Honest Visualization

The classic: starting a bar chart or line chart y-axis at something other than zero. This is sometimes defensible — a temperature anomaly chart showing changes from baseline doesn't need to start at absolute zero. A percentage change chart doesn't need to start at 0%.

But when a bar chart showing emissions reductions starts its y-axis at 85%, what looks like bars of very different heights is actually the range 85-100%. The visual impression is that some bars are 3-4× taller than others; the actual data difference is a few percentage points. In 2009, National Review published a climate chart with a y-axis from 0°F to 25°F that made the global temperature record look flat — the change from 11.5°F to 14.5°F over a century was nearly invisible against a 25-unit axis.

The check: what does this chart look like with the axis starting at zero? If the pattern disappears, ask yourself whether the truncation is justified.

"Emissions fell 30% since 2019" and "emissions are higher than 2000 levels" can both be true — they just use different start years. Choosing a start year that captures a peak (or a trough) to make a trend look better (or worse) than the long-run picture is one of the most common climate data manipulations.

The Skeptical Science "Escalator" graphic is a famous illustration of this: it shows how cherry-picked short-term cooling trends can be used to argue against long-term warming, by overlaying a series of misleading short windows on a clearly rising long-term trend. Each window is accurate. Together they obscure what's actually happening.

The check: show the trend from multiple start years and let your audience see whether the story changes. "Here's what the trend looks like from 1990, 2000, 2010, and 2019." If your argument depends on one specific start year, that's a signal.

Color scale choices dramatically affect the perceived magnitude of change. A diverging red-blue scale for temperature anomalies communicates alarm. The same data on a sequential blue-to-white scale communicates something much calmer. Neither is inherently wrong — the question is whether the visual weight matches what the data actually shows.

Perceptually non-linear scales are a subtler problem. Rainbow color maps (often a default) are not perceptually uniform — the same numerical difference between green and yellow looks smaller than the same difference between yellow and red. For climate temperature maps, this can make warming in some regions look more or less severe than it is. Viridis and other perceptually uniform scales were developed specifically to solve this problem.

The check: does the visual hierarchy match the data hierarchy? Are the places that look the most different actually the most different numerically?

Aggregating across different-sized units without normalization produces apparent comparisons that aren't really comparisons. A pie chart showing China's and Luxembourg's share of global emissions is technically accurate but tells you almost nothing meaningful — the comparison is overwhelmed by population size. Absolute figures and per-capita figures tell different stories; presenting only one without noting the other is selective.

A more subtle version: showing total climate finance flows without distinguishing grants from loans, or public from private. The headline number ($115.9 billion in 2022) obscures that much of it is loans that developing countries repay, and private finance that was "mobilized" rather than committed.