You look at a log axis running from 1 to 100, find the point halfway along, and read off 50 — because halfway between 1 and 100 is obviously (1 + 100) / 2 ≈ 50. Then you notice the tick sitting right there says 10, and it feels like the chart is lying. The move is tempting for a good reason: averaging the two ends is exactly how you read every linear axis you've ever met. It just happens to be the wrong rule here.
The tick says 10 because, on a log axis, equal distances mean equal ratios, not equal differences. The midpoint between two values isn't their average — it's their geometric mean, the square root of their product. Between 1 and 100 that's sqrt(1 × 100) = 10. The number 10 is to 1 what 100 is to 10: a factor of ten both ways. That symmetry of multiplication is what the axis is measuring, and 50 has no such symmetry (50 is 50× the start but only 2× the end).
If you only remember one rule: to read any point on a log axis, work with the exponent. Position is proportional to the log of the value, not the value itself.
Why 50 feels right
Every other graph you've read is linear. On a linear axis the gridlines are evenly spaced and they count up by a fixed amount — 10, 20, 30, 40 — so the middle gridline genuinely is the average of the two ends. Your eye learned that rule years before you ever saw a log plot, and it learned it so well that it fires automatically.
A log axis is visually a trap because it also has evenly spaced gridlines and a clean straight baseline. Nothing about the spacing screams "the rule changed." So the eye does what it always does: it sees a point halfway along and reports the halfway value. Halfway between 1 and 100, linearly, is about 50. The axis quietly disagrees.
The thing that actually changed is invisible: on a linear axis equal steps in distance add a constant; on a log axis they multiply by a constant. The gridlines look the same. What they encode does not.
The decade, fully worked
Take one decade — the span from 1 to 100 — and find the true midpoint.
The linear guess: (1 + 100) / 2 = 50.5. Wrong, because that's the arithmetic mean, and a log axis doesn't average.
The correct midpoint is the geometric mean: sqrt(1 × 100) = sqrt(100) = 10.
Here's why, in exponents. On a base-10 log axis, position is set by log10 of the value:
- log10(1) = 0
- log10(100) = 2
The visual midpoint is halfway between those positions: (0 + 2) / 2 = 1. Now undo the log — 10^1 = 10. The value sitting at the middle of the axis is 10, exactly as the tick says.
Check the quarter marks the same way, since they trip people up just as badly:
- One quarter along: log = 0.5, so the value is 10^0.5 = 3.16
- Three quarters along: log = 1.5, so the value is 10^1.5 = 31.6
Read the decade across and the ticks fall at:
1, 3.16, 10, 31.6, 100
Those five points are evenly spaced on the page and multiplicatively spaced in value — each is about 3.16× the one before (and 3.16 is sqrt(10), which is the whole pattern in one number). Compare the false linear reading you'd get by interpolating the pixels directly:
1, 25, 50, 75, 100
The linear read is off by more than a factor of two in the middle and gets worse as you go. At the visual midpoint it reports 50 where the truth is 10 — a fivefold error from one wrong assumption about what "halfway" means.
Where this bites in real life
This isn't a quirk of homework axes. Most of the scales you meet outside the classroom are logarithmic precisely because they span many orders of magnitude, and every one of them punishes a linear read.
- Bode plots. The frequency axis is log. The "middle" between 10 Hz and 1000 Hz is not 505 Hz; it's sqrt(10 × 1000) = 100 Hz. Misjudge it and you misplace a corner frequency by a decade.
- pH. A pH of 5 is not "a bit more acidic" than pH 7 — pH is -log10 of hydrogen-ion concentration, so each unit is a 10× change. pH 5 is 100× more acidic than pH 7.
- Richter and decibels. A magnitude-6 quake releases roughly 32× the energy of a magnitude-5. +20 dB is 100× the power, not double. Same rule: the number you read is an exponent in disguise — and decibels add a second twist, where the same +3 dB means a doubling for power but the +6 dB figure shows up for voltage.
The documentation behind plotting tools (GraphPad's Prism guide, Datawrapper's writeups on log scales, the Wikipedia article on logarithmic scale) keeps re-explaining this for the same reason: people interpolate linearly across a decade and report values that are wildly, confidently wrong. The confidence is the dangerous part, because the chart looks readable, so nobody double-checks.
How to read any log axis without getting fooled
- Find the powers of ten first. Locate the labeled decade boundaries (1, 10, 100, …). Those are your anchors.
- Convert your point to "fraction of the way through the decade." If a point sits halfway between 10 and 100, it's at log-position 1.5.
- Raise 10 to that position. 10^1.5 = 31.6. That's your value.
- Sanity-check with ratios, not differences. Ask "how many times bigger," never "how much bigger." If two points are the same distance apart on the page, they're in the same ratio — full stop. The same "which intuition am I running" catch is what saves you when a Bode plot tempts you into reading off a wrong answer from where the poles sit on the imaginary axis.
The single sentence to carry: on a log axis, the middle is the geometric mean, because the axis measures multiplication.
Check yourself
A log10 axis runs from 100 to 100,000. What value sits at the exact visual midpoint?
A) 50,050 — the average of the two ends.
B) 50,000 — round number near the middle.
C) 10,000 — one of the labeled gridlines.
D) About 3,162.
Correct answer: D.
Work in exponents. log10(100) = 2 and log10(100,000) = 5. The midpoint position is (2 + 5) / 2 = 3.5, and 10^3.5 = 3,162 (which is sqrt(100 × 100,000) = sqrt(10,000,000), the geometric mean).
A and B both apply the arithmetic-mean rule from linear axes — the exact habit that breaks here. C is tempting because 10,000 is a labeled tick and "feels central," but it sits at log-position 4, which is three-quarters of the way along, not halfway.
Close the gap
The fix here wasn't memorizing "use the geometric mean." It was noticing the moment your eye switches into linear mode and stopping to ask whether the axis measures differences or ratios. That catch — recognizing which intuition is firing and whether it fits the situation — is the part that's hard to install from a single explanation, and it's exactly what tracking your own missteps across problems is for.