Bar Charts & Pie Charts

1 Visualizing Comparisons and Compositions

Chart types covered in this lesson.

1.1 Learning objectives

1. Understand the difference between visualizing comparisons and visualizing compositions, and recall the appropriate chart types for these two types of analysis.
2. Create and customize bar charts using {ggplot2} for comparing categorical data, with geom_col(), geom_errorbar(), and position adjustments.
3. Create and customize pie charts using coord_polar() with geom_col().

1.2 Load packages

In this lesson we will use the following packages:

• {tidyverse} for data wrangling and data visualization

• {here} for project-relative file paths

pacman::p_load("tidyverse", "here")

1.3 Data: TB treatment outcomes in Benin

• Public Health Data Analysis: Compare subgroup metrics and understand data contributions.
• Focus on Benin’s TB Data: Investigate WHO-provided sub-national TB data on DHIS2 dashboard.
• Data Composition: Includes new and relapse TB cases started on treatment.
• Disaggregation Categories: Data broken down by time period, health facility, treatment outcome, and diagnosis type.

Let’s import the tb_outcomes data subset.

# Import data from CSV
tb_outcomes <- read_csv(here::here('data/benin_tb.csv'))

# Print data frame
tb_outcomes

Here are the detailed variable definitions for each column:

1. Time Frame Tracking (period and period_date): Quarterly records from 2015Q1 to 2017Q4.

2. Health Facility Identifier (hospital):

   • Data from St Jean De Dieu, CHPP Akron, CS Abomey-Calavi, Hopital Bethesda, Hopital Savalou, Hopital St Luc.

3. Treatment Outcome Categories (outcome):

   • completed: Treatment finished, outcome marked as completed.
   • cured: Treatment succeeded with sputum smear confirmation.
   • died: Patient succumbed to TB during treatment.
   • failed: Treatment did not succeed.
   • unevaluated: Treatment outcome not determined.

4. Diagnosis Categorization (diagnosis_type):

   • bacteriological: Diagnosis confirmed by bacteriological tests.
   • clinical: Diagnosis based on clinical symptoms, sans bacteriological confirmation.

5. Case Counts (cases): Quantifies the number of TB cases starting treatment.

1.3.1 Bar charts

• Bar Chart Advantages: Ideal for displaying counts and making categorical comparisons.
• Optimal Usage:
  • Effective for ordinal categories or time-based data.
  • Best when data is grouped into distinct categories.
• Comparison Tool: Bar charts excellently illustrate comparisons among groups.
• {ggplot2} Implementation: Use geom_col() for plotting categorical against numerical data.

Let’s exemplify this by visualizing the Number of cases per treatment outcomes in the tb_outcomes dataset:

# Basic bar plot example 1: Frequency of treatment outcomes
tb_outcomes %>% 
  # Pass the data to ggplot as a basis for creating the visualization
  ggplot(
    # Specify that the x and y axis variables 
    aes(x = outcome, y = cases)) + 
  #  geom_col() creates a bar plot
  geom_col() +
  labs(title = "Number of cases per treatment outcome")

• Aggregation: geom_col() sums up cases by outcomes, incorporating all periods, hospitals, and diagnosis types.
• Flexible Axes: Easily swap the x-axis variable to display other categorical data dimensions.

# Basic bar plot example 2: Case counts per hospital
tb_outcomes %>% 
  ggplot(aes(x = hospital, y = cases)) + 
  geom_col() +
  labs(title = "Number of Cases per Hospital")

• Horizontal Bar Plot Creation: Use coord_flip() to transform a vertical bar chart into a horizontal layout.
• Enhanced Category Visualization: Horizontal orientation can improve readability of categories.

# Basic bar plot example 3: Horizontal bars
tb_outcomes %>% 
  ggplot(aes(x = hospital, y = cases)) + 
  geom_col() +
  labs(title = "Number of Cases per Hospital") 

  # new code line here:
  coord_flip()

<ggproto object: Class CoordFlip, CoordCartesian, Coord, gg>
    aspect: function
    backtransform_range: function
    clip: on
    default: FALSE
    distance: function
    expand: TRUE
    is_free: function
    is_linear: function
    labels: function
    limits: list
    modify_scales: function
    range: function
    render_axis_h: function
    render_axis_v: function
    render_bg: function
    render_fg: function
    setup_data: function
    setup_layout: function
    setup_panel_guides: function
    setup_panel_params: function
    setup_params: function
    train_panel_guides: function
    transform: function
    super:  <ggproto object: Class CoordFlip, CoordCartesian, Coord, gg>

1.3.2 Stacked bar charts

• Stacked Bar Charts: Introduce a second categorical variable for deeper insight.
• ggplot() Customization: Use fill attribute to differentiate categories within the bars.

# Stacked bar plot: 
tb_outcomes %>% 
  ggplot(
    # Fill color of bars by the 'outcome' variable
    aes(x = hospital, 
        y = cases,
        # new code here
        fill = outcome)) + 
  geom_col()

• Stacked Plot Function: Retain primary categories on the axis while displaying subgroup contributions.
• Visual Segregation: Differentiate subgroups within bars through color-coded segments.

1.3.3 Grouped bar charts

• Grouped bar plots provide a side-by-side representation of subgroups within each main category.
• We can set the position argument to "dodge" in geom_col() to display bars side by side:

# Grouped bar plot: 
tb_outcomes %>% 
  ggplot(
    aes(x = hospital, 
        y = cases,
        fill = outcome)) +
  # Add position argument for side-by-side bars 
  geom_col(position = "dodge")

• Grouped bar charts are not ideal when there are too many groups.
• We can try this again but with a different grouping variable that has fewer categories:

# Grouped bar plot: split into 2 bars
tb_outcomes %>% 
  ggplot(
    # Fill color of bars by the 'diagnosis_type' 
    aes(x = hospital, 
        y = cases,
        # different variable here
        fill = diagnosis_type)) +
  geom_col(position = "dodge")

Question 1: Basic bar plot

Write the adequate code that generates a basic bar chart of the number of cases per quarter with period_date on the x axis

# PQ1 answer:
tb_outcomes %>% 
  ggplot(
    aes(x = period_date, y = cases)) + 
  geom_col()

Question 2: Stacked bar plot

Create a stacked bar chart to display treatment outcomes over different time periods

tb_outcomes %>% 
  ggplot(
    aes(x = period_date, y = cases, fill = outcome)) + 
  geom_col()

1.3.4 Adding error bars

• Showcasing data variability or uncertainty is done effectively with error bars.
• Error bars help illustrate the reliability of mean scores or the precision of data points.
• In {ggplot2}, adding error bars is achieved with the geom_errorbar() function.
• The error range is typically defined by standard deviation, standard error, or confidence intervals.
• Essential summary statistics like mean and standard deviation are necessary for error bars and need to be calculated.
• By integrating error bars into our grouped bar plots, we gain a more nuanced understanding of the data.

First, let’s create the necessary summary data since we need to have some kind of error measurement. In our case we will compute the standard deviation:

hosp_dx_error <- tb_outcomes %>%  
  group_by(period_date, diagnosis_type) %>% 
  summarise(
    total_cases = sum(cases, na.rm = T),
    error = sd(cases, na.rm = T))

hosp_dx_error

# A tibble: 24 × 4
# Groups:   period_date [12]
   period_date diagnosis_type  total_cases error
   <date>      <chr>                 <dbl> <dbl>
 1 2015-01-01  bacteriological         143 11.9 
 2 2015-01-01  clinical                 47  4.40
 3 2015-04-01  bacteriological         163 13.0 
 4 2015-04-01  clinical                 35  3.84
 5 2015-07-01  bacteriological         146 11.2 
 6 2015-07-01  clinical                 34  3.33
 7 2015-10-01  bacteriological         152 10.4 
 8 2015-10-01  clinical                 43  3.55
 9 2016-01-01  bacteriological         201 15.7 
10 2016-01-01  clinical                 71  7.01
# ℹ 14 more rows

Now, let use this data to create the plot:

# Recreate grouped bar chart and add error bars
hosp_dx_error %>% 
  ggplot(
    aes(x = period_date,
        y = total_cases,
        fill = diagnosis_type)) +
  geom_col(position = "dodge") +  # Dodge the bars
  #  geom_errorbar() adds error bars
  geom_errorbar(
    # Specify upper and lower limits of the error bars
    aes(ymin = total_cases - error, ymax = total_cases + error),
    position = "dodge"  # Dodge the error bars to align them with side-by-side bars
  )

1.4 Visualizing comparisons with normalized bar charts, pie charts, and donut charts

• Leet’s explore how compositions illustrate the contributions of individual parts to a whole.
• Consider using dedicated composition chart types that better represent these relationships.
• We’ll focus on part-to-whole charts to highlight how each piece fits into the overall picture.

1.4.1 Percent-stacked bar chart

• For showcasing compositions, we need to identify the parts and the whole they comprise.
• Stacked bar charts, previously discussed, serve as an acceptable starting point for visualizing these relationships.

# Regular stacked bar plot
tb_outcomes %>% 
  ggplot(
    aes(x = hospital, 
        y = cases,
        fill = outcome)) + 
  geom_col()

• They show us parts-of-wholes, but all the wholes are different sizes.
• The height of the bars represent the total number of cases, which is different at every location.
• Looking at the relative distribution of outcomes would be much easier if every bar were the same size.
• We can do this by creating a 100% stacked bar chart, where the total height of each bar is standardized to the same size, effectively showing proportions rather than counts or absolute values.

This is achieved by setting the position argument to "fill" in geom_col().

# Percent-stacked bar plot
tb_outcomes %>% 
  ggplot(
    aes(x = hospital, 
        y = cases,
        fill = outcome)) + 
  # Add position argument for normalized bars
  geom_col(position = "fill")

• All bars are now the same length, meaning all the wholes are now the same size. This now allows us to easily evaluate the contributions of the different parts to the whole.

1.4.2 Circular plots: Pie charts

• Let’s dive into circular data visualizations together, examining pie charts.
• We’ll start by aggregating data to tally total counts for each treatment outcome category.
• This step ensures each segment of our dataset is clearly represented for visualization.

outcome_totals <- tb_outcomes %>% 
  group_by(outcome) %>% 
  summarise(
    total_cases = sum(cases, na.rm = T))

outcome_totals

# A tibble: 6 × 2
  outcome     total_cases
  <chr>             <dbl>
1 completed           573
2 cured              1506
3 died                130
4 failed               30
5 lost                 87
6 unevaluated          15

A pie chart is basically a round version of a single 100% stacked bar.

# Single-bar chart (precursor to pie chart)
ggplot(outcome_totals, 
       aes(x = 4, # Set arbitrary x value
           y = total_cases,
           fill = outcome)) +
  geom_col()

• In ggplot2, we’ll explore how coord_*() functions can change a plot’s perspective, like tweaking aspect ratios or axis limits.
• We’ll transform our plot from linear to polar coordinates using coord_polar(), which will shape our data into slices for a pie chart.
• By mapping the y aesthetic to angles (using the theta argument), we’ll collaboratively create a visual that clearly displays the distribution of our categorical data.

# Basic pie chart
ggplot(outcome_totals, 
       aes(x=4, 
           y=total_cases, 
           fill=outcome)) +
  geom_col() +
  coord_polar(theta = "y") # Change y axis to be circular

1.5 Wrap Up!

Solutions

# PQ1 answer:
tb_outcomes %>% 
  ggplot(aes(x = period_date,
             y = cases)) + 
  geom_col()

# PQ2 answer:
tb_outcomes %>% 
  ggplot(
    aes(x = period_date, 
        y = cases,
        fill = outcome)) + 
  geom_col()

# PQ2 answer:
tb_outcomes %>% 
  ggplot(
    aes(x = period_date, 
        y = cases,
        fill = outcome)) + 
  geom_col()