Data Visualization Beyond Analysis: Learning by Exploring Data

Before reading the rest of this article, try this interactive chart for a minute or two:

Open the interactive COVID-19 chart

Instead of showing a single answer, the chart lets you click through multiple dimensions—date, region, age, sex, occupation, and other categories—and see the other charts update with the current selection.

Try a simple question:

What changes if I select a particular region and then narrow the data to a specific age group?

Then try another combination.

That small interaction is the starting point of the idea I want to discuss here.

What if data visualization were not only a way to explain data, but also a way to explore an unfamiliar field and generate questions?

I call this idea exploratory learning.

This is not a claim that interactive charts are automatically better for learning. The interesting part is how interaction is designed: what users can change, what questions they are prompted to ask, and how easily they can move between data and other sources of knowledge.

The Multidimensional Charts I Am Building

This chart is one example of the “multidimensional charts” I am building with DC.js.

Multiple charts are linked together, so when you select a condition in one chart, the other charts change accordingly.

The data is not limited to COVID-19.

So far, I have created multidimensional charts for:

• ☘️Biodiversity and nature observation data
• 📜Historical and Cultural Data
• 🗳️Election data
• 🌡️Weather data
• 🎮Game data
• 👥Population and social data
• 🏭 Industry and patent-related data
• ⚾Sports-related data
• 📁Directory usage
• Various other types of data

A list of the datasets, along with information about each dataset and links to the corresponding charts, is available here.

DC.js Multidimensional Chart Data List

Simply browsing this list can help you get an idea of what kinds of data can be explored from multiple dimensions.

As I experimented with different datasets, I gradually began to think that this system could be more than just a “data analysis tool.” It could also be an interface for exploring and understanding unfamiliar fields.

From “Looking” to “Exploring”

Suppose you have the following charts:

• 📅 Date
• 🗺️ Region
• 👤 Age group
• 💼 Occupation
• 👨‍👩‍👧 Sex
• 📊 Count

Now suppose these charts are linked together.

If you click “30s,” all the other charts are filtered to show only people in their 30s. Then, if you click “Engineer,” the data is narrowed down to:

People in their 30s × Engineers

If you then select a particular year, you can explore:

People in their 30s × Engineers × 2021

The user does not need to know the “correct answer” in advance. Instead, they can change the conditions themselves and discover things such as:

“Oh, this region has a particularly large number of people under these conditions.”

“This occupation is more common in this age group.”

“There is a sudden change during this period.”

This is the difference between simply “looking at a graph” and “exploring data.”

From Learning by “Reading Answers” to Learning by “Finding Answers”

In conventional learning, the process often looks something like this:

Read the text
    ↓
Understand the explanation
    ↓
Memorize the knowledge

With interactive data, however, it is possible to create a different process:

Look at the data
    ↓
Click something that interests you
    ↓
Change the conditions
    ↓
Compare the differences
    ↓
Wonder “Why?”
    ↓
Investigate further
    ↓
Gain a deeper understanding

In other words, learning does not have to be limited to “Reading the answer.” It can also involve “Finding the answer.”

This is closely related to Exploratory Data Analysis (EDA), in which data is investigated through repeated cycles such as:

Form a hypothesis → visualize it → examine the result → develop a new question → investigate again

This approach may be useful not only when experts analyze data, but also when people are learning about an unfamiliar field.

Combining Wikipedia with Interactive Data

One combination I find particularly interesting is:

Wikipedia × Interactive Data

Text-based resources such as Wikipedia are useful for understanding:

• An overview of a field
• History
• Terminology
• Background
• Causal relationships
• Related people and organizations

Interactive data visualization, on the other hand, is useful for exploring:

• Magnitude of values
• Changes over time
• Regional differences
• Differences between age groups
• Differences between categories
• Combinations of multiple conditions

In other words, the following cycle becomes possible:

Read an article
  ↓
“So that's what this field is about.”
  ↓
Look at the data
  ↓
“What does it actually look like?”
  ↓
Click and explore
  ↓
“Oh, there is a difference like this.”
  ↓
Read the article again
  ↓
“Now I understand why.”

Text and data visualization do not have to compete with each other. They are simply good at different things.

Learning About Rhinoceros Beetles Through Data

This idea is not limited to specialized fields.

Suppose someone knows nothing about the Japanese rhinoceros beetle.

They could first use Wikipedia or another reference source to learn about:

• Taxonomy
• Geographic distribution
• Habitat
• Food
• Seasonal activity
• Reproduction

But if observation data is available, they can explore the subject from additional perspectives, such as:

• Number of observations by month
• Number of observations by region
• Number of observations by elevation
• Surrounding vegetation
• Organisms observed together
• Differences between daytime and nighttime

Then, instead of merely reading “They are common in summer,” you can investigate:

“Is July really the peak?”

You may then come up with new questions:

“Does the peak occur at different times in different regions?”

“Does the season shift at higher elevations?”

“What plants are commonly recorded together with rhinoceros beetles?”

Here, the data itself does not necessarily become a “textbook.” It may be more useful to think of it as:

Data becoming a learning material that generates the next question.

Interactive Exploration Can Generate Questions

An important aspect of exploratory learning is that users are not simply receiving information. They can manipulate it themselves.

For example:

All data
   │
   ├─ Select a region
   │      ↓
   │   View Region A only
   │
   ├─ Select an age group
   │      ↓
   │   View people in their 30s only
   │
   └─ Select a year
          ↓
       View 2024 only

When users can do this, they can create their own questions:

“What happens under these conditions?”

This ability to create a question for yourself is important in exploratory learning.

When you simply look at a static graph, you are essentially answering a question prepared by its creator. With interactive visualization, however, the user creates a question and asks the data directly.

The “Multidimensional Charts” I Am Building

I am building a system with dc.js that allows users to explore multidimensional data through multiple linked charts.

For example, separate charts can represent:

• Date
• Region
• Age
• Occupation
• User
• Directory
• Category

The key feature is that a selection in one chart is reflected in the other charts.

For example, if “Fukuoka” is selected:

Region: Fukuoka
       ↓
┌──────────────┐
│ By age       │
│ 20s  ███     │
│ 30s  ██████  │
│ 40s  ████    │
└──────────────┘

┌──────────────┐
│ By occupation│
│ Technical █████
│ Sales     ███
│ Office    ██████
└──────────────┘

This makes it possible to ask questions such as:

“How does the occupational distribution change when Region A is selected?”

“What happens to regional differences when I look only at people in their 30s?”

“Which age group is most common in this category?”

I think of this not simply as a “dashboard,” but as an interface for walking around inside the data.

Data Visualization Is Not Only About “Explaining”

There seem to be at least two broad ways to use data visualization.

Visualization for Explanation

Creator
  ↓
Select important information
  ↓
Create an easy-to-understand chart
  ↓
Show it to the audience

This is extremely important in news articles and presentations. The goal is to communicate something like, “This number changed in this way.”

Visualization for Exploration

Data
  ↓
User interacts with it
  ↓
Change the conditions
  ↓
Compare
  ↓
Develop a question
  ↓
Explore further

Here, the goal is for the user to discover meaning from the data.

This is not a matter of saying that one is better than the other. Different visualization designs are appropriate depending on whether the goal is to “communicate” or to “enable exploration.”

Interactive Does Not Automatically Mean Better Learning

This is an important point.

It would be too strong to say that “an interactive graph is always easier to understand than a static graph.” It is also not simply a matter of adding more and more interaction.

Research on interactive visualization and learning suggests that, beyond the amount of interaction, the kinds of operations provided and whether they are appropriate for the learning task are important.

If users are given too much freedom, they may not know where they are supposed to look.

Therefore, when designing exploratory learning experiences, it is not enough to say:

“Let's make everything freely interactive.”

For example, it can be important to:

• Show users where to look first
• Present questions such as “What happens under this condition?”
• Link related charts together
• Make the current selection state easy to understand
• Provide an easy way to return to the original state
• Make it possible to check the meaning and source of the data

In other words:

“Being able to explore freely” and “being easy to explore” are not the same thing.

The Possibility of a “Learning Mode”

This leads to another interesting possibility.

What if interactive data visualization were designed from the beginning with “learning” as one of its purposes?

For example:

Question 1

Between 2020 and 2021, which regions experienced the larger increase?

The user explores the chart.

↓

Question 2

Does this pattern change when you focus only on people in their 30s?

The user changes the conditions again.

↓

Question 3

How do other sources explain this change?

At this point, the user returns to text or external resources.

The learning cycle becomes:

Question
 ↓
Explore the data
 ↓
Discovery
 ↓
Hypothesis
 ↓
Read articles and sources
 ↓
Understanding
 ↓
Next question

This is somewhat different from simply creating a “textbook with graphs.” It is a learning material in which users manipulate the data themselves and discover answers to questions.

Learning About an Unfamiliar Field

If we take this idea further, there are many interesting possibilities.

Suppose you suddenly have to learn about one of these fields:

• Medicine
• Biology
• Environment
• Economics
• Sports
• Agriculture
• History
• Geography
• Demographics
• Industry
• IT

Normally, you might follow a process such as:

Read an introductory book
↓
Learn the terminology
↓
Read reference materials
↓
Look at statistics

An alternative route could be:

Read an overview
↓
Explore the data
↓
Find something interesting
↓
Develop a question
↓
Look up the terminology
↓
Look at the data again

This approach may allow people to gradually discover the structure of a field while looking at the data, even if they do not understand the whole field from the beginning.

Perhaps This Is Somewhere Between “Data Analysis” and “Learning”

Normally, when we say data analysis, we often imagine:

There is a defined objective, and we use data to find an answer.

On the other hand, when we say learning, we often imagine:

Acquiring knowledge that is already known.

Exploratory learning may lie somewhere in between.

             Data analysis
                 ↑
                 │
       “I want to find an answer”
                 │
          Exploratory learning
                 │
       “I want to understand this world”
                 │
                 ↓
              Learning

Exploratory learning does not necessarily require a clearly defined analytical objective from the beginning.

It can start simply with curiosity:

“I wonder what is in this dataset.”

As you explore, you may discover something interesting:

“This is interesting.”

That can lead to another question:

“Why is it like this?”

Then you investigate further. This cycle itself may become a form of learning.

A New Way of Positioning the Charts I Am Building

At first, I thought of this system simply as a “chart for analyzing multidimensional data.” However, after putting many different datasets into it, I began to see another possibility.

For example, when looking at datasets such as:

• COVID-19 data
• Biodiversity and nature observation data
• Demographic data
• Election data
• Patent data
• Sports club participation data
• Directory usage

I sometimes feel that I am doing something less like “analysis” and more like asking:

“What exists in this world represented by the data?”

This may be somewhat different from a conventional BI dashboard.

In BI, people often check predefined KPIs. In exploratory visualization, on the other hand, it is possible to:

“Explore the data even when you do not yet know what you are going to discover.”

What I Would Like to Build Next

If I were to develop this idea further, I would like to build not simply dashboards, but:

“Explorable learning materials.”

For example:

STEP 1: Overview

Briefly explain the field in text.

STEP 2: Look at the Data

Display related data using multiple charts.

STEP 3: Explore

Allow users to filter and explore the data freely through clicks and other interactions.

STEP 4: Present Questions

Ask questions such as, “What happens under this condition?”

STEP 5: Investigate Further

Link to Wikipedia, research papers, statistical sources, and other materials.

STEP 6: Look at the Data Again

Explore the data from a different perspective than before.

This creates a cycle of:

Text → Data → Question → Investigation → Data

“Explore an Unknown World Through Data”

When people hear the term “data visualization,” they may think of:

“Making numbers easier to understand.”

Of course, that is important.

But interactive visualization may have another possibility.

It can become:

An interface for exploring an unfamiliar field.

You read an article and think:

“That is what this is about.”

Then you manipulate the data and ask:

“Is that really the case?”

You can then change the conditions yourself:

“What happens under these conditions?”

You discover something and wonder:

“Why?”

Then you investigate further.

By repeating this process, you are not simply receiving knowledge. You can discover the structure of a field for yourself.

I would like to call this:

“Exploratory learning.”

Data visualization is not only for analysis.

It may also be used as a:

“Tool for exploring and understanding an unfamiliar world.”

References

Charts and Datasets

• DC.js Multidimensional Chart Data List
• COVID-19 Infection Status — Multidimensional Chart

Data Visualization and Interaction

• Kazuki Ogiwara, “The Importance of Interaction in Data Visualization”
• “A Self-Study Note on a Stanford Course That Systematically Covers Data Visualization from History to Implementation — JavaScript / Python”
• Research by Yi, J. S. et al. on classifications of interaction in visualization

Interactive Visualization and Learning

Research on the effects of interactive visualization on learning suggests that, beyond simply having interaction or not, the compatibility between the learning content and the design of the available interactions is important.

Therefore, this article does not claim that “making a visualization interactive will necessarily improve learning outcomes.” Instead, it takes the position that appropriately designed interactions may support exploration, comparison, and hypothesis formation.

Conclusion

This idea is not yet a “finished learning theory.” It is more like a hypothesis I have developed while actually building multidimensional charts.

However, I think the perspectives of:

“From looking at data to exploring data.”

“From reading answers to finding answers.”

suggest that data visualization may have possibilities beyond analysis.

Data visualization may not only be a tool for analysis.

It may also be a tool for exploring and understanding an unfamiliar world.

In the future, I would like to experiment with questions such as:

“How can we design charts that are easier to explore?”

“What kinds of questions lead to learning?”

“What is the appropriate balance between free exploration and guided exploration?”