr>          <dbl>       <dbl>       <dbl>     <dbl>
## 1 Fred Funk       2004        1728        2774      70.4
## 2 Scott Verplank  2004        1638        2619      69.9
## 3 Craig Bowden    2004        1494        2401      71.5
## 4 Joe Durant      2004        1530        2563      70.7
## 5 Tom Byrum       2004        1494        2380      70.5
## 6 Jose Coceres    2004        1152        1819      71.0

You can see that we have the total number of holes played, the total number of putts made, and the scoring average (lower is better in golf) made each year by various PGA golfers.

Perhaps we want to know who has the highest or lowest scoring average. We can sort based on the column score.avg in ascending order with arrange() like this:

pga1 %>% 
  arrange(score.avg)  
## # A tibble: 2,269 x 5
##    name            year total.holes total.putts score.avg
##    <chr>          <dbl>       <dbl>       <dbl>     <dbl>
##  1 Tiger Woods     2007        1080        1736      67.8
##  2 Tiger Woods     2009        1116        1763      68.1
##  3 Tiger Woods     2006         936        1528      68.1
##  4 Tiger Woods     2005        1332        2124      68.7
##  5 Rory McIlroy    2014        1152        1830      68.8
##  6 Vijay Singh     2004        1980        3216      68.8
##  7 Luke Donald     2011        1206        1878      68.9
##  8 Jim Furyk       2006        1584        2539      68.9
##  9 Rory McIlroy    2012         972        1551      68.9
## 10 Steve Stricker  2013         846        1348      68.9
## # … with 2,259 more rows

So, the lowest average score was 67.794 by Tiger Woods in 2007. The second lowest was 68.052 by Tiger Woods in 2009.

To sort data in descending order, we need to put a negative sign (hyphen) - in front of the column name:

pga1 %>% 
  arrange(-score.avg)  
## # A tibble: 2,269 x 5
##    name              year total.holes total.putts score.avg
##    <chr>            <dbl>       <dbl>       <dbl>     <dbl>
##  1 David Gossett     2004        1026        1693      75.0
##  2 Kevin Muncrief    2004         900        1490      73.5
##  3 Chris Couch       2004         990        1635      73.5
##  4 Hidemichi Tanaka  2006        1314        2136      73.5
##  5 David Duval       2008         918        1497      73.2
##  6 Brad Faxon        2010         972        1567      73.1
##  7 Eric Axley        2009        1206        1916      73.1
##  8 Hirofumi Miyase   2004        1134        1857      73.1
##  9 Greg Kraft        2010         828        1370      73.1
## 10 Stephen Gangluff  2012         918        1507      73.0
## # … with 2,259 more rows

The worst average score on the PGA tour was by David Gossett in 2004 with an average of 75.013.

arrange() can also sort data in ascending alphabetical order, if you put a column with character data into the function, such as the name column in our data:

pga1 %>% 
  arrange(name) 
## # A tibble: 2,269 x 5
##    name            year total.holes total.putts score.avg
##    <chr>          <dbl>       <dbl>       <dbl>     <dbl>
##  1 Aaron Baddeley  2004        1530        2411      71.6
##  2 Aaron Baddeley  2005        1476        2303      71.3
##  3 Aaron Baddeley  2006        1368        2125      71.2
##  4 Aaron Baddeley  2007        1440        2261      70.1
##  5 Aaron Baddeley  2008        1314        2072      70.2
##  6 Aaron Baddeley  2009        1170        1826      71.2
##  7 Aaron Baddeley  2010        1692        2686      71.0
##  8 Aaron Baddeley  2011        1350        2129      70.2
##  9 Aaron Baddeley  2012        1296        2021      71.1
## 10 Aaron Baddeley  2013        1188        1835      71.5
## # … with 2,259 more rows

It turns out that Aaron Baddeley is the player that is closest to the beginning of the alphabet. You cannot sort in a descending way on character data.

If you wish to sort over multiple columns, you just need to separate your column names with a comma inside arrange(). So, to first sort by year, and then by score.avg, we’d do the following:

pga1 %>% 
  arrange(year, score.avg)  
## # A tibble: 2,269 x 5
##    name            year total.holes total.putts score.avg
##    <chr>          <dbl>       <dbl>       <dbl>     <dbl>
##  1 Vijay Singh     2004        1980        3216      68.8
##  2 Ernie Els       2004        1044        1653      69.0
##  3 Tiger Woods     2004        1296        2048      69.0
##  4 Phil Mickelson  2004        1422        2287      69.2
##  5 Retief Goosen   2004         990        1580      69.3
##  6 Sergio Garcia   2004        1152        1919      69.8
##  7 Stewart Cink    2004        1746        2731      69.8
##  8 Stephen Ames    2004        1710        2756      69.9
##  9 Scott Verplank  2004        1638        2619      69.9
## 10 David Toms      2004        1332        2174      70.0
## # … with 2,259 more rows

You can see that the lowest year in this dataset is 2004, and so it put all the rows with 2004 at the top, and then sorted each of these rows by score.avg. So, Vijay Singh had the lowest score with 68.839 in 2004.



4.5.6 Chaining together

Just as an extra note, the real power of these tidyverse commands, is by chaining them all together. So, one thing we could do is select from our original pga dataset the columns above, then create a new column called putt.avg which is equal to the total number of putts total.putts divided by the total number of holes total.holes columns, and then arrange in ascending order by the new putt.avg column. Here it is in one bit of code:

pga %>%
  select(name, year, total.holes, total.putts, score.avg)  %>%
  mutate(putt.avg = total.putts / total.holes) %>%
  arrange(putt.avg)
## # A tibble: 2,269 x 6
##    name            year total.holes total.putts score.avg putt.avg
##    <chr>          <dbl>       <dbl>       <dbl>     <dbl>    <dbl>
##  1 Brian Gay       2013        1422        2173      71.2     1.53
##  2 Kevin Na        2011        1512        2331      70.4     1.54
##  3 Greg Chalmers   2011        1548        2388      70.6     1.54
##  4 Justin Leonard  2014        1350        2083      70.9     1.54
##  5 Steve Stricker  2005        1116        1723      71.1     1.54
##  6 Aaron Baddeley  2013        1188        1835      71.5     1.54
##  7 Jordan Spieth   2015        1584        2448      68.9     1.55
##  8 Jordan Spieth   2014        1764        2730      69.9     1.55
##  9 Tim Clark       2007        1152        1784      69.9     1.55
## 10 Jonas Blixt     2012        1260        1952      70.2     1.55
## # … with 2,259 more rows

It turns out that Brian Gay in 2013 had the fewest putts per hole.



4.6 Wide versus Long Data

Sometimes you will need to rearrange your data for some data analysis and more commonly for data visualization. Dataframes can come in two broad shapes. There is wide data where each row is a separate subject and each column has variables that may or may not contain the same type of data. The other is long data where values of the same type are only in one column, and the same subject may appear on several rows.

This is easier to understand with a visual aid. The image below show data that is in wide format. Each row contains all the information for each subject - those being three NHL hockey teams. We see the division that each team belongs to, and the total goals they scored in the seasons beginning in 2016, 2017, 2018.

These data are also untidy - as we prefer to have each row contain all the information for one unique value. To do that we need to turn the data into long format which looks like this:

Here, each row contains information related to the team, division, the total number of goals scored, and the year that that piece of data came from. Because each row contains unique information, we call this type of data tidy. Essentially, you can see we took columns 3,4, and 5 from the wide data and made these longer.

It can take a bit of practice to fully recognize what data are in wide or long format - sometimes it’s not so clear in really large datasets. This does become important though, because some visualization and analyses can only be done with data in wide format, and others with data in long format. It’s therefore important to be able to switch our dataframes through these two formats.

Fortunately, the tidyverse has built in functions that can turn your dataframe from long format to wide format and vice versa.

pivot_longer makes the dataframes longer by increasing the number of rows by combing the number of columns. This is called long form data, which is needed to tidy data for graphing and some analysis.

First, we’ll make some small dataframe examples manually. The first dataframe is a wide dataframe. You could imagine this as showing five subjects and their reaction times at two different timepoints. We call this dataframe df.wide:

# wide data has more than one score of each type per individual in rows

df.wide <- data.frame(
                 name = c("James", "Tyler", "Stephen", "Jennifer", "Carmen"),
                 time1 = c(15, 17, 14, 13, 11),
                 time2 = c(16, 19, 20, 21, 23)
                 )


df.wide
##       name time1 time2
## 1    James    15    16
## 2    Tyler    17    19
## 3  Stephen    14    20
## 4 Jennifer    13    21
## 5   Carmen    11    23

The second dataframe we’ll create manually contains the same data, but in a long tidy format:

# long data (also called tidy data) has only one column for each type of score
# Every row has the information to make each score unique

df.long <- data.frame(
                 name = c("James", "Tyler", "Stephen", "Jennifer", "Carmen"),
                 score = c(15, 17, 14, 13, 11, 16, 19, 20, 21, 23),
                 time = rep(c("time1", "time2"), each = 5)
)

df.long 
##        name score  time
## 1     James    15 time1
## 2     Tyler    17 time1
## 3   Stephen    14 time1
## 4  Jennifer    13 time1
## 5    Carmen    11 time1
## 6     James    16 time2
## 7     Tyler    19 time2
## 8   Stephen    20 time2
## 9  Jennifer    21 time2
## 10   Carmen    23 time2

4.6.1 Wide to Long

To go from wide to long format we can use pivot_longer. We need to tell R which columns contain the data that are currently in wide format and need to be turned into long format. Here, we put cols = 2:3 to indicate that the data are in columns 2 and 3. We also use names_to to tell it that we want the new column to be called time:

# turn columns 2 and 3 into one long column:
df.wide %>% 
  pivot_longer(cols = 2:3, names_to="time")  
## # A tibble: 10 x 3
##    name     time  value
##    <fct>    <chr> <dbl>
##  1 James    time1    15
##  2 James    time2    16
##  3 Tyler    time1    17
##  4 Tyler    time2    19
##  5 Stephen  time1    14
##  6 Stephen  time2    20
##  7 Jennifer time1    13
##  8 Jennifer time2    21
##  9 Carmen   time1    11
## 10 Carmen   time2    23

Our example dataset was very simple, only containing an identifier (name) column and the two columns with data. What if we had other columns in our dataset? Here we add a column called group that indicates if our subject are in a control or treatment group:

df.wide$group <- c("control", "control", "treatment", "treatment", "treatment")

df.wide
##       name time1 time2     group
## 1    James    15    16   control
## 2    Tyler    17    19   control
## 3  Stephen    14    20 treatment
## 4 Jennifer    13    21 treatment
## 5   Carmen    11    23 treatment

To convert this dataframe to a long dataframe, we still use the same code indicating that the data to make long are contained in columns 2 and 3. Because each individual is always in the same group (e.g. James is always a control, Carmen is always in treatment), R will just make this a new column:

df.wide %>% 
  pivot_longer(cols = 2:3, names_to="time")  
## # A tibble: 10 x 4
##    name     group     time  value
##    <fct>    <chr>     <chr> <dbl>
##  1 James    control   time1    15
##  2 James    control   time2    16
##  3 Tyler    control   time1    17
##  4 Tyler    control   time2    19
##  5 Stephen  treatment time1    14
##  6 Stephen  treatment time2    20
##  7 Jennifer treatment time1    13
##  8 Jennifer treatment time2    21
##  9 Carmen   treatment time1    11
## 10 Carmen   treatment time2    23

4.6.2 Long to Wide

If we start with data in the long format, we can convert the dataframe to wide format using pivot_wider(). Here, we need to provide the name of the column that contains the data to become new wide columns. We do that below with values_from = score. We also need to state which column contains the new column headers. In this example they are in the column time, so we do that with names_from = time.

head(df.long)
##       name score  time
## 1    James    15 time1
## 2    Tyler    17 time1
## 3  Stephen    14 time1
## 4 Jennifer    13 time1
## 5   Carmen    11 time1
## 6    James    16 time2
df.long %>% 
  pivot_wider(values_from = score, names_from = time)
## # A tibble: 5 x 3
##   name     time1 time2
##   <fct>    <dbl> <dbl>
## 1 James       15    16
## 2 Tyler       17    19
## 3 Stephen     14    20
## 4 Jennifer    13    21
## 5 Carmen      11    23

Now our data are in a wide format.

Again, it is more common that we have more columns in our dataset. Say we had rated the confidence of each participant on the tasks prior to testing them. We may have a column called confidence that might look like this:

# if we had more columns in our dataset:
df.long$confidence <- c(3, 8, 9, 4, 10) # this repeats all the way down the dataset

df.long
##        name score  time confidence
## 1     James    15 time1          3
## 2     Tyler    17 time1          8
## 3   Stephen    14 time1          9
## 4  Jennifer    13 time1          4
## 5    Carmen    11 time1         10
## 6     James    16 time2          3
## 7     Tyler    19 time2          8
## 8   Stephen    20 time2          9
## 9  Jennifer    21 time2          4
## 10   Carmen    23 time2         10

Even when we have more columns, we still can make our data go wide in the same way:

df.long %>% 
  pivot_wider(values_from = score, names_from = time)
## # A tibble: 5 x 4
##   name     confidence time1 time2
##   <fct>         <dbl> <dbl> <dbl>
## 1 James             3    15    16
## 2 Tyler             8    17    19
## 3 Stephen           9    14    20
## 4 Jennifer          4    13    21
## 5 Carmen           10    11    23

Here, it just truncates the confidence column into one column to go along with name.

4.6.3 Real Data Example.

It can be quite hard to work out which columns to select when converting data from long to wide. It mainly comes from practice and from trial-and-error. You can always check if you got it right, and correct yourself if you did not. In this section, we’ll just show an example of converting dataframes between these two formats using some real data.

The dataset we’ll use is called wheels1.csv. We’ll read it in and assign the name wheel:

wheel <- read_csv("data/wheels1.csv")

head(wheel)
## # A tibble: 6 x 14
##   strain id      day1  day2   day3   day4 dob   mother sex   wheel startdate
##   <chr>  <chr>  <dbl> <dbl>  <dbl>  <dbl> <chr> <chr>  <chr> <dbl> <chr>    
## 1 B6     692ao 12853  8156.  9028. 12516. 4/18… 633ax  male      9 10/9/2005
## 2 B6     656aa  2644  4012.  5237   7404. 12/1… 593ar  male      1 9/12/2005
## 3 B6     675ag  4004. 3054.  3816.  3761  2/9/… 593ad  male      5 10/13/20…
## 4 B6     675ai 11754. 8863  11784  11684  2/9/… 593ad  male      4 10/13/20…
## 5 B6     656af  6906. 5322. 10424.  8468. 12/1… 593ar  male      2 10/13/20…
## 6 B6     656al  6517  4440   5334.  9291  12/1… 554aa  male      8 10/9/2005
## # … with 3 more variables: total <dbl>, age <dbl>, group <chr>

We’re mainly interested in the id, strain and columns with day in, so we shall use select() to select these columns:

wheeldf <- wheel %>% select(id, strain, day1,day2,day3,day4)

head(wheeldf)
## # A tibble: 6 x 6
##   id    strain   day1  day2   day3   day4
##   <chr> <chr>   <dbl> <dbl>  <dbl>  <dbl>
## 1 692ao B6     12853  8156.  9028. 12516.
## 2 656aa B6      2644  4012.  5237   7404.
## 3 675ag B6      4004. 3054.  3816.  3761 
## 4 675ai B6     11754. 8863  11784  11684 
## 5 656af B6      6906. 5322. 10424.  8468.
## 6 656al B6      6517  4440   5334.  9291

These data represent the number id of individual mice, the strain of that mouse, and the number of wheel revolutions made on day1, day2, day3 and day4. It’s currently in wide format.

Let’s convert it to long format, by taking the data in each day column and forcing them to go long:

#convert to long
wheeldf.long <- wheeldf %>% 
  pivot_longer(cols = 3:6, names_to="day")  

wheeldf.long
## # A tibble: 320 x 4
##    id    strain day    value
##    <chr> <chr>  <chr>  <dbl>
##  1 692ao B6     day1  12853 
##  2 692ao B6     day2   8156.
##  3 692ao B6     day3   9028.
##  4 692ao B6     day4  12516.
##  5 656aa B6     day1   2644 
##  6 656aa B6     day2   4012.
##  7 656aa B6     day3   5237 
##  8 656aa B6     day4   7404.
##  9 675ag B6     day1   4004.
## 10 675ag B6     day2   3054.
## # … with 310 more rows

If we wished to make our new wheeldf.long dataset to go back to a wide format, we could do this:

# to make it go wider again, we do:
wheeldf.long %>% 
              pivot_wider(
                 names_from = day,
                 values_from = value
                    )
## # A tibble: 80 x 6
##    id    strain   day1   day2   day3   day4
##    <chr> <chr>   <dbl>  <dbl>  <dbl>  <dbl>
##  1 692ao B6     12853   8156.  9028. 12516.
##  2 656aa B6      2644   4012.  5237   7404.
##  3 675ag B6      4004.  3054.  3816.  3761 
##  4 675ai B6     11754.  8863  11784  11684 
##  5 656af B6      6906.  5322. 10424.  8468.
##  6 656al B6      6517   4440   5334.  9291 
##  7 692an B6     11366.  6473  11790. 15724 
##  8 705ab B6     13794. 10868  14374. 12729 
##  9 692at B6     10486. 12650. 15851  15790 
## 10 692au B6      4592.  5819   9272  11379 
## # … with 70 more rows

4.7 Joins

A common requirement in data science is to join together two datasets. Usually these two datasets will have some bits of data in common. For instance, you may have one dataset that contains subject names and lots of demographic information. Another dataset might contain the subject names and their performance on various tests. Ideally, you’d have all of this information in one dataframe. The tidyverse is excellent at joining dataframes quickly and automatically. It is far more foolproof than cutting, copying and pasting in excel which should always be a no-no.

In this course, we won’t actually ever ask you to do any joining. All the datasets that we provide are already cleaned up. However, we are including this small subsection in this guide to introduce you to the topic, should it be something you ever need to do on your own. Joining is surprisingly a really big topic by itself, so we just scratch the surface here. For much more information on this topic, we highly recommend Prof Jenny Bryan’s guides as a good starting point.

To illustrate the power of joining, lets make up some data.

The first dataframe, x contains ten subjects who are referred to by an id and their ages (age).

x <- data.frame("id" = 1:10, "age" = c(21,25,17,34,25,33,22,27,29,24))

head(x)
##   id age
## 1  1  21
## 2  2  25
## 3  3  17
## 4  4  34
## 5  5  25
## 6  6  33

The second dataframe, y, contains the id of each individual again, but this time we also have data on their height, as well as how many hours they spent on an activity and some work.

y <- data.frame(
  "id" = 1:10, 
  "height_cm" = c(156, 155, 154, 149, 153, 152, 151, 150, 147, 155), 
  "activity_hr"= c(3,5,3,6,7,4,2,8,4,5), 
  "work_hr" =c(40,35,38,46,50,42,40,46,41,40)
  )

head(y)
##   id height_cm activity_hr work_hr
## 1  1       156           3      40
## 2  2       155           5      35
## 3  3       154           3      38
## 4  4       149           6      46
## 5  5       153           7      50
## 6  6       152           4      42

Currently we have ‘age’ in a separate dataframe. Perhaps we want to join it together with all the other information in our dataframe y. We can use full_join() to do this. full_join() joins two different dataframes based on one or more shared variable(s). The following will join based on id.

x %>% 
  full_join(y)
## Joining, by = "id"
##    id age height_cm activity_hr work_hr
## 1   1  21       156           3      40
## 2   2  25       155           5      35
## 3   3  17       154           3      38
## 4   4  34       149           6      46
## 5   5  25       153           7      50
## 6   6  33       152           4      42
## 7   7  22       151           2      40
## 8   8  27       150           8      46
## 9   9  29       147           4      41
## 10 10  24       155           5      40

As you can see, R joined the two dataframes together and age is now included in the new dataframe.

For this to work, you must have at least one column in each of your two dataframes that exactly matches by name. If you do not have this, R will not know how to join the dataframes together. If you have multiple matching columns, then R will match based on all columns that match - so be careful!