Lesson 7 — Data Visualization for Communication

(chart choice, storytelling, interactive plots, and honest interpretation)

Why this matters (motivation)¶

In business and economics, most decisions are made from:

• charts in slide decks,

• dashboards,

• short memos with figures.

Even a correct model can be ignored if the story is unclear.
And a misleading chart can create confidence in the wrong conclusion.

The visualization mindset: “What question does this answer?”¶

A practical checklist (before you plot)¶

1. Unit of observation: customer? transaction? month? firm?

2. Aggregation rule: sum, mean, median, rate per 1,000?

3. Comparability: are groups measured the same way?

4. Missingness: are you hiding missing values?

5. Uncertainty: do you need error bars, bands, or sample sizes?

Chart chooser (fast and useful)¶

Common business questions and “good default” charts¶

• “Which product category grew the most?” → sorted bar chart of growth rates

• “Is sales increasing over time?” → line chart (with smoothing optional)

• “Are there segments of customers?” → scatter plot colored by cluster; or boxplots by segment

• “Is marketing associated with sales?” → scatter plot + fitted line; or faceted by category

• “Where are we missing data?” → missingness bar chart by variable or by group

Honest charts: common ways charts mislead (often unintentionally)¶

1) Truncated axes¶

A bar chart with a y-axis that doesn’t start at 0 can exaggerate differences.

2) Aggregation traps¶

Averages can hide important patterns:

• A rising average can be driven by a small number of extreme observations.

• Aggregating across regions/segments can hide opposite trends.

3) Cherry-picking time windows¶

Trends can flip depending on start/end dates.

4) Too much ink / too many categories¶

Clutter weakens interpretation. If there are many categories:

• show top 10 + “other,” or

• facet into multiple charts, or

• allow interaction (filtering).

5) Causality by implication¶

A chart can quietly suggest a causal story (e.g., “marketing causes sales”) even if the analysis only shows correlation.

Building a “data story” (how to present visuals)¶

Template: chart caption that actually helps¶

• What: what the chart shows (variable + unit + time)

• Pattern: what stands out (trend, difference, outlier)

• Meaning: what this might imply for the decision

Example caption:

“Sales in Category A increased steadily after 2021, while Category C declined. The change coincides with a shift in marketing spend; however, the relationship is correlational and may reflect seasonality or pricing differences.”

Visualization and uncertainty (optional but powerful)¶

Sometimes “point estimates” are not enough. Options:

• confidence intervals / error bars (when you have sampling uncertainty),

• rolling averages (to reduce noise),

• showing sample size (n) explicitly,

• distribution plots instead of averages.

Mini case: communicating a regression result visually¶

From Week 6, you may have estimated:

• sales ~ marketing + price + season controls

How to visualize for communication:

• Scatter plot of sales vs marketing with fitted line (simple picture)

• Facet by category or region if patterns differ

• Show time trend of sales and marketing side-by-side

• Boxplots of sales by season/quarter to highlight seasonality

Mini-lab (Google Colab)¶

In-class checkpoints (core)¶

1. Create one chart for each purpose:

   • comparison (bar)

   • trend (line)

   • distribution (hist/box)

   • relationship (scatter)

2. For each chart, add:

   • title with variable and unit

   • readable axis labels

   • a short caption (“headline + meaning”)

In-class checkpoints (storytelling)¶

3. Choose ONE question and build a mini “dashboard” (2–4 charts):

   • Example questions:

     • “Which category is growing and why?”

     • “Which segment has higher churn risk signals?”

     • “How did the key metric change before/after a policy/event date?”

4. Write a short narrative:

   • 1 headline sentence

   • 3 bullet points (evidence)

   • 1 caveat

In-class checkpoints (optional extensions)¶

5. Add an uncertainty element:

   • show sample sizes

   • or show a distribution instead of an average

   • or add a rolling average line (time series)

6. Make an interactive Plotly version of at least one plot with hover labels and filters.

Submission (after class)

• Colab link (view permission) or PDF export.

• Include your 2–4 chart “dashboard” and written narrative.

Tool box: minimal code patterns (for your notebook)¶

(In your Colab notebook, you can include these as “snippets” for students to modify.)

AI check (responsible use for visualization)¶

Good prompt examples

• “Given this question and these columns, suggest 3 chart options and explain what each would show.”

• “Write a caption template that includes a caveat and avoids causal claims.”

• “What is a common visualization pitfall for comparing averages across groups?”

Bad prompt example

• “Write a persuasive story that marketing caused sales to rise” (without justification)

Review questions (quiz / reflection)¶

1. Which chart type would you use for: (i) trend, (ii) distribution, (iii) relationship—and why?

2. What is one example of a misleading chart design choice and how to fix it?

3. Why can aggregation hide important patterns? Give one example from business/economics.