6 Data types
Data can be stored in lots of different forms. For example, "TRUE" and TRUE are stored as two different types, even though they look very similar to us.
The main different data types are:
- logical
- TRUE
- FALSE
- double (numeric)
- 12.5
- 19
- 99999
- integer (numeric)
- 2L
- 34L
- character
- “hello”
- “my name is”
- factors
- dates
- 2019-06-01
- datetime (POSIXct/POSIXlt)
- 2019-06-01 12:00:00
Let’s have a look at each one in detail:
6.1 Logical
A logical variable can only have two real values, TRUE or FALSE. I say two real values, because you can also have things like NA, but that’s true of any data type. We look at NA values a little bit later on.
Logical variables are used a lot in response questionnaires, where the answer to the question is either “Yes” or “No” (TRUE or FALSE). I would recommend converting any character strings like “Yes” or “No” or “TRUE” or “FALSE” to a logical variable rather than leaving them as characters, because it’ll make your analysis less verbose (use fewer lines of code), even if it doesn’t change the underlying logic.
To test whether something is stored as logical, we use the is.logical() function:
is.logical(TRUE)## [1] TRUEis.logical("TRUE")## [1] FALSETo convert a value to logical, use the as.logical() function:
as.logical(1)## [1] TRUEas.logical(0)## [1] FALSEas.logical("TRUE")## [1] TRUEas.logical("FALSE")## [1] FALSEBe careful though, just because a conversion seems obvious to you, doesn’t mean you’ll get the expected result! For instance, what do you think as.logical(2) should return? See for yourself.
6.2 Double
The best way to think of a double value is as a number. It can be a whole number (but see Integers) or a decimal. R will often take care of any implicit number conversion that needs to be done under the hood, so the only thing you really need to keep in mind is that when you assign a number, be it a whole number or decimal, it will be stored as double by default.
As an aside, it’s called double because it’s stored using double precision.
To check whether a value is stored as double (or more generally numeric), use the is.double() and is.numeric() functions:
is.double(2)## [1] TRUEis.numeric("not numeric")## [1] FALSEis.double(2L) # see the next section for why this returns FALSE## [1] FALSETo convert a value to a double, use the as.double() or as.numeric() functions:
as.double("5")## [1] 5as.numeric("10")## [1] 10as.double("im going to cause a warning")## Warning: NAs introduced by coercion## [1] NA6.3 Integer
Whilst also storing numeric data (like double), integers are specific to whole numbers. Also, by default, even when you assign a whole number, like this: number <- 1, R will store that value as double rather than as an integer. For the most part, I let R take care of how it stores numbers, unless I explicitly need it to be of a certain type. That is pretty rare though.
To check if something is an integer, use the is.integer() function:
is.integer(2)## [1] FALSEis.integer(2L)## [1] TRUETo convert to an integer, use the L suffix or the as.integer() function:
1L## [1] 1as.integer(1)## [1] 1One small different between the L suffix and the as.integer() function though, is that the former will not convert values that would lose information as an integer to integer, it will keep it as a numeric. as.integer() will always return an integer:
1.5L # This will still return 1.5## [1] 1.5is(1.5L)## [1] "numeric" "vector"as.integer(1.5) # This will return 1 as an integer## [1] 1is(as.integer(1.5))## [1] "integer" "double" "numeric"
## [4] "vector" "data.frameRowLabels"Some of the keen-eyed among you may have noticed something. The as.integer() function doesn’t round the value, it gives you the “floor” value:
as.integer(1.9999999999999)## [1] 1This is something to bear in mind, converting a non-integer into an integer will not give you the closest integer to the value you provided. Instead, it will essentially take the value you’ve provided and always round down the nearest whole number.
6.4 Character
Sometimes called characters, or character strings, or just strings, characters store text. If you assign a value within quotation marks, regardless of what’s inside the quotation marks, it will be stored as character. For example,
"5" stores a character string with the text “5”, not the number 5. This is particularly important when you want to start combining variables. For example, {r, error = TRUE} "5" + 5 doesn’t work, because you’re trying to add text to a number, which doesn’t make sense.
To check whether something is stored as a character, use the is.character() function:
is.character("hello")## [1] TRUEis.character(5)## [1] FALSEis.character(TRUE)## [1] FALSETo convert something to a character, use the as.character() function:
as.character(5)## [1] "5"as.character(TRUE)## [1] "TRUE"6.5 Factors
Factors are a unique but useful data type in R. Essentially, factors store different levels that represent some sort of grouping. For example, say you were collecting some information on people from different countries, the column that holds which country the respondent is from could be stored as a factor, with the levels England, Spain, France, etc.
A factor level is made up of two things. A label and a number that represents that group. In my countries example, our factor would have the labels “England”, “Spain”, “France” and the values 1, 2, 3. This means that internally, a factor is essentially a collection of integers representing the level position and character strings representing the level label.
To create a factor, we just use the factor() function:
factor(c("England", "France", "Spain"))## [1] England France Spain
## Levels: England France SpainIt’s also worth remembering that you can have levels that don’t appear in the data you have. For example, in a questionnaire, you may provide the options “None”, “Some”, “All”. But in your responses, you may see that no-one chose the “None” option. In that case, you would still create a factor with three levels, even though only two of them appear.
You can also specify whether a factor is ordered. You would use an ordered factor when the levels have meaningful order. For instance, in the above example, it would make sense that “Some” is better than “None”, and “All” is better than “Some”. To create an ordered factor, just specify ordered = TRUE in your function. By default, the factor will be ordered in the order the values appear, unless you specify levels (see below).
To convert something to a factor, use the factor() function if you want to specify levels and labels, or as.factor() to do it for you:
factor(c("Some", "All"), levels = c("None", "Some", "All"))## [1] Some All
## Levels: None Some Allfactor(c("Some", "All"), levels = c("None", "Some", "All"), ordered = TRUE)## [1] Some All
## Levels: None < Some < Allas.factor(c("Some", "All"))## [1] Some All
## Levels: All SomeNotice the difference in the output of those three lines. The first allows us to specify the levels (i.e. the values that were possible). The second does the same but we also specify the ordering of the levels, and the third just converts the provided values and generates the levels based on that data.
Note: An important change in R version 4.0.0 is that R will no longer automatically convert strings (characters) to factors when you import data using data.frame() or read.table(). Prior to 4.0.0, it would automatically convert strings to characters unless otherwise specified.
6.6 Dates
Dates in any language are tricky. Different countries store dates in different formats and different bits of software store dates in different ways (looking at you Excel). This can make storing values as dates tough.
The most common way of creating a date is to use the as.Date() function. To use this function, you just need to provide your date as a character string:
as.Date("2019/01/01")## [1] "2019-01-01"But Adam, how does R know which one is the month and which is the day? Good question, thank you for asking. By default, R expects your character string to be in the order “Year/Month/Day”. If you don’t provide it in that format, you’ll get a nonsense output:
as.Date("01/12/2019")## [1] "1-12-20"If your data is in a different format however, you can specify the format:
as.Date("01/12/2019", format = "%d/%m/%Y")## [1] "2019-12-01"Here, we’re telling R that the string is in the format “Day/Month/Year”. A list of the different codes that can be used in the format parameter can be found here, or by typing “R date codes” into Google.
Because nothing in life is simple, sometimes you’ll get some data that has the date stored as a number. This is because the source of that data has the date stored as the number of days that have passed since an origin date. Because it’s a number, our as.Date(..., format = ...) doesn’t work. Instead, we can still use the as.Date() function, but we need to specify what the origin date is that the number refers to.
By default, when importing from Excel in Windows, the origin date is December 31st 1899. More commonly, the date January 1st 1970 (also known as the epoch date) is used.
Anyway, to specify your origin, we use the origin parameter, like this:
as.Date(18262, origin = "1970/01/01")## [1] "2020-01-01"Notice the format I’ve provided the origin in. It’s the same as the default that R expects, and I would recommend copying that format wherever possible. If you’re someone who just wants to watch the world burn, then you can specify a format for your origin as well…’
as.Date(18262, origin = as.Date("01/01/1970", format = "%d/%m/%Y"))## [1] "2020-01-01"but where’s the humanity in that?
Testing whether something is a date is not as simple as the other data types unfortunately. Instead, can use the inherits function and ask if it inherits the class date:
inherits(as.Date("2020/01/01"), 'Date')## [1] TRUEOr can use the is() or class() functions. If the first value returned is “Date”, then you know it’s a date:
is(as.Date("2020/01/01"))## [1] "Date" "oldClass"class(as.Date("2020/01/01"))## [1] "Date"6.7 Datetimes (POSIXct/lt)
If you thought dates were annoying, datetimes are like dates’ little brother who didn’t get enough attention as a child and so acts up all the time. One of the reasons for this is that datetimes aren’t actually called datetimes. They’re called POSIXct or POSIXlt (depending on how the date is stored internally) in R. So whenever you see that dreadful word, just remember “ah, Adam told me that means datetime” and you’ll be fine.
Another thing that makes datetimes tough is that in addition to dates, datetimes (as you may have guessed) also store the time. The issue with that is that time is a more relative concept - there are lots of different time zones, so how do you know which one you’re referring to? By default, R has a locale for where you currently are and will use that location for your timezone. You override that default using the Sys.setlocale() function, or you can use the tz parameter when creating your datetime as we’ll see below.
6.7.1 POSIXct & POSIXlt
These two data types both store the same thing; dates and times. But there’s a couple of differences in how they do it.
Starting with POSIXct, the “ct” stands for “calendar time”. The POSIXct datatype stores the number of seconds since the origin (just like our dates are stored as the number of days since an origin).
For POSIXlt, the “lt” stands for “local time”. Instead of storing the datetime as the number of seconds since a point in time, POSIXlt stores datetimes as a list of time attributes, like “year”, “month”, “day”, “minute”, “second”, etc.
Ultimately, they both store the same information. I personally use POSIXct when converting to/from datetimes, but it shouldn’t make a great deal of difference which one you choose as the differences are just how R stores the value internally.
Note: While the “ls” in POSIXlt categorically does not stand for “list time”, thinking about the POSIXlt datatype as being like a “list time” can be useful in remembering how the POSIXct and POSIXlt objects differ in how they represent datetimes; POSIXlt stores a ‘list’ of attributes, POSIXct doesn’t.
With these annoyances aside however, creating datetimes isn’t all that different to creating dates except that we use the as.POSIXct() or as.POSIXct() functions instead. We just provide a character string (with a format specification if necessary), or a number with an origin. Remember though that now our difference from the origin should be in seconds, not days.
as.POSIXct("2020/01/01 12:00:00")## [1] "2020-01-01 12:00:00 UTC"as.POSIXct("2020/01/01 12:00:00", tz = "NZ")## [1] "2020-01-01 12:00:00 NZDT"as.POSIXct(1577880000, origin = "1970/01/01")## [1] "2020-01-01 12:00:00 UTC"Similar to dates, there is no is.POSIXct() function in base R, so we use the inherits() or is() and class() functions instead:
inherits(as.POSIXct("2020/01/01 12:00:00"), "POSIXct")## [1] TRUEis(as.POSIXct("2020/01/01 12:00:00"))## [1] "POSIXct" "POSIXt" "oldClass"class(as.POSIXct("2020/01/01 12:00:00"))## [1] "POSIXct" "POSIXt"6.8 NA and NULL
The R language has two kind of related values, NA and NULL.
6.8.1 NULL
NULL indicates the absence of a value. It means that a value is missing (or has length zero). A null value has no ‘type’ because it represents an absence of something, so passing a null value to any of the is.[type]() functions will return FALSE. Instead, checking whether a value is NULL is done with the is.null() function:
is.character(NULL)## [1] FALSEis.null(NULL)## [1] TRUEis.character() returns FALSE because NULL doesn’t have a data type (it’s the absence of a value), whereas is.null() returns TRUE because it is NULL.
What happens when you compare values to NULL using our logical operators? Well, this is where the idea of NULL having a length of 0 comes in. If you type length(NULL), you’ll see it will return 0. So when you’re comparing a value against NULL like this:
"Am I NULL?" == NULL## logical(0)You get an output that is of length 0. And that makes sense, because NULL is missing/has 0 zero length. So basically you’re then comparing something ("Am I NULL?") to nothing (NULL). Therefore, your output is nothing (or length 0).
Dealing with NULL can be a bit tricky, so the important thing to remember is that NULL represents something missing, and have length 0. Unlike, NA…
6.8.2 NA
On the other hand, NA (not available) represents an invalid value. This most often occurs when you try to convert one datatype to another where R can’t assign an appropriate value.
NA isn’t of length 0 like NULL because it is a value, it’s just invalid. In other words, there’s something there, but it doesn’t really make sense what it is.
For example, attempting to parse a character string to a date format that doesn’t match will result in a NA value:
as.Date("10/01/20", format = "%m%Y%d")## [1] NAR has tried to parse a value from one type into another. The value isn’t NULL because it clearly isn’t missing, but it couldn’t be converted to a date type, so it’s NA.
6.8.2.1 NA Types
Unlike NULL, there are different NA values for each datatype (although they’ll all look like NA in the console). In the example above, we actually created a NA that has the Date type:
errant_date <- as.Date("10/01/20", format = "%m%Y%d")
is(errant_date)## [1] "Date" "oldClass"This is because R knows what type the NA should be, but it couldn’t assign in a proper value.
The same behaviour can be observed for other data types:
as.logical("not a logical")## [1] NAis(as.logical("not a logical"))## [1] "logical" "vector"To test whether something is NA, we use the is.na() function. You don’t need to worry about what type the NA is, this will test if it is an NA of any type.
is.na(NA_character_) # this will be an NA that is of type 'character'## [1] TRUEis.na(NA_integer_) # this will be an NA that is of type 'integer'## [1] TRUE6.8.2.2 Dealing with NAs
Dealing with NAs is often contextual. Attempting to perform a mathematical calculation on a vector of values that contains at least one NA will often return NA:
sum(1,2,NA)## [1] NAmean(c(1,NA))## [1] NANA + 1## [1] NAIn some cases, those NAs will represent real issues with the values and so removing the NAs or converting them to 0 will just mask the error without fixing it.
Alternatively, data imports can often return NA values because of differing data types or similar and so converting those values to 0 or removing them outright may be appropriate.
Ultimately, how you deal with NA values is a question that you’ll need to answer when it happens and depending on the situation. I will give you a helping hand though and say that if you want to just remove the NA values when summing or calculating an average or similar, then these functions often have an na.rm parameter than can be used to remove the NA values from the supplied list of values:
sum(1,2,NA, na.rm = TRUE)## [1] 36.9 NaN and Infinity
A special case is NaN (not a number). NaNs are distinct to NA in that they represent a valid value. More precisely, NaN represents not real numbers (numeric values that cannot be represented with numbers). For example, dividing 0/0.
Inf and -Inf are similar constructs. They represent infinity and minus infinity respectively. They are valid values but they are not representable with numbers, so they have their own reserved words.
NaN, Inf and -Inf are all of the numeric type and do not have equivalent values in other data types.
6.10 Implicit Conversion
Converting between types using the as.x() functions is nice and easy because the type that you’re converting to is explicitly defined. However, there is also implicit conversion in R. That is where R basically avoids an error by converting a variable into a different data type so that the function can run correctly. Let’s look at a proper example.
paste("I'm a character", 100)## [1] "I'm a character 100"The paste() function provides a pretty good example of implicit conversion in the wild. paste() expects character strings when it’s called, so what happens when you provide it with a number like we did above? Why doesn’t it complain?
Well it’s because R automatically converts it for us. It takes that number, converts it to a character and then pastes things together. This way, you can be a bit more fast and loose with your data types because R will do the type conversion for you. Maybe the best way to describe it is that it’s a bit like a programming autocorrect.
Remember this concept because we’ll revisit it later when we look at coercoin.
6.10.1 Dangers of implicit conversion
As you can imagine, this automatic conversion between types when the user didn’t specifically ask for it can cause problems. For example, when I teach R courses I often ask the attendees what they think R is going to return when I type "TRUE" == TRUE. Without fail, the majority will say that it’s going to be FALSE; one is a character string and the other is a logical value.
My heart always dies a little bit then when I show them that it actually returns TRUE. “But why?”, they rightly ask. And it’s all because of implicit conversion. When we use the == operator, R will try and coerce the variables so that they’re the same type. In this case, it successfully coerces "TRUE" to TRUE, and so returns TRUE.
There are, of course, going to be cases where this makes things easier. For instance, say you’ve got some data where there are two columns of logical values but one has incorrectly been imported as a character column. This implicit conversion will save you from having to change the character column. But implicit conversion will inevitably cause you problems in the future.
So my advice is simply to be aware of it. Don’t fight it, don’t worry about it unduly, just keep it in the back of your mind for that one time where you can’t seem to figure out why you can divide things by TRUE.
6.11 Questions
- Why are
2and2Ldifferent? - What is an ordered factor and how is it different to a character string?
- Why does
as.Date("19/01/2019", format = "%d/%m/%y")return the date 19th Jan 2020 and not 19th Jan 2019? - Why does
as.numeric(TRUE)return 1? What willas.logical(2)return? - Look at the
identical()function. How is that different from==?