N noduly critical thinking · reading data
🔥0 · 0/5 All modules

Reading Data Critically

A chart is an argument made with ink. Same numbers, different chart — different conclusion. Learn the dozen design choices that can flip a story without changing a single data point.

The five-second checklist

Before you believe a chart, read it slowly. Every time:

  1. Axes. What do they measure? Do they start at zero? Are they linear or log? Two y-axes?
  2. Scale and units. Per-capita or total? Adjusted for inflation? Same units across categories?
  3. Window. Why this time range? What happens before or after?
  4. Sample. Who's in the data? Who's missing? How were they chosen?
  5. Comparison. Compared to what? Where's the baseline?

Why this matters

Edward Tufte calls bad data graphics "chartjunk" — visual noise that hides or distorts the underlying numbers. Cairo and Wainer call the same techniques "graphical lies." Both agree: a misleading chart is rarely fake — it just shows true numbers in a deceptive shape.

The fixes are simple: zero the baseline, label every axis, give context, show the comparison, prefer rates over totals when populations differ.

Trick #1 — Truncated y-axis

The same data, two stories

Five companies' quarterly revenue (millions). Identical numbers, identical bars. On the left, the y-axis starts at zero; on the right, it starts at 95. Move the slider — see how dramatic the "growth" looks when the axis is truncated.

✓ Honest — y-axis at 0
Differences are real, but small (~10%). The chart shows that.
⚠️ Misleading — y-axis truncated
Same numbers, but the differences look enormous. Eye reads ratio of bar heights — bars now lie about the ratio.
Rule: for bar charts of magnitudes, start the y-axis at zero. For change over time of a roughly-constant quantity (stock price, temperature) a non-zero baseline can be honest — but label it clearly.

Trick #2 — Cherry-picked window

"The market is collapsing" — or is it?

A stock index over time. Use the slider to choose the visible window. Watch how the same series tells the opposite story depending on where you start and stop.

Rule: always show the longest comparable window unless there is a specific, named reason to zoom in. Demand that reason.

Trick #3 — Per-capita vs. total

Big country, small problem?

Compare four countries on the same metric. Toggle between total and per-capita — the ranking flips entirely.

Total
Larger populations dominate when you ignore per-capita.
Quick math
USA: 333 M people · 700,000 events → 2,103/M
India: 1428 M · 1,400,000 events → 980/M
Iceland: 0.37 M · 1,200 events → 3,243/M
Brazil: 215 M · 400,000 events → 1,860/M
Iceland leads when adjusted for population — invisible in totals.
Rule: when comparing populations of different sizes, use rates (per-capita, per-100k, per-1M). Totals are useful only when "how many" is the actual question.

Trick #4 — Correlation is not causation

Spurious correlations are easy to find

Two real series can correlate at r > 0.9 with no causal link. Some classics from Tyler Vigen's collection — the data is real; the connection isn't.

Rule: a correlation means something but rarely means causation. Possibilities: A causes B, B causes A, a third factor C causes both, or pure coincidence at small sample size.

Catalog of chart tricks

Fix the deceptive chart

"Country X's pollution is exploding!"

Below is a real-style headline chart pushing the narrative "Country X is now the world's biggest polluter." It uses three design choices to make the story land. Toggle each fix on and watch the conclusion change.

⚠️ Deceptive — all three tricks active

Real-world lesson: any one of these "tricks" can be defensible if explicitly called out (a stock chart often non-zero, a recent window if you flag the broader trend, totals if asking the absolute-emissions question). The deception isn't in the choice — it's in not telling the reader what was chosen.

Connect the dots

Quiz

15 questions on chart literacy.

0
Score
0
Streak
1/15
Question

Flashcards

Tap to flip. Twenty terms every chart-reader needs.

1 / 20
Mastery: —
Space flip · J/ next · K/ prev · 1/2 grade
Daily Chart Audit
A real-style misleading chart every day. Spot the trick.

Teacher mode

Lesson outline, quick-reference card, and a printable worksheet with answer key.

Lesson outline (45 min)

  • 5 min · Hook — Project two versions of the same bar chart side-by-side (one with truncated axis, one zeroed). Ask which company is "winning." Reveal they're the same data.
  • 10 min · Concept — Axes, scale, sample, window, comparison. Tufte's "data-ink ratio" and the difference between chartjunk and clarity.
  • 10 min · Demonstrations — Run each lab above on the projector: truncation slider, cherry-pick window, per-capita toggle, spurious correlation.
  • 15 min · Practice — Students bring printed news charts and audit them with the five-step checklist. Defend the audit to a partner.
  • 5 min · Wrap — Each student creates a single chart correctly displaying their own attendance / grade / hour data. Honest version only.

Quick reference — the five-step audit

Axes
zero, log, dual
Where does it start, and is the scaling honest for the comparison?
Scale
total vs. rate
Per-capita changes most rankings; raw totals usually flatter big populations.
Window
start, end
Why this range? Show before and after when you can.
Sample
N, selection
Who's included, who's excluded, and was it random or volunteer?
Comparison
vs. baseline
Compared to what? A number alone is rarely informative.
Causation
correlation, confound
A line going up next to another line going up is not a causal arrow.

Worksheet