---
title: "GGPlot2 Introduction"
author: "Arno Kimeswenger"
output:
  html_document:
    code_folding: none # no hide/show butten in html document
    number_sections: true
    css: "styles.css"
editor_options: 
  markdown: 
    wrap: 72
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(
  warning = TRUE,
  message = TRUE,
  error = TRUE
)
options(warn = 1)
```


# Preliminary Work

```{r}
library("tidyverse")
```


# Load Data

Have a look at variables and scale levels:

```{r, tidy = FALSE}
data_toy_properties <- tribble(
  ~variable_name,   ~values,                              ~scale_level,                ~quantity1,   ~quantity2, ~r_type,
  "austrian",       "TRUE and FALSE",                     "nominal",                   "discrete",   2,          "logical",
  "marital_status", "married, single, divorced, widowed", "nominal",                   "discrete",   4,          "factor",
  "name",           "string",                             "nominal",                   "discrete",   Inf,        "character",
  "?",              "?",                                  "nominal",                   "continuous", Inf,        "?",
  "age_cut",        "child, adolescent, adult",           "ordinal",                   "discrete",   3,          "ordered factor",
  "?",              "?",                                  "ordinal",                   "continuous", Inf,        "?",
  "year_birth",     "2023, 2024, etc.",                   "metric (interval-scaled)",  "discrete",   Inf,        "integer",
  "temperature",    "[20, 50] °C",                        "metric (interval-scaled)",  "continuous", Inf,        "double",
  "test_score",     "{0, 1, ..., 100} points",            "metric (ratio-scaled)",     "discrete",   101,        "integer",
  "height",         "[0, 300] cm",                        "metric (ratio-scaled)",     "continuous", Inf,        "double"
)

data_toy_properties
```

Or more compact:

```{r}
data_toy_properties_wide_1 <- data_toy_properties |>
  select(variable_name, scale_level, quantity1) |>
  pivot_wider(
    names_from = quantity1, # take values from quantity1 (discrete and continuous) and make columns
    values_from = variable_name, # these values are inserted in the new columns discrete and continuous
    # there is more than one value per cell to insert, therefore a function is used in each cell
    values_fn = ~ paste(.x, collapse = ", ") # short form of anonymous function
    # values_fn = function(x) paste(x, collapse = ", ")  # long form of anonymous function
  ) |>
  rename(
    "scale level \\ quantity" = scale_level # rename column scale_level
  )

data_toy_properties_wide_1
```

The following is the "inverse" map with pivot_longer. pivot_longer is used often to prepare data for plots.

```{r}
data_toy_properties_wide_1 |>
  pivot_longer(
    cols = c("discrete", "continuous"), # chose columns
    names_to = "quantity_1", # the names (discrete and continuous) will be stored in new variable quantity_1
    values_to = "variable_name" # the values (austrian, marital_status, name)
  ) |>
  separate_rows(variable_name, sep = "\\s*,\\s*") |> # to get for e.g. austrian, marital_status, name 3 rows instead of one
  rename("scale_level" = "scale level \\ quantity") |> # rename column name
  select(variable_name, scale_level, quantity_1) # new ordering
```

And how it is used in R:

```{r}
data_toy_properties |>
  select(r_type, scale_level, quantity1) |>
  pivot_wider(
    names_from = quantity1, # take values from quantitiy1 (discrecte and continuous) and make columns
    values_from = r_type, # these values are inserted in the new columns discrete and continuous
    # there is more than one value per cell to insert, therefore a function is used in each cell
    values_fn = ~ paste(.x, collapse = ", ") # short form of anonymous function
    # values_fn = function(x) paste(x, collapse = ", ")  # long form of anonymous function
  ) |>
  rename(
    "scale level \\ quantity" = scale_level
  )
```

Read data with warning, because we do not specify the column types:

```{r}
data_toy_temp <- read_csv("toy.csv")
data_toy_temp |> head(3)
```

Store values for marital_status:

```{r}
marital_status_values <- data_toy_temp |>
  pull(marital_status) |>
  unique()
marital_status_values
```

Store values for age.cat:

```{r}
age_cat_values <- data_toy_temp |>
  pull(age_cat) |>
  unique()
age_cat_values
```

Change ordering (maybe it was already ok) and erase NA

```{r}
age_cat_values <- age_cat_values[c(3, 2, 1)]
age_cat_values
```

Read data again:

```{r}
data_coltypes <- cols(
  name = col_character(),
  austrian = col_logical(),
  marital_status = col_factor(marital_status_values),
  year_birth = col_integer(),
  age_cat = col_factor(levels = age_cat_values, ordered = TRUE),
  temperature = col_double(),
  test_score = col_integer()
)

data_toy <- read_csv("toy.csv", col_types = data_coltypes)

data_toy |>
  head(3)
```

```{r}
data_toy |>
  glimpse()
```

Levels of all factor variables:

```{r}
data_toy |>
  select(where(is.factor)) |>
  summary()
```

Levels of all ordered variables:

```{r}
data_toy |>
  select(where(is.ordered)) |>
  summary()
```

Full summary:

```{r}
data_toy |>
  summary()
```


# Some Boxplots

## Simple boxplot

year_birth has na-values hence warnings:

```{r, fig.width=5}
data_toy |> # data
  ggplot(aes(y = year_birth)) + # aesthetics (define what is on y axis) is used for all geom objects
  geom_boxplot() # define geom object (plot type here boxplot)
```

We can define aesthetics in geom_boxplot instead of ggplot.
If we have more geom objects, then this can be necessary.
Here it makes no difference.

```{r, fig.width=5}
data_toy |> # data
  ggplot() + # empty command
  geom_boxplot(aes(y = year_birth)) # aesthetics is defined only for this geom object
```

To avoid warning, we could use filter to exclude rows where year_birth is NA.
Or we use na.rm = TRUE in ggplot to do the same.

```{r, fig.width=5}
data_toy |>
  ggplot(aes(y = year_birth)) +
  geom_boxplot(na.rm = TRUE)
```

Change dimensions in code chunk with e.g. {r, fig.width=1, fig.height=1.2} (the real text size stays the same at 11pt).
Be careful: Here in r chunk we can only use inch not cm.
So if we want to get the same text/plot ratio in e.g. pdf, we should also use inch later in ggsave.

```{r, fig.width=1, fig.height=1.2}
data_toy |>
  ggplot(aes(y = year_birth)) +
  geom_boxplot(na.rm = TRUE)
```

Or with {r, fig.width=2, fig.height=2.4}

```{r, fig.width=2, fig.height=2.4}
data_toy |>
  ggplot(aes(y = year_birth)) +
  geom_boxplot(na.rm = TRUE)
```

Save last plot. If we use width and height we change again the text/plot ratio because the textsize stays constant at 11pt.
If we want to have the same text/plot ratio as in above r chunk, we should use inch in ggsave.
Formats png and jpg are no vector graphics, I prefer pdf and svg in most cases.

```{r}
ggsave("my_first_plot_a.pdf", width = 2, height = 2.4, units = "in") # compare text/plot ratio, should be as in last plot
ggsave("my_first_plot_b.pdf", width = 1, height = 1.2, units = "in") # compare text/plot ratio, should be as in plot above last plot
ggsave("my_first_plot_c.pdf", width = 2, height = 2.4, units = "in", dpi = 100) # compare file size, does not change for pdf (vector graphics) so dpi is unnecessary
ggsave("my_first_plot_d.pdf", width = 2, height = 2.4, units = "in", dpi = 600) # compare file size, does not change for pdf (vector graphics) so dpi is unnecessary
ggsave("my_first_plot_e.png", width = 2, height = 2.4, units = "in", dpi = 100) # here dpi makes a difference
ggsave("my_first_plot_f.png", width = 2, height = 2.4, units = "in", dpi = 600) # here dpi makes a difference compare file size with plot e
```

Or change text size for (ideally all) plots in document:

```{r, fig.width=2, fig.height=2.4}
data_toy |>
  ggplot(aes(y = year_birth)) +
  geom_boxplot(na.rm = TRUE) +
  theme_grey(base_size = 16)
```

```{r}
ggsave("my_first_plot_g.pdf", width = 2, height = 2.5, units = "in")
```

It is also possible to store the plot as variable:

```{r}
plot1 <- data_toy |>
  ggplot(aes(y = year_birth)) +
  geom_boxplot(na.rm = TRUE)
```

Next we can show the plot:

```{r, fig.width=5}
plot1
```

Modify the plot e.g. theme:

```{r, fig.width=5}
plot1 +
  # theme_classic()
  # theme_dark()
  # theme_minimal()
  # theme_grey()
  theme_bw()
```

And of course save the plot:

```{r}
ggsave("my_first_plot_h.pdf", plot1, width = 2, height = 2.4, units = "in")
```


## Colors and transparency

```{r, fig.height=2, fig.width=10}
fh1 <- rgb(r = 97 / 255, g = 164 / 255, b = 215 / 255) # fh lightblue
fh1a <- rgb(r = 97 / 255, g = 164 / 255, b = 215 / 255, alpha = 0.15) # transparancy: fh1 15% and background (maybe white) 85%
factor <- 0.15
fh1b <- rgb(t(col2rgb(fh1) / 255 * factor + 1 * (1 - factor))) # fh1 15% and white 85%
fh2 <- rgb(r = 22 / 255, g = 48 / 255, b = 114 / 255) # fh darkblue

ggplot() +
  geom_rect(aes(xmin = 0, xmax = 1, ymin = 0, ymax = 1), fill = fh1) + # fh1
  geom_rect(aes(xmin = 1, xmax = 2, ymin = 0, ymax = 1), fill = fh1a) + # fh1a, grids are darker, because background is gray and not white
  geom_rect(aes(xmin = 2, xmax = 3, ymin = 0, ymax = 1), fill = fh1b) + # fh1b, no transparency, so we do not see the grids
  geom_rect(aes(xmin = 3, xmax = 4, ymin = 0, ymax = 1), fill = fh2) + # fh2
  theme_bw() # it is important to use white background, otherwise transparancy of fh1a is different, see grid lines
```


## Some properties

```{r}
data_toy |>
  ggplot(aes(y = year_birth)) +
  geom_boxplot(
    color = fh2, # lines are plotted in fh2
    fill = fh1, # areas are plotted in fh1
    linewidth = 1, # width of lines
    median.linewidth = 1, # with of median line
    outlier.shape = 23, # shape of outlier
    outlier.size = 3, # size of outlier
    outlier.stroke = 1, # width of outlier lines
    na.rm = TRUE
  ) +
  labs(
    title = "Year of Birth",
    y = ""
  ) +
  scale_x_continuous(breaks = NULL) + # This ensures no x-axis labels or ticks
  scale_y_continuous( # control the position of breaks, normally I keep standard values
    breaks = 1920 + (0:4) * 20,
    minor_breaks = 1930 + (0:4) * 20
  ) +
  theme(
    panel.background = element_rect(fill = fh1b), # background color
    panel.grid.major.y = element_line(color = "black", linetype = "solid", linewidth = 0.1), # major y-grid
    panel.grid.minor.y = element_line(color = "black", linetype = "dashed", linewidth = 0.05), # minor y-grid
    # panel.grid.minor.y = element_blank(),  # or if we do not like to see minor grid
    text = element_text(face = "bold", size = 11), # font properties
    axis.ticks = element_blank() # no ticks e.g. right of values 1940, 1960, ... on y-axis
  )
```


## More than one geom in one plot

We could store the above plot as variable e.g. plot2 and reuse it here. but it can make a difference which geom object (violin, boxplot) and gridlines etc. are used first.

```{r}
data_toy |>
  ggplot(aes(x = 0, y = year_birth)) + # geom_violin and geom_point need x value
  geom_violin( # violin plot
    linewidth = 1,
    color = fh2,
    fill = fh1,
    alpha = 0.3,
    width = 4.5,
    na.rm = TRUE
  ) +
  geom_boxplot( # boxplot
    color = fh2,
    fill = fh1,
    linewidth = 1,
    median.linewidth = 1,
    outlier.shape = 23,
    outlier.size = 3,
    outlier.stroke = 1,
    na.rm = TRUE
  ) +
  stat_boxplot( # error bars
    geom = "errorbar",
    color = fh2,
    linewidth = 1,
    na.rm = TRUE
  ) +
  geom_point( # data points
    position = position_jitter(width = 0.2), # jittered is used (randomly x value)
    color = fh2,
    na.rm = TRUE
  ) +
  labs(
    title = "Year of Birth",
    x = "",
    y = ""
  ) +
  scale_x_continuous(
    breaks = NULL # This ensures no x-axis labels or ticks
  ) +
  scale_y_continuous( # control the position of breaks, normally I keep standard values
    breaks = 1920 + (0:4) * 20,
    minor_breaks = 1930 + (0:4) * 20
  ) +
  theme(
    panel.background = element_rect(fill = fh1b), # background color
    panel.grid.major.y = element_line(color = "black", linetype = "solid", linewidth = 0.1), # major y-grid
    panel.grid.minor.y = element_line(color = "black", linetype = "dashed", linewidth = 0.05), # minor y-grid
    # panel.grid.minor.y = element_blank(),  # or if we do not like to see minor grid
    text = element_text(face = "bold", size = 11), # font properties
    axis.ticks = element_blank() # no ticks e.g. right of values 1940, 1960, ... on y-axis
  )
```


## Plot by factor variable

Plot by marital_status

```{r, fig.width=8}
data_toy |>
  ggplot(aes(x = marital_status, y = year_birth)) + # we get 4 boxplot because marital_status has 4 values
  geom_boxplot(
    color = fh2,
    fill = fh1,
    linewidth = 1,
    median.linewidth = 1,
    outlier.shape = 23,
    outlier.size = 3,
    outlier.stroke = 1,
    na.rm = TRUE
  ) +
  stat_boxplot(
    geom = "errorbar",
    color = fh2,
    linewidth = 1,
    na.rm = TRUE
  ) +
  labs(
    title = "Year of Birth\nBy marital status",
    x = "",
    y = ""
  ) +
  scale_y_continuous( # control the position of breaks, normally I keep standard values
    breaks = 1920 + (0:4) * 20,
    minor_breaks = 1930 + (0:4) * 20
  ) +
  theme(
    panel.background = element_rect(fill = fh1b),
    panel.grid.major.y = element_line(color = "black", linetype = "solid", linewidth = 0.1),
    panel.grid.major.x = element_blank(),
    panel.grid.minor.y = element_line(color = "black", linetype = "dashed", linewidth = 0.05),
    text = element_text(face = "bold", size = 11),
    axis.ticks = element_blank()
  )
```


## Using variable for property

We again plot by marital_status, but furthermore we use this variable for color and fill.
There is no real advantage in using different colors here for the 3 boxes.

```{r, fig.width=8}
factor2 <- 0.3
darkred_alpha <- rgb(t(col2rgb("darkred") / 255 * factor2 + 1 * (1 - factor2)))
darkgreen_alpha <- rgb(t(col2rgb("darkgreen") / 255 * factor2 + 1 * (1 - factor2)))
darkblue_alpha <- rgb(t(col2rgb("darkblue") / 255 * factor2 + 1 * (1 - factor2)))
gray42_alpha <- rgb(t(col2rgb("gray42") / 255 * factor2 + 1 * (1 - factor2)))

data_toy |>
  ggplot(aes(x = marital_status, y = year_birth)) +
  geom_boxplot(aes(color = marital_status, fill = marital_status), # each boxplot gets its own color, we do not use default colors, therefore we use scale_color_manual and scale_fill_manual later
    linewidth = 1,
    median.linewidth = 1,
    outlier.shape = 23,
    outlier.size = 3,
    outlier.stroke = 1,
    na.rm = TRUE
  ) +
  stat_boxplot(aes(color = marital_status),
    geom = "errorbar",
    linewidth = 1,
    na.rm = TRUE
  ) +
  labs(
    title = "Year of Birth\nBy marital status",
    x = "",
    y = ""
  ) +
  scale_color_manual(values = c("single" = "darkred", "married" = "darkblue", "divorced" = "darkgreen", "widowed" = "gray42")) + # use our own color selection for color
  scale_fill_manual(values = c("single" = darkred_alpha, "married" = darkblue_alpha, "divorced" = darkgreen_alpha, "widowed" = gray42_alpha)) + # use our own color selection for fill
  theme(
    legend.position = "none", # legend is not necessary here
    axis.ticks = element_blank()
  )
```


## Using facet grid

Here we use marital_status again:

```{r, fig.width=8}
data_toy |>
  ggplot(aes(y = year_birth)) + # we do not use x="marital_status" because we use facets
  geom_boxplot(aes(color = marital_status, fill = marital_status),
    linewidth = 1,
    median.linewidth = 1,
    outlier.shape = 23,
    outlier.size = 3,
    na.rm = TRUE
  ) +
  stat_boxplot(aes(color = marital_status),
    geom = "errorbar",
    linewidth = 1,
    na.rm = TRUE
  ) +
  labs(
    title = "Geburtsjahr\nnach Familienstand",
    x = "",
    y = ""
  ) +
  facet_grid(. ~ marital_status, # facet creates again 4 boxplots
    labeller = as_labeller(c("single" = "ledig", "married" = "verheiratet", "divorced" = "geschieden", "widowed" = "verwitwet")) # be careful here!
  ) +
  scale_x_continuous(
    breaks = NULL # This ensures no x-axis labels or ticks
  ) +
  scale_y_continuous( # control the position of breaks, normally I keep standard values
    breaks = 1920 + (0:4) * 20,
    minor_breaks = 1930 + (0:4) * 20
  ) +
  scale_color_manual(values = c("single" = "darkred", "married" = "darkblue", "divorced" = "darkgreen", "widowed" = "gray42")) + # use our own color selection for color
  scale_fill_manual(values = c("single" = darkred_alpha, "married" = darkblue_alpha, "divorced" = darkgreen_alpha, "widowed" = gray42_alpha)) + # use our own color selection for fill
  theme(
    legend.position = "none", # we do not need legend here
    axis.ticks = element_blank() # no ticks on x-axis
  )
```

But we can use another factor to get a "matrix". Here we use austrian for rows and marital_status for columns.

```{r, fig.width=8}
data_toy |>
  filter(!is.na(austrian)) |> # we do not want to get a row for austrian == NA
  ggplot(aes(y = year_birth)) + # we do not use x="marital_status" because we use facet
  geom_boxplot(aes(color = marital_status, fill = marital_status),
    linewidth = 1,
    median.linewidth = 1,
    outlier.shape = 23,
    outlier.size = 3,
    na.rm = TRUE
  ) +
  stat_boxplot(aes(color = marital_status),
    geom = "errorbar",
    linewidth = 1,
    na.rm = TRUE
  ) +
  labs(
    title = "Geburtsjahr\nnach Staatsbürgerschaft und Familienstand",
    x = "",
    y = ""
  ) +
  facet_grid(austrian ~ marital_status,
    labeller = as_labeller(c(
      "single" = "ledig", "married" = "verheiratet", "divorced" = "geschieden", "widowed" = "verwitwet", # be careful here
      "TRUE" = "Österreicher/in", "FALSE" = "kein/e Österreicher/in" # be careful here
    ))
  ) +
  scale_x_continuous(
    breaks = NULL # This ensures no x-axis labels or ticks
  ) +
  scale_y_continuous( # control the position of breaks, normally I keep standard values
    breaks = 1920 + (0:4) * 20,
    minor_breaks = 1930 + (0:4) * 20
  ) +
  scale_color_manual(values = c("single" = "darkred", "married" = "darkblue", "divorced" = "darkgreen", "widowed" = "gray42")) + # use our own color selection for color
  scale_fill_manual(values = c("single" = darkred_alpha, "married" = darkblue_alpha, "divorced" = darkgreen_alpha, "widowed" = gray42_alpha)) + # use our own color selection for fill
  theme(
    legend.position = "none", # we do not need legend here
    axis.ticks = element_blank() # no ticks on x-axis
  )
```

Maybe we want to use different colors for Austrian as well. For that we define a new variable in our data_toy.

```{r}
# for demonstration only
data_toy |>
  mutate(
    color = interaction(marital_status, austrian, sep = "_")
  ) |>
  distinct(color)
```

```{r, fig.width=8}
factor2 <- 0.3 # blend with white
darkred_alpha <- rgb(t(col2rgb("darkred") / 255 * factor2 + 1 * (1 - factor2)))
darkgreen_alpha <- rgb(t(col2rgb("darkgreen") / 255 * factor2 + 1 * (1 - factor2)))
darkblue_alpha <- rgb(t(col2rgb("darkblue") / 255 * factor2 + 1 * (1 - factor2)))
gray42_alpha <- rgb(t(col2rgb("gray42") / 255 * factor2 + 1 * (1 - factor2)))

factor3 <- 0.5 # blend with black
darkred_alpha2 <- rgb(t(col2rgb("darkred") / 255 * factor3 + 0 * (1 - factor3)))
darkgreen_alpha2 <- rgb(t(col2rgb("darkgreen") / 255 * factor3 + 0 * (1 - factor3)))
darkblue_alpha2 <- rgb(t(col2rgb("darkblue") / 255 * factor3 + 0 * (1 - factor3)))
gray42_alpha2 <- rgb(t(col2rgb("gray42") / 255 * factor3 + 0 * (1 - factor3)))

# blend with white
darkred_alpha3 <- rgb(t(col2rgb(darkred_alpha2) / 255 * factor2 + 1 * (1 - factor2)))
darkgreen_alpha3 <- rgb(t(col2rgb(darkgreen_alpha) / 255 * factor2 + 1 * (1 - factor2)))
darkblue_alpha3 <- rgb(t(col2rgb(darkblue_alpha2) / 255 * factor2 + 1 * (1 - factor2)))
gray42_alpha3 <- rgb(t(col2rgb(gray42_alpha2) / 255 * factor2 + 1 * (1 - factor2)))

data_toy |>
  filter(!is.na(austrian)) |>
  mutate(color = interaction(marital_status, austrian, sep = "_")) |> # add the color e.g. single_FALSE
  ggplot(aes(y = year_birth)) + # we do not use x = "marital_status" because we use facet
  geom_boxplot(aes(color = color, fill = color),
    linewidth = 1,
    median.linewidth = 1,
    outlier.shape = 23,
    outlier.size = 3,
    na.rm = TRUE
  ) +
  stat_boxplot(aes(color = color),
    geom = "errorbar",
    linewidth = 1,
    na.rm = TRUE
  ) +
  labs(
    title = "Geburtsjahr\nnach Staatsbürgerschaft und Familienstand",
    x = "",
    y = ""
  ) +
  facet_grid(austrian ~ marital_status,
    labeller = as_labeller(c(
      "single" = "ledig", "married" = "verheiratet", "divorced" = "geschieden", "widowed" = "verwitwet", # be careful here
      "TRUE" = "Österreicher/in", "FALSE" = "kein/e Österreicher/in" # be careful here
    ))
  ) +
  scale_x_continuous(
    breaks = NULL # This ensures no x-axis labels or ticks
  ) +
  scale_y_continuous( # control the position of breaks, normally I keep standard values
    breaks = 1920 + (0:4) * 20,
    minor_breaks = 1930 + (0:4) * 20
  ) +
  scale_color_manual(values = c(
    "single_FALSE" = "darkred", "single_TRUE" = darkred_alpha2,
    "married_FALSE" = "darkblue", "married_TRUE" = darkblue_alpha2,
    "divorced_FALSE" = "darkgreen", "divorced_TRUE" = darkgreen_alpha2,
    "widowed_FALSE" = "gray42", "widowed_TRUE" = gray42_alpha2
  )) + # use our own color selection for color
  scale_fill_manual(values = c(
    "single_FALSE" = darkred_alpha, "single_TRUE" = darkred_alpha3,
    "married_FALSE" = darkblue_alpha, "married_TRUE" = darkblue_alpha3,
    "divorced_FALSE" = darkgreen_alpha, "divorced_TRUE" = darkgreen_alpha3,
    "widowed_FALSE" = gray42_alpha, "widowed_TRUE" = gray42_alpha3
  )) + # use our own color selection for fill
  theme(
    legend.position = "none", # we do not need legend here
    axis.ticks = element_blank() # no ticks on x-axis
  )
```