Variable Class

Session 3 practical exercises

Column names are now sorted. But a clean name does not mean the column contains the right type of data — and in R, type matters more than you might expect. A number stored as text cannot be used in calculations. A date stored as text cannot be used to compute time intervals or epidemiological weeks. R will not warn you about this automatically. It will just quietly produce wrong results, or throw an error at the worst possible moment.

So the second step of any cleaning pipeline is to check what R thinks each variable is, and fix anything that is wrong.

Part 2 · Inspecting variable types

You have already met names(). Now you need a function that shows you the type of each column at a glance. Run this on your cleaning object:

glimpse(imd_cleaning)

You should see something like this:

Rows: 2,724
Columns: 16
$ disease               <chr> "Mening", "Mening", "Mening", "Mening", "Mening"…
$ region_id             <chr> "R1", "R1", "R1", "R1", "R1", "R1", "R1", "R1", …
$ age_years             <dbl> 1, 3, 0, 0, 2, 6, 55, 17, 3, 38, 3, 14, 14, 16, …
$ age_months            <chr> NA, NA, "2", "5", NA, NA, NA, NA, NA, NA, NA, NA…
$ sex                   <chr> "M", "F", "F", "M", "M", "M", "M", "F", "M", "F"…
$ year                  <dbl> 1999, 1999, 1999, 1999, 1999, 1999, 1999, 1999, …
$ key_date              <chr> "1999-01-01", "1999-01-01", "1999-01-01", "1999-…
$ symptom_onset_date    <chr> "1999-01-01", "1999-01-01", "1999-01-01", "1999-…
$ clinical_presentation <chr> "Sepsis", "Mening", "Mening", "Sepsis", "Both", …
$ death                 <chr> "No", "No", "No", "Yes", "No", "Yes", "Yes", "No…
$ country               <chr> "ES", "ES", "ES", "ES", "ES", "ES", "ES", "ES", …
$ imported              <chr> "No", "No", "No", "No", "No", "No", "No", "No", …
$ diagnostic_date       <lgl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ serogroup             <chr> "B_NeisMen", "C_NeisMen", "C_NeisMen", "B_NeisMe…
$ type_of_case          <chr> "Prim", "Prim", "Prim", "Prim", "Prim", "Prim", …
$ lab_confirmed         <chr> "Yes", "Yes", "Yes", "Yes", "Yes", "No", "Yes", …

Look at the type tags in angle brackets: <chr>, <dbl>, <lgl>. These are R’s way of telling you what it thinks each column is.

Action — Before reading on, look at the output and identify which columns have a type mismatch — where what R thinks the variable is does not match what it should actually be.

Action — Try out the other functions for examining data: str(), summary(), head()

The main variable types you will encounter

Tag	Type	Example
`<chr>`	Character (text)	`"Male"`, `"Narnia"`, `"Yes"`
`<dbl>`	Double (number)	`23`, `1999`, `3.5`
`<lgl>`	Logical	`TRUE`, `FALSE`, `NA`
`<date>`	Date	`1999-03-12`

A column with dates showing as <chr> is one of the most common problems in real surveillance data. R sees "1999-03-12" as just a string of characters — it has no idea it represents a point in time.

Did you realize the ghost column? Hint: it also bears one strange new variable type you hadn’t seen before. Take a moment to look for it, and try to guess what may be happening

Part 3 · Fixing variables types

Fixing numbers

The age_months column is stored as <chr>. This means R cannot use it in any calculation. The fix is straightforward: inside a mutate(), you overwrite the column with its numeric version using as.numeric()

Remember that mutate() can modify existing variables

mutate(
  variable = function(variable)
)

Action — Add this mutate() to your cleaning pipeline to fix age_months, run it, and check glimpse() again. Did age_months change to <dbl>?

You will likely see this warning in your console:

Warning: NAs introduced by coercion

What does “NAs introduced by coercion” mean?

Do not panic. This is R being honest with you. When R tries to convert a character column to numeric, it succeeds for values like "2" or "8". But some cells may contain text that cannot be converted — like a genuine missing value entered as blank or a data entry error. R converts those to NA and tells you about it.

In this dataset this is expected and correct behavior. Warnings are only that: warnings. They want you to be aware something happened that is not an error inherently

Fixing dates

Dates are the most common source of pain in surveillance data — and the most important to get right, since almost every epidemiological analysis depends on them.

Look at your glimpse() output again. Three date columns are stored as <chr>:

key_date
symptom_onset_date
diagnostic_date

The lubridate package (part of tidyverse) provides a family of functions to convert text into real dates. The function name tells R what order to expect the date components:

ymd() → year, month, day (e.g. "1999-03-12")
dmy() → day, month, year (e.g. "12/03/1999")
mdy() → month, day, year (e.g. "03/12/1999")

Action — Check the actual values in your date columns to determine which format each one uses

Action — Add the three date conversions to your mutate(). Use the right function for each column based on what you observed.

Why does date format matter so much?

"03/12/1999" is that the 3rd of December, or the 12th of March? Without knowing the format, R cannot tell. If you apply the wrong function, R may either return NA for every row, or worse — silently parse the dates incorrectly and give you plausible-looking wrong values.

Always check the raw values before converting. Always.

Action — After adding all conversions, run glimpse() one more time. Confirm that age_months is now <dbl> and that key_date and symptom_onset_date are now <date>. What about diagnostic_date?

Exercise summary

Your columns now have the right types. Numbers are numbers, dates are dates. In E3 you will tackle the messier problem: categorical variables with inconsistent labels.

We used these functions for achieving our goal:

Function	Package	What it does
`glimpse()`	dplyr	Shows column types and first values — your primary inspection tool
`mutate()`	dplyr	Creates or modifies columns
`as.numeric()`	base R	Converts a column to numeric; non-convertible values become `NA`
`ymd()` / `dmy()`	lubridate	Converts text to dates; function name = expected order of components

💡 Show solution - only after trying yourself!

This is the full mutate() block to add to your pipeline after the name-cleaning step:

  # Change variables type
  mutate(
    age_months         = as.numeric(age_months),
    key_date           = ymd(key_date),
    symptom_onset_date = ymd(symptom_onset_date),
    diagnostic_date    = ymd(diagnostic_date)
  )

--- title: "Variable Class" subtitle: "Session 3 practical exercises" --- ```{r} #| include: false library(webexercises) pacman::p_load(tidyverse, rio, here, janitor) imd_raw <- import(here("data", "raw", "IMD_Sample_Dirty.xlsx")) imd_cleaning <- imd_raw %>% clean_names() %>% rename( age_years = age, year = key_year ) ``` Column names are now sorted. But a clean name does not mean the column contains the right *type* of data — and in R, type matters more than you might expect. A number stored as text cannot be used in calculations. A date stored as text cannot be used to compute time intervals or epidemiological weeks. R will not warn you about this automatically. It will just quietly produce wrong results, or throw an error at the worst possible moment. So the second step of any cleaning pipeline is to check what R *thinks* each variable is, and fix anything that is wrong. ## Part 2 · Inspecting variable types You have already met `names()`. Now you need a function that shows you the type of each column at a glance. Run this on your cleaning object: ``` r glimpse(imd_cleaning) ``` You should see something like this: ```{r} #| echo: false glimpse(imd_cleaning) ``` Look at the type tags in angle brackets: `<chr>`, `<dbl>`, `<lgl>`. These are R's way of telling you what it thinks each column is. **Action** — Before reading on, look at the output and identify which columns have a type mismatch — where what R thinks the variable is does not match what it should actually be. **Action** — Try out the other functions for examining data: `str()`, `summary()`, `head()`  ::: {.callout-note collapse="true"} ## The main variable types you will encounter | Tag | Type | Example | |----------|------------------|-------------------------------| | `<chr>` | Character (text) | `"Male"`, `"Narnia"`, `"Yes"` | | `<dbl>` | Double (number) | `23`, `1999`, `3.5` | | `<lgl>` | Logical | `TRUE`, `FALSE`, `NA` | | `<date>` | Date | `1999-03-12` | A column with dates showing as `<chr>` is one of the most common problems in real surveillance data. R sees `"1999-03-12"` as just a string of characters — it has no idea it represents a point in time. ::: **Did you realize the ghost column?** Hint: it also bears one strange new variable type you hadn't seen before. Take a moment to look for it, and try to guess what may be happening ## Part 3 · Fixing variables types ### Fixing numbers The `age_months` column is stored as `<chr>`. This means R cannot use it in any calculation. The fix is straightforward: inside a `mutate()`, you overwrite the column with its numeric version using `as.numeric()` Remember that `mutate()` can modify existing variables ``` r mutate( variable = function(variable) ) ``` **Action** — Add this `mutate()` to your cleaning pipeline to fix `age_months`, run it, and check `glimpse()` again. Did `age_months` change to `<dbl>`? You will likely see this warning in your console: ``` r Warning: NAs introduced by coercion ``` ::: callout-tip ## What does "NAs introduced by coercion" mean? Do not panic. This is R being honest with you. When R tries to convert a character column to numeric, it succeeds for values like `"2"` or `"8"`. But some cells may contain text that cannot be converted — like a genuine missing value entered as blank or a data entry error. R converts those to `NA` and tells you about it. In this dataset this is expected and correct behavior. Warnings are only that: warnings. They want you to be aware something happened that is not an error inherently ::: ### Fixing dates Dates are the most common source of pain in surveillance data — and the most important to get right, since almost every epidemiological analysis depends on them. Look at your `glimpse()` output again. Three date columns are stored as `<chr>`: - `key_date` - `symptom_onset_date` - `diagnostic_date` The `lubridate` package (part of tidyverse) provides a family of functions to convert text into real dates. The function name tells R what order to expect the date components: - `ymd()` → year, month, day (e.g. `"1999-03-12"`) - `dmy()` → day, month, year (e.g. `"12/03/1999"`) - `mdy()` → month, day, year (e.g. `"03/12/1999"`) **Action** — Check the actual values in your date columns to determine which format each one uses **Action** — Add the three date conversions to your `mutate()`. Use the right function for each column based on what you observed. ::: {.callout-note collapse="true"} ### Why does date format matter so much? `"03/12/1999"` is that the 3rd of December, or the 12th of March? Without knowing the format, R cannot tell. If you apply the wrong function, R may either return `NA` for every row, or worse — silently parse the dates incorrectly and give you plausible-looking wrong values. Always check the raw values before converting. Always. ::: **Action** — After adding all conversions, run `glimpse()` one more time. Confirm that `age_months` is now `<dbl>` and that `key_date` and `symptom_onset_date` are now `<date>`. What about `diagnostic_date`? ### Other functions for changing class You just used two ver relevant functions: `as.numeric()` and `ymd()` for numbers and dates. But other relevant functions exist and are needed: - `as.character()` transforms column values into text - `as.integer()` alternative version for numeric transformation - `as.factor()` a factor variable is slightly different, and will be covered further into the course - Many other `as.something()` functions exist that cover topics beyond the course (matrix, tibbles, vectors, etc.) ------------------------------------------------------------------------ ```{r} #| echo: false opts1 <- c( "It shows the first 6 rows of a dataframe", "It prints a summary of means and quartiles for each column", answer = "It shows each column with its type tag and first few values", "It checks for missing values in each column" ) ``` **What does `glimpse()` show you that `head()` does not?** `r longmcq(opts1)` ------------------------------------------------------------------------ ```{r} #| echo: false opts2 <- c( answer = "`<chr>`", "`<dbl>`", "`<date>`", "`<lgl>`" ) ``` **A date column that has not yet been converted will appear in `glimpse()` as:** `r longmcq(opts2)` ```{r} #| echo: false opts3 <- c( "It means the conversion failed and all values are now missing", "It means the column contained non-numeric values that became NA — always a problem", answer = "It means some values could not be converted to numeric and became NA — expected if the column had missing entries", "It means you used the wrong function and should try `as.character()` instead" ) ``` **You run `as.numeric()` on a column and see "NAs introduced by coercion". What does this mean?** `r longmcq(opts3)` ::: {.callout-note collapse="true"} ## 💬 Explanation "NAs introduced by coercion" is R being transparent, not alarming. When converting text to numbers, any value that cannot be parsed as a number — blanks, `"unknown"`, stray letters — becomes `NA`. This is the correct behaviour. The question is always whether the number of new NAs is what you would expect given your data. ::: ```{r} #| echo: false opts4 <- c( "`mdy()` — month, day, year", "`dmy()` — day, month, year", answer = "`ymd()` — year, month, day", "Any of them — lubridate detects the format automatically" ) ``` **Which function would you use to convert the string `"1999-03-12"` to a date?** `r longmcq(opts4)` ------------------------------------------------------------------------ ## Exercise summary **Your columns now have the right types.** Numbers are numbers, dates are dates. In E3 you will tackle the messier problem: categorical variables with inconsistent labels. We used these functions for achieving our goal: | Function | Package | What it does | |------------------------|------------------------|------------------------| | `glimpse()` | dplyr | Shows column types and first values — your primary inspection tool | | `mutate()` | dplyr | Creates or modifies columns | | `as.numeric()` | base R | Converts a column to numeric; non-convertible values become `NA` | | `ymd()` / `dmy()` | lubridate | Converts text to dates; function name = expected order of components | ::: {.callout-tip collapse="true"} ## 💡 Show solution - only after trying yourself! This is the full `mutate()` block to add to your pipeline after the name-cleaning step: ``` r # Change variables type mutate( age_months = as.numeric(age_months), key_date = ymd(key_date), symptom_onset_date = ymd(symptom_onset_date), diagnostic_date = ymd(diagnostic_date) ) ``` ::: ```{=html} <script> document.addEventListener("DOMContentLoaded", function() { var radiogroups = document.getElementsByClassName("webex-radiogroup"); for (var i = 0; i < radiogroups.length; i++) { radiogroups[i].onchange = radiogroups_func; } }); </script> ```