Log scales suit data spanning orders of magnitude

What it is

When data spans orders of magnitude (one series at 10, another at 1,000, a third at 100,000), the linear axis is a disaster: small values crush against the floor while the largest owns the page — 99% of the data squeezed into 1% of the space. The log scale turns "equal ratio" into "equal distance": 10-to-100 and 1,000-to-10,000 take the same length, each decade granted its share. It does more than "enlarge small values" — it changes the semantics of comparison: distance along the axis maps directly onto multiplicative relations, and doubling, exponential growth, and cross-magnitude distribution shapes (log-normal) return as linearly readable structure.

Why it happens

The linear axis's failure under cross-magnitude data is geometric necessity: the maximum fixes the range, and every other value's resolvable height collapses proportionally. The log transform maps multiplicative structure onto additive: geometric sequences become equally spaced on a log axis, exponential growth draws as a straight line, doubling reads as equal segments — precisely the questions cross-magnitude scenes most need answered. Log-normal distributions (income, durations, file sizes — products of multiplicative processes) go from a linear-axis skews-tail back to a symmetric bell, and the shape toolkit (mean ± SD, box plots) becomes usable again.

Where it stops holding

The log axis holds only for positive values: zero and negatives have no logarithm; data containing zeros (zero counts are common) either shifts by +1 (declared) or abandons the log. The comparison semantics change: axis distance = ratio, absolute difference no longer readable — "three more units" has no geometric expression on this axis, and difference-asking tasks (balance comparisons) should not go logarithmic. Bars cannot ride a log axis (no zero exists, so length encoding has no origin — see the broken-axis and zero-baseline groups). Reader literacy is a real threshold, treated in the next leaf.

Applying it

• Quantify the trigger: a max/min ratio past 100 (about two decades) puts the log axis under evaluation; multiplicative metrics (growth rates, exchange rates, fold-changes) go first.
• Label ticks with raw values (1, 10, 100, 1000), minor ticks at the 2× and 5× positions; mark key ratios (2×, 10×) with explicit reference lines.
• Verification: ask readers "by what factor does the largest group exceed the smallest?"; directly readable on the log axis but not on the linear one is exactly the gain from switching.

Related

• Same group: U3.03.2 Untrained readers misread logarithmic axes · U3.03.3 The scale type must be prominently labelled
• Nearby: U3.02.5 Facets or log scales usually replace the broken axis · U2.08.1 Histogram bin width changes the conclusion
• Search terms: logarithmic scale · orders of magnitude · log-normal · multiplicative process