130-missing-values.qmd

---
title: "Missing Values"
subtitle: "How to operate on values that are not known."
author: "Otho Mantegazza _ Dataviz for Scientists _ Part 1.3"
editor_options: 
  chunk_output_type: console
---

```{r}
#| include: false

library(dplyr)
library(readr)
library(magrittr)
library(tidyr)
library(palmerpenguins)
library(tibble)
library(ggplot2)
library(readr)
library(here)
library(stringr)
library(broom)

theme_update(axis.title = element_text(size = 15),
             axis.text = element_text(size = 15),
             strip.text = element_text(size = 15))
```

## What is a missing value?

::: {.column width="70%"}

A missing value is a data point that's missing, no one knows what the value was, so you can't conduct any operation on it.

In R missing values are written **NA**.

:::

## Operation on missing values

::: {.column width="70%"}

Most operation on missing values return NAs.

Ask yourself what is one plus "a number that I don't know" (**NA**). The answer is "I don't know" (**NA**). 

```{r}
NA + 1
```

```{r}
NA / 4
```

```{r}
NA == 1
```

```{r}
NA == NA
```

:::

## Testing if a value is missing

::: {.column width="70%"}

To test if an element is a missing value we must use the function **is.na()**.

```{r}
1 %>% is.na()
```

```{r}
'hello' %>% is.na()
```

```{r}
TRUE %>% is.na()
```

```{r}
NA %>% is.na()
```

:::

## Missing values in a variable

::: {.column width="70%"}

We can assign missing values to a variable and place them in a vector.

```{r}
missing_value <- NA
```

```{r}
missing_value %>% is.na()
```

:::

## Missing values in a vector

::: {.column width="70%"}

When we use **is.na()** on a vector, it returns a vector of booleans, with **TRUE** in the position where the values are missing.

```{r}
vector_with_na <- c(1, 5, NA, 10)
```

```{r}
vector_with_na %>% is.na()
```

<br>

Remember, we can't use the **==** statement to test if a vector stores NAs.

```{r}
vector_with_na == NA
```

:::

## Missing values in a vector

::: {.column width="70%"}

We can count missing values in a vector with **is.na() %>% sum()**.

```{r}
vector_with_na %>% is.na() %>% sum()
```

:::

## Operating on missing values

::: {.column width="70%"}

Most functions such as **mean()**, **median()**, **sd()** give you the chance to remove missing values with the argument **na.rm = TRUE**.

```{r}
vector_with_na %>% mean()
```

```{r}
vector_with_na %>% mean(na.rm = T)
```

Also, **ggplot**'s function remove NAs automatically for us, if they would hinder computations.

:::

## Data with missing values

::: {.column width="70%"}

Often data have NAs in them, for example the **penguins** dataset does.

To count missing values per each column, use this lines of code:

:::

```{r}
penguins %>% 
  summarise(
    across(
      everything(),
      ~is.na(.) %>% sum()
    )
  )
```

## Data with missing values

::: {.column width="70%"}

We can also count them stratified by the fixed variables, for example:

:::

```{r}
penguins %>% 
  group_by(island) %>% 
  summarise(
    across(
      everything(),
      ~is.na(.) %>% sum()
    )
  )
```

## Strategy for missing values

::: {.column width="70%"}

Often data have missing values.

The most important thing, when you get new data is to figure out how many missing values it contains, and where they are.

Afterward you can decide if you want to remove them, or to impute them.

**Learn more** about missing values at [R4DS](https://r4ds.had.co.nz/transform.html#missing-values).

:::

## Exercise

::: {.column width="70%"}

Identify the columns of the penguins dataset that contain NAs. 

Substitute the missing values:

- With 0s.
- With the mean of the column.

:::