1.3_Data_Visualization.qmd

---
title: "1.3 Data Visualization"
subtitle: | 
  | Workshop: "Handling Uncertainty in your Data"
  |
  | Dr. Mario Reutter & Juli Nagel
  | (slides adapted from Dr. Lea Hildebrandt)
format:
  revealjs:
    smaller: true
    scrollable: true
    slide-number: true
    theme: serif
    chalkboard: true
    width: 1280
    height: 720
from: markdown+emoji
---

# Data Viz

```{css}
code.sourceCode {
font-size: 1.4em;
}

div.cell-output-stdout {
font-size: 1.4em;
}
```

Why should we visualize our data?

::: incremental
-   check whether data make sense (unusual observations?)

-   honestly and comprehensively present the data

-   (visually) check whether data fits the assumptions of statistical tests
:::

. . .

It's fun! (And plots are probably the most important information in papers!)

## ggplot

We will use a package called `ggplot2` (part of the `tidyverse`). `ggplot2` is a very versatile package that allows us to make beautiful, publication-ready(-ish) figures.

`ggplot2` follows the ["grammar of graphics" by Leland Wilkinson](https://ggplot2-book.org/mastery), a formal guide to visualization principles. A core feature are the *layers* each plot consists of. The main function to "start" plotting is `ggplot()` - we will then add layers of data and layers to tweak the appearance.

![Layers of a ggplot](images/ggplotlayers.PNG){fig-alt="Depiction of how a plot is built up from different layers in ggplot2" width="502"}

## Activity 1: Set Up

1.  Within your precision workshop project in RStudio, create a new script called `DataVisualisation1.R`.

2.  Make sure you have the following two files downloaded into your project folder (we already used them in Intro to `R` presentation): [`ahi-cesd.csv`](https://psyteachr.github.io/quant-fun-v2/ahi-cesd.csv) and [`participant-info.csv`](https://psyteachr.github.io/quant-fun-v2/participant-info.csv).

3.  Copy and run the code below to load the `tidyverse` package and the data files:

```{r}
#| echo: true
#| eval: false
#| 
library(tidyverse) 

dat <- read_csv("ahi-cesd.csv")
pinfo <- read_csv("participant-info.csv")
```

```{r}
#| echo: false
#| eval: true
#| message: false
library(tidyverse) 

dat <- read_csv("Data/ahi-cesd.csv")
pinfo <- read_csv("Data/participant-info.csv")
```

## Activity 1: Set Up

4.  Run the following code to combine both files and select our variables of interest:

```{r}
#| echo: true
all_dat <- 
  dat %>% 
  inner_join(
    pinfo, # combine dat with pinfo
    by = c("id", "intervention") # common variables that tell R which data belongs together
  ) %>% 
  arrange(id, occasion) #joining messes up the order of the data frame => arrange again

# we throw out several variables even though they would be important for a comprehensive data analysis
summarydata <- 
  all_dat %>% 
  select(
    id, ahiTotal, cesdTotal, # ID & questionnaire scores
    sex, age, educ, income # demographic variables
  ) 
```

::: notes
what happens in the code chunk?
:::

## Look at the Data

Have a look at the types of data:

```{r}
#| echo: true
glimpse(summarydata)
```

What do you see?

. . .

All variables are loaded as numeric. However, are all of those numeric?

. . .

`sex`, `educ` and `income` don't seem to really be numbers but **factors** with individual categories (factor **levels**)! \
We should convert these data to factor. Checking and adjusting data types (as part of data wrangling) will be important for plotting and analyzing the data, you might otherwise get strange/wrong results!

## Activity 2: Transform to factor

Copy and run the below code to change the categories to factors.

```{r}
#| echo: true
summarydata1 <- 
  summarydata %>%
  mutate(
    sex = as_factor(sex),
    educ = as_factor(educ),
    income = as_factor(income)
  )
```

. . .

-   If you mutate a new column with the same name as the old one, it will overwrite the column.

. . .

```{r}
#| echo: true
glimpse(summarydata1)
```

## Activity 2: Transform to factor

```{r}
#| echo: true
glimpse(summarydata1)
```

. . . 

```{r}
#| echo: true
summarydata1 %>% pull(educ) %>% unique()
```

. . .

-   At first glance, the data look the same. But the 1s and 2s in `sex` are now e.g. not treated as numbers anymore, but as factor levels (categories). This e.g. changes how the data behave in different analyses, or plotting (continuous vs. categorical data).

## Set labels

A simple change to a factor is not always helpful. We still don't know what a `1` in `sex` or a `5` in `educ` stands for:


-   `sex`: 1 = *female*, 2 = *male*

-   `educ`: 1 = *no graduation*, 2 = *school graduation*, 3 = *vocational training*, 4 = *bachelor's degree*, 5 = *post graduate*

-   `income`: 1 = *low*, 2 = *middle*, 3 = *high*

:::aside
There is very sparse information on the variables at <https://doi.org/10.5334/jopd.35>, so I guesstimated some of the factor levels.
:::

## Set labels

`match_case()` allows us to label our numeric data with more human-readable descriptions. `if_else()` is a useful shorthand for cases where we only have two categories.

```{r}
#| echo: true
summarydata2 <- 
  summarydata %>% 
  mutate(
    sex = if_else(sex == 1, "female", "male") %>% as_factor(),
    educ = educ %>% 
      case_match(
        1 ~ "no graduation",
        2 ~ "school graduation",
        3 ~ "vocational training",
        4 ~ "bachelor's degree",
        5 ~ "post grad"
      ) %>%
      as_factor(), # need to transform to factor in the end
    income = income %>% 
      case_match(1 ~ "low", 2 ~ "middle", 3 ~ "high") %>% 
      as_factor()
  )
```

## Set labels

```{r}
#| echo: true
summarydata2 %>% pull(educ) %>% unique()
```

Factor is now ordered by occurrence in data! :( (E.g., the order on the axis of a plot would be wrong.)

## Set factor order (levels)

Using `factor()`, we can explicitly order the categories using the argument `levels`.

```{r}
#| echo: true
summarydata %>% 
  mutate(
    income = income %>%
      case_match(1 ~ "low", 2 ~ "middle", 3 ~ "high") %>% 
      as_factor()
  ) %>% 
  pull(income) %>% unique()
```

. . .

```{r}
#| echo: true
summarydata %>% 
  mutate(
    income = income %>% 
      case_match(1 ~ "low", 2 ~ "middle", 3 ~ "high") %>% 
      factor(levels = c("low", "middle", "high")) # factor(), not as_factor()!
  ) %>% 
  pull(income) %>% unique()
```

# Activity 3: Barplot

A bar plot is a plot that shows counts of categorical data (factors), where the height of each bar represents the count of that variable.

We will plot male and female participants.

## The First Layer

-   The first line (or layer) sets up the base of the graph: the data to use and the aesthetics (`aes()`) (what will go on the x and y axis, how the plot will be grouped).

-   `aes()` includes anything that is *directly* related to your data: e.g., what goes on the x and y axis, or whether the plot should be *grouped* by a variable in your data.

-   We can provide `x` and `y` as arguments, however, in a bar plot, the count per category is calculated automatically, so we don't need to put anything on the y-axis ourselves.

```{r}
#| output-location: column-fragment
#| echo: true
ggplot(summarydata1, aes(x = sex))
```

## The Second Layer

The next layer adds a **geom** or a shape. In this case we use `geom_bar()` as we want to draw a bar plot.

-   Note that we are adding layers, using a [`+`](https://rdrr.io/r/base/Arithmetic.html) between layers. This is a very important difference between pipes and visualization.

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "2"

ggplot(summarydata1, aes(x = sex)) +
  geom_bar()
```

## The Second Layer with color

-   Adding `fill` to the first layer will separate the data into each level of the grouping variable and fill it with a different color. <span style="font-size: 16px;">Note that `fill` colors the inside of the bar, while `colour` colors the bar's outlines.</span>

-   We can get rid of the (in this case redundant legend) with `show.legend = FALSE`.

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "1"
ggplot(summarydata2, aes(x = sex, fill = sex)) +
  geom_bar() #geom_bar(show.legend = FALSE)
```

## The Next Layers - Improving the Plot

We might want to make the plot a bit prettier and easier to read. What would you improve?

. . .

We could add better axis labels, and custom colors. We can do so with the functions `scale_x_discrete()` and `scale_y_continuous()`, which adjust the x and y axes.

Both functions can change several aspects of our axes; here, we use the argument `name` to set a new axis name.

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "3,4|5"
ggplot(summarydata2, aes(x = sex, fill = sex)) +
  geom_bar(show.legend = FALSE) +
  scale_x_discrete(name = "Participant Sex") + 
  scale_y_continuous(name = "Number of participants")
```

::: notes
There's also the counterparts `scale_x_continuous()` and `scale_y_discrete()`. What do you think, why do we use the ones mentioned above and when would we use the other ones?
:::

## Themes: Changing the Appearance

There are a number of built-in themes that you can use to change the appearance (background, whether axes are shown etc.), but you can also tweak the themes further manually.

We will now change the default theme to `theme_minimal()`, but you can also try other themes (just type "theme\_" and see what the autocomplete brings up).

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "5"
ggplot(summarydata2, aes(x = sex, fill = sex)) +
  geom_bar(show.legend = FALSE) +
  scale_x_discrete(name = "Participant Sex") + 
  scale_y_continuous(name = "Number of participants") +
  theme_minimal()
```

## Colors

There are various ways to change the colors of the bars. You can manually indicate the colors you want to use but you can also easily use pre-determined color palettes that are already checked for color-blind friendliness.

A popular palette is **viridis**. We can simply add a function/layer to your ggplot named `scale_fill_viridis_d()` (d for discrete). The function has an *option* parameter that takes 5 different values (A - E).

-   Run the code below. Try changing the option to either A, B, C or D and see which one you like!

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "6"

ggplot(summarydata2, aes(x = sex, fill = sex)) +
  geom_bar(show.legend = FALSE) +
  scale_x_discrete(name = "Participant Sex") +
  scale_y_continuous(name = "Number of participants") +
  theme_minimal() +
  scale_fill_viridis_d(option = "E")
```

## Transparency

You can also add transparency to your plot, which can be helpful if you plot several layers of data that overlap.

To do so, you can simply add `alpha` to the `geom_bar()` - try changing the value of `alpha` (between 0 and 1):

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "3"

ggplot(summarydata2, aes(x = sex, fill = sex)) +
  geom_bar(show.legend = FALSE, 
           alpha = .8) +
  scale_x_discrete(name = "Participant Sex") +
  scale_y_continuous(name = "Number of participants") +
  theme_minimal() +
  scale_fill_viridis_d(option = "E")
```

## Grouped Plots

Imagine that you have several *factors* that you want to use to group your data, such as gender and income. In this case, you could use a **grouped bar plot**:

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "1|4"

ggplot(summarydata2, aes(x = sex, fill = income)) +
  geom_bar(
    # the default are stacked bars; we use "dodge" to put them side by side
    position = "dodge",
    alpha = .8
  ) +
  scale_x_discrete(name = "Participant Sex") +
  scale_y_continuous(name = "Number of participants") +
  theme_minimal() +
  scale_fill_viridis_d(option = "E")
```

:::notes
Without `position = dodge`, you would get a stacked barplot
:::

## Facetting

You could also use **facets** to divide your data visualizations into several subplots: `facet_wrap` for one variable.

```{r}
#| echo: true
#| code-line-numbers: "7"

ggplot(summarydata2, aes(x = sex, fill = sex)) +
  geom_bar(show.legend = FALSE, alpha = .8) +
  scale_x_discrete(name = "Participant Sex") +
  scale_y_continuous(name = "Number of participants") +
  theme_minimal() +
  scale_fill_viridis_d(option = "E")  +
  facet_wrap(vars(income)) # here, you need to use vars() around variable names
```

:::notes
What is problematic here? We needed percentages for females and males for a better comparison
:::

## Facetting 2

You could also use **facets** to divide your data visualizations into several subplots: `facet_grid` for a matrix of (combinations of) two variables.

```{r}
#| echo: true
#| code-line-numbers: "7-11"

ggplot(summarydata2, aes(x = sex, fill = sex)) +
  geom_bar(show.legend = FALSE, alpha = .8) +
  scale_x_discrete(name = "Participant Sex") +
  scale_y_continuous(name = "Number of participants") +
  theme_minimal() +
  scale_fill_viridis_d(option = "E")  +
  facet_grid(
    rows = vars(income),
    cols = vars(educ),
    labeller = "label_both" # this adds the variable name into the facet legends
  )
```

## A closer look

```{r}
ggplot(summarydata2, aes(x = sex, fill = sex)) +
  geom_bar(show.legend = FALSE, alpha = .8) +
  scale_x_discrete(name = "Participant Sex") +
  scale_y_continuous(name = "Number of participants") +
  theme_minimal() +
  scale_fill_viridis_d(option = "E")  +
  facet_grid(
    rows = vars(income),
    cols = vars(educ),
    labeller = "label_both" # this adds the variable name into the facet legends
  )
```

# Activity 4: The Violin-Boxplot

There are a number of different figure types you can plot with the different geoms, e.g. `geom_point()`, `geom_histogram()`, `geom_line()`...

We now want to plot a form of a boxplot that becomes more popular: The violin-boxplot (which combines, i.e. overlays, the violin plot with a boxplot).

## Violin-Boxplot

Let's look at the code. How does the code differ from the one for the barplot above?

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "|2|4|6"

ggplot(summarydata1, aes(x = income, 
                         y = ahiTotal, # new variable!
                         fill = income)) +
  geom_violin(trim = FALSE, # smooth on edges
              alpha = .4) +
  geom_boxplot(width = .2, # small boxplot contained in violin
               alpha = .7) +
  scale_x_discrete(
    name = "Income",
    # set new labels
    labels = c("Below Average", "Average", "Above Average")) +
  scale_y_continuous(name = "Authentic Happiness Inventory Score")+
  theme_minimal() +
  # no need to switch of axis for every geom individually
  theme(legend.position = "none") + 
  scale_fill_viridis_d()
```

::: notes
In this case, not the **count** on the y-axis, but another cont. variable!
:::

## Layer Order

The order of layers is crucial, as the plot will be built up in that order (later layers on **top**):

::: columns
::: column
```{r}
#| echo: true
ggplot(summarydata1, aes(x = income, y = ahiTotal)) +
  geom_violin() +
  geom_boxplot()
```
:::

::: column
```{r}
#| echo: true
ggplot(summarydata1, aes(x = income, y = ahiTotal)) +
  geom_boxplot() +
  geom_violin() 
```
:::
:::

## Scatterplot

If we have continuous data of two variables, we often want to make a **scatter plot**:

```{r}
#| echo: true
#| output-location: column-fragment
#| code-line-numbers: "1|2|5"

ggplot(summarydata1, aes(x = age, y = cesdTotal)) +
  geom_point() +
  # if you don't want the shaded CI, add se = FALSE to this
  geom_smooth(method = lm) 
```

## Saving your Figures

You can use `ggsave()` to save your plots. If you don't tell `ggsave()` which plot you want to save, by default it will save **the last plot you created**.

You just have to enter the name of the file to be saved (in your working directory) like this:

```{r}
#| echo: true
#| eval: false
ggsave("violin-boxplot.png")
```

Check whether indeed the last plot was saved!

. . .

You can also specify the dimensions of your plot to be saved:

```{r}
#| echo: true
#| eval: false
ggsave("violin-boxplot.png",
       width = 6.5, #width of a typical page in inches minus border (according to APA format)
       height = 6.5 / sqrt(2), #golden ratio :)
       units = "in")
```
or
```{r}
#| echo: true
#| eval: false
ggsave("violin-boxplot.png",
       width = 1920,
       height = 1080,
       units = "px") #full HD picture in pixels: 1920 x 1080
```

## Saving your Figures 2

You can also assign the plot to a variable in your environment and then tell `ggsave()` which object to save. This is a bit safer.

Run the code for the violin-boxplot again and save the plot in an object called `viobox`. You'd then have to explicitly tell `ggsave()` to save the object `viobox`:

```{r}
#| echo: true
#| eval: false
#| code-line-numbers: "1,20"

viobox <- 
  ggplot(summarydata1, aes(x = income, y = ahiTotal, fill = income)) +
  geom_violin(trim = FALSE, alpha = .4) +
  geom_boxplot(width = .2, alpha = .7) +
  scale_x_discrete(
    name = "Income",
    labels = c("Below Average", "Average", "Above Average")) +
  scale_y_continuous(name = "Authentic Happiness Inventory Score")+
  theme_minimal() +
  theme(legend.position = "none") + 
  scale_fill_viridis_d()

ggsave("violin-boxplot-stored.png", plot = viobox)
```

::: notes
Do not add `ggsave()` to the plot with a `+`. Instead run it on a separate line!

If plot is assigned to object, it won't be displayed unless you type `viobox` in the console!
:::

# Thanks!

Check out [Chapter 13 of QuantFun](https://psyteachr.github.io/quant-fun-v2/visualisation.html) for further exercises and tips for data visualization!

The free book [R for data science](https://r4ds.had.co.nz/) was written by (some of) the creators of the `tidyverse`, and includes lots of plotting and data science exercises. (For more advanced plotting, check the [ggplot2 book](https://ggplot2-book.org/).)

Also keep in mind: Coding is a lot of googling things!

You can also check out the [R Graph Gallery](https://r-graph-gallery.com/) for code for different types of plots.