How to Read a Results Chart Before You Read the Caption
You are three slides into a case study and there it is: a line climbing off the top right corner of the frame, under a confident caption. Does the steepness belong to the results or to the drawing? You want to know before the deck moves on.
The claim is in the drawing, not just the numbers
A chart is not a neutral rendering of data. It is an argument made out of data, and whoever drew it made choices the numbers did not dictate. The same figures can look like a cliff or a gentle slope depending on three decisions: where the y-axis starts, whether the line is cumulative or periodic, and whether a second axis is doing work.
You are staring at a steep line, you have no reason to think anyone lied, and you still cannot say where the steepness lives.
The ten second axis check
Take a hypothetical metric, reported before a campaign and after it, moving from 41 to 46. Invented numbers, illustrating a mechanism, not a result anyone reported. Plotted from 0 to 50, that line rises by a tenth of the chart’s height. Plotted from 40 to 50, the same five points fill half the height and climb at close to forty five degrees. The bottom four fifths of the range is gone, so the part that moved got the whole canvas.
Wikipedia’s entry on misleading graphs documents this construction and notes that studies found people overestimate the difference even after being told the axis was truncated. Knowing about it does not immunise you, so make the check mechanical. Read the first and last value printed on the y-axis before you look at the slope. If the bottom one is not zero, ask whether this would still look dramatic from zero.
Cumulative totals only go up
A running total adds each period to everything before it, so as long as the periods are non-negative, as they are for followers, leads, or revenue, it can rise or hold flat and it cannot fall. That is a property of the construction, not an observation about any campaign, so a cumulative chart can survive a bad quarter without ever showing you one.
Read the axis label and chart title for words like total, cumulative, or to date, then scan for a single downward segment. If there is not one, you are looking at a running total, and a flat stretch is then the strongest signal on the chart: those periods contributed close to nothing.
Such a chart answers how much so far, not whether this is getting better or worse. Only the periodic version, each day or week standing on its own, answers that. If the caption talks about acceleration, the chart you want is the one you were not shown.
When a second axis does the persuading
Picture a hypothetical chart from an equally hypothetical campaign. One line is follower count, 12,000 to 14,000 on the left axis. The other is monthly revenue, 80,000 to 95,000 on the right. Both climb and cross near the middle: followers went up, revenue followed. Invented figures again, illustrating a construction, not a reported result.
That impression is cheap, because nothing requires the two axes to share a scale, a starting value, or any relationship at all. On shared, zero-based axes, the same two series might not look related at all. The Wikipedia entry lists this among its documented distortions: giving one line its own y-axis makes comparison between the series misleading.
So the check is not visual. Ask whether the case study states a mechanism: why followers should move revenue, by what route, with what lag. Shared position on a chart is a layout decision.
The three checks before you read the caption
- Does the y-axis start at zero? If not, would the same data still look dramatic from zero?
- Is the line cumulative or periodic? Look for total, cumulative, or to date in the label, and for the absence of any downward segment. If cumulative, ask what the periodic version would show.
- Is there a second y-axis? If so, look for a stated reason the two series should move together. If the only connection offered is the shared chart, treat the coupling as decoration.
The caption’s job versus the chart’s job
A caption can assert a conclusion the chart does not support, and it usually gets away with it. Reverse the order: run the three checks, decide what the chart shows on its own terms, then read what you are told you just saw.
A chart is a claim like any other in the document, and it earns the same treatment as the headline number. If you work through how to read a case study without being misled, the chart is one more item on that list. And when a second axis pairs a flattering series with another, remember why the best metric on a dashboard is rarely the real story.
Try it on a published one
Browse the case study library and run the three checks on whatever chart you land on. That is how they become automatic. If you have published work worth a close reading, you can submit your own.
FAQ
Why does a chart with a truncated y-axis look so much steeper than the same data plotted from zero?
Truncating the axis shrinks the visible range, so the same absolute change fills more vertical space. Read the first and last value printed on the axis before judging the slope, because the studies noted in Wikipedia’s entry on misleading graphs found people still overestimate the difference when told the axis was truncated.
Is a cumulative total chart dishonest if the line never goes down?
No. A running total of non-negative values can only rise or stay flat, so a line that never falls is what the construction produces, not evidence of anything. The real question is what the periodic version would show, each day or week plotted separately.
How do I know if a dual axis chart is misleading me?
There is no requirement that the two y-axes share a scale or a starting point, so two unrelated series can be scaled until they rise together or cross. Look for a stated mechanism connecting the metrics, a reason one should move the other. If the only thing joining them is their shared position on one chart, that is a layout decision, not a finding.
Sources
- Misleading graph, Wikipedia, on truncated graphs, axis changes, and distinct y-axes. Fetched 2026-09-02.
