First or last observations

• To keep only the first 10 observations:

head(dat, n = 10)
##    speed dist
## 1      4    2
## 2      4   10
## 3      7    4
## 4      7   22
## 5      8   16
## 6      9   10
## 7     10   18
## 8     10   26
## 9     10   34
## 10    11   17

• To keep only the last 5 observations:

tail(dat, n = 5)
##    speed dist
## 46    24   70
## 47    24   92
## 48    24   93
## 49    24  120
## 50    25   85

Random sample of observations

• To draw a sample of 4 observations without replacement:

library(dplyr)
sample_n(dat, 4, replace = FALSE)
##   speed dist
## 1    13   34
## 2    15   20
## 3    18   56
## 4    20   32

Based on row or column numbers

If you know what observation(s) or column(s) you want to keep, you can use the row or column number(s) to subset your dataset. We illustrate this with several examples:

• keep all the variables for the \(3^{rd}\) observation:

dat[3, ]

• keep the \(2^{nd}\) variable for all observations:

dat[, 2]

• You can mix the two above methods to keep only the \(2^{nd}\) variable of the \(3^{rd}\) observation:

dat[3, 2]
## [1] 4

• keep several observations; for example observations \(1\) to \(5\), the \(10^{th}\) and the \(15^{th}\) observation for all variables:

dat[c(1:5, 10, 15), ] # do not forget c()
##    speed dist
## 1      4    2
## 2      4   10
## 3      7    4
## 4      7   22
## 5      8   16
## 10    11   17
## 15    12   28

• remove observations 5 to 45:

dat[-c(5:45), ]
##    speed dist
## 1      4    2
## 2      4   10
## 3      7    4
## 4      7   22
## 46    24   70
## 47    24   92
## 48    24   93
## 49    24  120
## 50    25   85

• tip: to keep only the last observation, use nrow() instead of the row number:

dat[nrow(dat), ] # nrow() gives the number of rows
##    speed dist
## 50    25   85

This way, no matter the number of observations, you will always select the last one. This technique of using a piece of code instead of a specific value is to avoid “hard coding”. Hard coding is generally not recommended (unless you want to specify a parameter that you are sure will never change) because if your dataset changes, you will need to manually edit your code.

As you probably figured out by now, you can select observations and/or variables of a dataset by running dataset_name[row_number, column_number]. When the row or column number is left empty, the entire row/column is selected. Note that all examples presented above also works for matrices:

mat <- matrix(c(-1, 2, 0, 3), ncol = 2, nrow = 2)
mat
##      [,1] [,2]
## [1,]   -1    0
## [2,]    2    3
mat[1, 2]
## [1] 0

Based on variable names

To select one variable of the dataset based on its name rather than on its column number, use dataset_name$variable_name:

dat$speed
##  [1]  4  4  7  7  8  9 10 10 10 11 11 12 12 12 12 13 13 13 13 14 14 14 14 15 15
## [26] 15 16 16 17 17 17 18 18 18 18 19 19 19 20 20 20 20 20 22 23 24 24 24 24 25

Accessing variables inside a dataset with this second method is strongly recommended compared to the first if you intend to modify the structure of your database. Indeed, if a column is added or removed in the dataset, the numbering will change. Therefore, variables are generally referred to by its name rather than by its position (column number). In addition, it is easier to understand and interpret code with the name of the variable written (another reason to call variables with a concise but clear name). There is only one reason why I would still use the column number; if the variables names are expected to change while the structure of the dataset do not change.

To select variables, it is also possible to use the select() command from the powerful dplyr package (for compactness only the first 6 observations are displayed thanks to the head() command):

head(select(dat, speed))
##   speed
## 1     4
## 2     4
## 3     7
## 4     7
## 5     8
## 6     9

This is equivalent than removing the distance variable:

head(select(dat, -dist))
##   speed
## 1     4
## 2     4
## 3     7
## 4     7
## 5     8
## 6     9

Based on one or multiple criterion

Instead of subsetting a dataset based on row/column numbers or variable names, you can also subset it based on one or multiple criterion:

• keep only observations with speed larger than 20. The first argument refers to the name of the dataset, while the second argument refers to the subset criteria:

subset(dat, dat$speed > 20)
##    speed dist
## 44    22   66
## 45    23   54
## 46    24   70
## 47    24   92
## 48    24   93
## 49    24  120
## 50    25   85

• keep only observations with distance smaller than or equal to 50 and speed equal to 10. Note the == (and not =) for the equal criteria:

subset(dat, dat$dist <= 50 & dat$speed == 10)
##   speed dist
## 7    10   18
## 8    10   26
## 9    10   34

• use | to keep only observations with distance smaller than 20 or speed equal to 10:

subset(dat, dat$dist < 20 | dat$speed == 10)
##    speed dist
## 1      4    2
## 2      4   10
## 3      7    4
## 5      8   16
## 6      9   10
## 7     10   18
## 8     10   26
## 9     10   34
## 10    11   17
## 12    12   14

• to filter out some observations, use !=. For instance, to keep observations with speed not equal to 24 and distance not equal to 120 (for compactness only the last 6 observations are displayed thanks to the tail() command):

tail(subset(dat, dat$speed != 24 & dat$dist != 120))
##    speed dist
## 41    20   52
## 42    20   56
## 43    20   64
## 44    22   66
## 45    23   54
## 50    25   85

Discretize a continuous variable

To transform a continuous variable into a discrete variable:

dat$speed_discr <- cut(dat$speed,
  breaks = c(0, 12, 15, 19, 26), # cut points
  right = FALSE # closed on the left, open on the right
)
dat[c(1:2, 23:24, 49:50), ] # display some observations
##    speed dist speed_dist  speed_cat speed_discr
## 1      4    2          8  low speed      [0,12)
## 2      4   10         40  low speed      [0,12)
## 23    14   80       1120 high speed     [12,15)
## 24    15   20        300 high speed     [15,19)
## 49    24  120       2880 high speed     [19,26)
## 50    25   85       2125 high speed     [19,26)

Recode categorical variables

It is possible to recode labels of a categorical variable if you are not satisfied with the current labels. In this example, we change the labels as follows:

• “small distance” \(\rightarrow\) “short distance”
• “big distance” \(\rightarrow\) “large distance”

dat$dist_cat <- recode(dat$dist_cat,
  "small distance" = "short distance",
  "big distance" = "large distance"
)
head(dat)
##   speed dist speed_dist  speed_cat speed_discr       dist_cat
## 1     4    2          8  low speed      [0,12) short distance
## 2     4   10         40  low speed      [0,12) short distance
## 3     7    4         28  low speed      [0,12) short distance
## 4     7   22        154  low speed      [0,12) large distance
## 5     8   16        128 high speed      [0,12) large distance
## 6     9   10         90 high speed      [0,12) short distance

Change reference level

For some analyses, you might want to change the order of the levels. For example, if you analyze data about a control group and a treatment group, you may want to set the control group as the reference group. By default, levels are ordered by alphabetical order or by its numeric value if it was change from numeric to factor.

• to check the current order of the levels (the first level being the reference):

levels(dat$dist_cat)
## [1] "short distance" "large distance"

In this case, “short distance” being the first level it is the reference level. It is the first level because it was initially set with a value equal to 1 when creating the variable.

• to change the reference level:

dat$dist_cat <- relevel(dat$dist_cat, ref = "large distance")
levels(dat$dist_cat)
## [1] "large distance" "short distance"

Large distance is now the first and thus the reference level.

Remove NAs

A simple solution is to remove all observations (i.e., rows) containing at least one missing value. This is done by keeping observations with complete cases:

dat_complete <- dat[complete.cases(dat), ]
dat_complete
##   variable1 variable2
## 1         6         3
## 2        12         7
## 4         3         1

Be careful before removing observations with missing values, especially if missing values are not “missing at random”. This is, however, beyond the scope of the present article.

Impute NAs

Instead of removing observations with at least one NA, it is possible to impute them, that is, replace them by some values such as the median or the mode of the variable. This can be done easily with the command impute() from the package imputeMissings:

library(imputeMissings)
dat_imputed <- impute(dat) # default method is median/mode
dat_imputed
##   variable1 variable2
## 1         6         3
## 2        12         7
## 3         6         9
## 4         3         1

When the median/mode method is used (the default), character vectors and factors are imputed with the mode. Numeric and integer vectors are imputed with the median. Again, use imputations carefully. Other packages offer more advanced imputation techniques. However, we keep it simple and straightforward for this article as advanced imputations is beyond the scope of introductory data manipulations in R.

Thanks for reading. I hope this article helped you to manipulate your data in RStudio. See other articles related to R here. As always, if you find a mistake/bug or if you have any questions do not hesitate to let me know in the comment section below, raise an issue on GitHub or contact me. Get updates every time a new article is published by subscribing to this blog.