Data Exploration and Visualization

Using visualization as a thinking tool to discover patterns, relationships, and uncertainties in science data investigations.

Context: Electrical Fire Incidents and Contributing Factors

Throughout this module, we continue working with our running example of investigating electrical fire incidents, focusing on how visualization helps reveal patterns and relationships that inform our understanding of contributing factors.

Why This Phase Matters

Once data has been cleaned and structured, the next step is to explore it visually. Exploration is not about proving a conclusion, it is about discovering patterns, relationships, anomalies, and uncertainties that help refine understanding of a real-world problem.

In investigations of electrical fires, visualization helps reveal:

This phase is closely connected to modeling and interpretation. What you see here may lead you to revise your research question, request additional data, or rethink earlier assumptions.

A Critical Stance Toward LLMs

Throughout this phase, LLMs function as idea generators and explanation partners, not authorities.

Any chart, interpretation, or narrative produced with LLM assistance must be:

Exploration and Visualization Workflow

The following seven-step workflow guides the process of exploring data visually and communicating findings responsibly.

Step 1: Clarify the Visualization Goal

Before creating any chart, you may restate my investigative question and decide what kind of insight you are looking for.

Electrical fire examples:

Each question implies a different visualization goal:

Prompting technique: Zero-shot prompting

“Given this research question about electrical fire causes, what type of insight should a visualization aim to show: trend, comparison, distribution, or relationship?”

Step 2: Select Chart Type and Map Variables

Once the goal is clear, choose a chart that matches it and decide how variables map to:

Examples:

Prompting technique: Few-shot prompting

“Here is my data on fire safety analysis.
Based on these, suggest an appropriate chart for comparing electrical fire incidents across wire materials.”

I treat the output as a suggestion, not a final decision.

Step 3: Build a First-Draft Visualization

At this stage, the goal is exploration, not polish.

Generate an initial chart and ask:

Prompting technique: Prompt chaining

Prompt 1:

“Create a simple bar chart comparing electrical fire incidents by wire material.”

Prompt 2:

“Explain what patterns or differences are most noticeable in this chart.”

Step 4: Interpret What the Data Shows

Interpretation involves describing what the data actually displays, not explaining why it happens.

Separate:

Example interpretation:

Observed: “Buildings with wiring older than 30 years appear more frequently in the dataset.”

Speculation: “Older insulation materials may degrade over time, increasing fire risk.”

Prompting technique: Chain-of-thought prompting

“Step by step, describe the main patterns in this electrical fire visualization.
Clearly separate observed data patterns from possible explanations. Use the attached data as context”

Step 5: Critique the Visualization

Before communicating results, check whether the visualization could mislead.

Key questions:

Prompting technique: Self-consistency

“Review this visualization for potential sources of misinterpretation.
Then review it again from the perspective of a skeptical reader.”

This helps surface issues you might overlook.

Step 6: Communicate Findings

Once the visualization is accurate and interpretable, translate it into communication products:

Each figure includes:

Step 7: Review, Revise, and Archive

Data visualization is rarely final on the first attempt.

You should:

This ensures transparency and reproducibility.

Key Takeaway

Exploration and visualization are not just about presentation, they are thinking tools. In investigating electrical fire data, visualizations help surface relationships, challenge assumptions, and guide next steps in analysis.

LLMs can accelerate this process by suggesting options and explanations, but the responsibility for accuracy, interpretation, and ethical use always remains with the investigator.

Enjoy Reading This Article?

Here are some more articles you might like to read next:

  • On Conducting Interviews: A guide for UX and HAI Researchers
  • What AI Sycophancy Means for K-12 Students
  • Prompting Techniques for Science Data Analysis and Sensemaking
  • Prompt Engineering Framework for Data Sensemaking
  • Intro to Large Language Models