The Field Guide · No. 51
Reading a log-scale chart: equal steps are not equal amounts
A curve that flattens on a log chart can still track a count that keeps rising.
Updated
Take two hypothetical countries that each start with 1,000,000 people. Country A adds 100,000 people a year. Country B grows by 10 percent a year. On an ordinary arithmetic chart, Country A's line is straight and Country B's curves upward. On a log chart the picture flips: Country B becomes the straight line, and Country A's curve flattens, even though A is still adding the same 100,000 people every year.
On a log axis, spacing follows multiples, not amounts. The distance from 1 to 10 is the same as the distance from 10 to 100, and the distance from 100 to 1,000 is the same again. Each of those steps is a tenfold increase, so the step from 100 to 1,000 looks exactly as long as the step from 1 to 10, even though the first adds 900 (1,000 minus 100) and the second adds 9 (10 minus 1). Inside each cycle, the tick marks crowd together as the values rise.
Log scales do two jobs well. The CDC's epidemiology course says that when the range of the data is so large that an ordinary graph does not work well, a logarithmic axis is the alternative. It also turns a steady percentage change into a straight line. An outbreak that grows by the same percentage each week draws a straight line, while the same growth on an arithmetic chart draws a curve that steepens. The course's own example plots age-adjusted death rates for five of the 15 leading causes of death in the United States from 1958 to 2002, and it uses the log scale to compare how quickly each cause's death rate changed.
Log charts are harder to read, and in one study the scale changed how worried people were. In a 2020 experiment, Romano and colleagues randomly showed roughly 2,000 US residents the same COVID-19 death counts, on either a linear graph or a logarithmic one. Those who saw the linear graph answered the questions about it correctly more often and made better predictions. The authors offer a conjecture: a flat stretch on a log curve can look like a plateau, which might make people feel the end is near. They recommend a linear graph as the default. The UK Office for National Statistics says to avoid log scales except where there is no other clear way to show the data.
Read the axis before the slope
Start with the labels. The CDC's example axis has four cycles, each a multiple of ten, such as 0.1 to 1 and then 1 to 10.
Then read the slope as a rate. A straight line on a log chart means a constant percentage change from a constant rate. The slope shows how fast something grows in percentage terms, not how many people it adds.
A flattening curve means the percentage growth is slowing. It does not mean the count is falling. In the CDC example, Country A still gains 100,000 people a year, but each year's gain is a smaller share of a larger population, so the curve flattens.
What the 2020 experiment found, and what it cannot show
Half of the respondents saw a linear graph of US COVID-19 deaths and half saw a log graph of the same data, which came from worldometers.info and ran from 15 February to 18 April. Respondents answered the policy and worry questions first, then the questions about the graph, one of which used a hypothetical infection. Last, they predicted deaths for 25 April, one week after the experiment launched.
The linear group reported more worry about the health crisis, though not about the economic one. It wanted nonessential businesses kept closed for longer, and it supported closing them less strongly. The authors say they cannot know the mechanism behind these results.
The study has limits. The graphs showed one data series, and the sample was US residents recruited through Cloud Research. The policy results are the least steady part. In the first model for each policy question, the difference is not statistically significant: the support result has a coefficient of 0.0406 (p = 0.621), and the closure-length result is 2.295 more days (p = 0.464). Later models show larger gaps, including 17.38 more days (p = 0.014). The paper also cites a Canadian study that found the graph scale had no effect on respondents.
What to remember
- The gap from 1 to 10 matches the gap from 100 to 1,000, because each step is a tenfold increase.
- A straight line means a steady percentage change. A flattening curve means the percentage growth is slowing, not that the count is falling.
- In a 2020 experiment, people who saw COVID-19 death counts on a log graph answered questions about them less accurately and made less accurate predictions than people who saw a linear graph.
- The UK Office for National Statistics says to avoid log scales unless there is no other clear way to show the data.
From the record
In particular, we find that when people are exposed to a logarithmic scale they have a less accurate understanding of how the pandemic unfolded until now, make less accurate predictions on its future, and have different attitudes and policy preferences than when they are exposed to a linear scale.
Asked often
What does a straight line on a log chart mean?
A constant percentage change from a constant rate. The slope is a rate of growth, so a straight line keeps the same percentage pace while the number itself climbs.
If a line flattens on a log chart, is the count falling?
Not necessarily. Flattening means the percentage growth is slowing. The number can still rise every year, as long as each year's gain is a smaller percentage of a larger total.
Why do outbreak charts often use a log scale?
A log scale can show values from 1 to 1,000 on one chart, and steady percentage growth shows up as a straight line. The paper notes that the Financial Times, The Guardian and the New York Times used log scales for COVID-19 death graphs in 2020, and that the New York Times described the scale as a better way to show exponential growth.
Should I trust a log chart?
Read the axis labels before you read the slope. Both the paper's authors and the UK Office for National Statistics prefer linear graphs as the default, because they are easier for readers to understand. A log chart still earns its place when the range is too large for a linear one to show clearly.
Read the news better, every morning.
We send you the day's real progress in science, medicine and beyond, told plainly and sourced. Free, forever.
Free for good. One click to leave.