Same data, different truths: perspective and framing in charts
TL;DR
Difficulty: Intermediate
The same numbers can honestly support opposite conclusions, because every chart needs a framing: a unit, a baseline and something to divide by.
- Show absolute change next to relative change, or a small base looks huge.
- Rates change in percentage points; pick one unit for every rate.
- Totals grow with the crowd, so add a per customer view.
- Index to a meaningful base, and adjust money across years for inflation.
- Show the parts next to the whole (Simpson's paradox) and ask who is missing (survivorship).
- Choose the framing on purpose, use it for good and bad news alike, and write it under the chart.
What to read first
The marketing lead says the quarter was great. The finance lead says it was worrying. Both used the same spreadsheet, and both are right. This chapter takes one invented quarter at an outdoor gear brand in Limerick and shows the eight framing choices that turn the same data into opposite stories, then builds the one board slide that shows both.
Two people, one spreadsheet, opposite conclusions
It happens in almost every quarterly meeting. The marketing lead puts up a slide that says the business is flying. The finance lead puts up a slide that says it is slipping. Both took their numbers from the same spreadsheet, and neither made anything up. So who is right?
Usually both are, and that is the uncomfortable part. A number on its own says very little. It only means something once you decide what to compare it with, which unit to show it in, and whose point of view to take. Those decisions are called the framing. The previous chapter, how charts lie and how to catch them, was about charts that distort the data. This one is about charts that show the data perfectly and still steer you, because the framing was chosen for you.
Framing is not a small effect. In a famous experiment, Amos Tversky and Daniel Kahneman gave people the same choice twice, once described in terms of lives saved and once in terms of lives lost. The outcomes were identical. The preferences flipped.1 If the wording of a question can do that, the choice of axis, unit and baseline on a chart can do it too.
This article walks through the framing choices that matter most in a business: absolute against relative change, percentage change against percentage points, totals against per customer figures, indexing, money with and without inflation, Simpson's paradox, survivorship and the choice of baseline. Each one gets its own example. At the end, the whole thing comes together in one honest board slide.
Example: the quarter that started an argument
Shannon Outfitters is an outdoor gear brand in Limerick. It sells through a web shop, a small social media shop, and one physical shop in the city. Its analyst, Sam, prepares the board report. The quarter in question is July to September 2026, compared with the same quarter a year earlier. All numbers in this article are invented for the example.
| Channel | Q3 2025 | Q3 2026 |
|---|---|---|
| Web shop | €176,000 | €224,000 |
| Social shop | €4,000 | €10,000 |
| Limerick shop | €220,000 | €196,000 |
| Total revenue | €400,000 | €430,000 |
| Customers | 5,000 | 6,200 |
| Web conversion rate | 2.0% | 2.4% |
| Return rate | 8% | 10% |
Gráinne, the marketing lead, and Cian, the finance lead, each have ten minutes with Aoife, the owner. Every claim they make in the sections below is true. Watch how far apart they end up.
Relative change makes small things look big
A change can be told in two ways. The absolute change is the plain difference: the thing went up by €6,000. The relative change is that difference measured against where you started: the thing went up by 150%. Both are correct. They answer different questions. The absolute change tells you how much it matters to the business. The relative change tells you how fast it moved.
Here is the absolute change, in the same unit as the data (euro, customers, orders). is the relative change, as a percentage. is the value now and is the value you compare against. Because sits underneath the fraction, a small starting value makes any change look enormous. The full explanation is in Foundations: percentages.
That denominator is where the trouble lives. Gerd Gigerenzer and colleagues describe what happened in the UK in 1995, when a warning said that newer contraceptive pills doubled the risk of a dangerous blood clot, an increase of 100%. In the studies behind the warning, about one in 7,000 women on the older pills had a clot, against about two in 7,000 on the newer ones. The relative increase really was 100%. The absolute increase was one woman in 7,000. Many women stopped taking the pill, and the authors cite an estimated 13,000 additional abortions in England and Wales the following year.2 The same paper notes that a screening programme described as reducing deaths by 25% meant, in absolute terms, one fewer death in every 1,000 women screened.2 Relative numbers make better headlines. Absolute numbers make better decisions.
Example: Gráinne's social shop and Cian's Limerick shop
Gráinne opens with the social shop. Its revenue went from €4,000 to €10,000. Using the formula: . Her slide says "Social sales up 150%".
Cian opens with the Limerick shop. Its revenue went from €220,000 to €196,000. The relative change is , downwards. On a chart of percentages, a fall of 10.9% next to a rise of 150% looks like a rounding error.
Now the absolute changes. The social shop gained . The Limerick shop lost . The web shop gained , which is a relative change of about 27.3%. In euro, the social shop is the smallest story of the three, and the Limerick shop lost four times what the social shop gained.
The honest version is not the second chart on its own. It is both, or a single chart in euro with the percentage written on each bar. Aoife needs the euro to judge the size and the percentage to judge the speed.
Percentage change and percentage points are different units
When the number you are changing is itself a percentage, such as a conversion rate or a return rate, there are two ways to describe the change, and they can differ by a factor of ten or more. A percentage point is the plain difference between two percentages. A percentage change is that difference measured against the starting percentage.
is the change in percentage points. and are the two rates, both already written as percentages. The relative change of a rate uses the formula from the previous section with in place of . See Foundations: percentages for both, and Foundations: ratios and rates for what a conversion rate is.
Neither unit is wrong. The problem starts when someone picks the unit that sounds best, rate by rate.
Example: the same two rates, described by both leads
The web conversion rate went from 2.0% to 2.4%. In points: percentage points. As a percentage change: .
The return rate went from 8% to 10%. In points: percentage points. As a percentage change: .
Now watch what each lead says.
| Claim | Who says it | Unit chosen | What it hides |
|---|---|---|---|
| "Conversion up 20%" | Gráinne | percentage change | it is 0.4 points |
| "Returns up only 2 points" | Gráinne | points | it is a rise of a quarter |
| "Conversion up a mere 0.4 points" | Cian | points | it is a fifth more orders per visit |
| "Returns up 25%" | Cian | percentage change | it is 2 in every 100 orders |
Each line is true. Gráinne uses percentage change for the good news and points for the bad news. Cian does the opposite. The fix is boring and effective: pick one unit for all rates in a report, and say which one. I would choose percentage points for rates, with the percentage change in brackets when it helps: "conversion rose 0.4 points (a fifth more orders per visit), returns rose 2 points (a quarter more returns per order)".
Totals hide how many people are behind them
A total grows when the business grows, even if every single customer is spending less. A figure per customer (or per order, per visit, per head of population) takes the size of the crowd out. Neither is the truth on its own. The total tells you whether the business got bigger. The figure per customer tells you whether it got better at what it does with each person.
is the value per customer. is the total, for example revenue in the period. is the number of customers in the same period. It is a plain mean, so everything in Foundations: the mean applies, including the warning that one very large customer can pull it up.
The same idea is behind every official figure quoted "per head". A country's output, a region's crime figures or a city's spending only compare fairly once you divide by the number of people, because a bigger place will almost always have a bigger total.
Example: more customers, less from each
Gráinne's slide says "24% more customers". She is right: .
Cian divides revenue by customers. In Q3 2025, . In Q3 2026, . The change is , downwards. His slide says "Each customer spent 13% less".
Both slides are true, and they are true together: Shannon Outfitters found many more customers, and on average each one bought less. That combination is the real finding, and it raises the question Aoife actually needs answered: are the new customers smaller, or are the old ones shrinking? Neither total answers it. A split by new and returning customers would.
Indexing puts different scales on one chart
Sometimes you want to compare how several things grew, but they live on wildly different scales. Web visits are in the hundreds of thousands, orders in the thousands, revenue in euro. Draw them on one axis and the biggest series flattens the others into a line along the bottom. Draw them on two axes and you invite the dual axis tricks from the previous chapter.
Indexing solves this. You pick one period as the base, set every series to 100 at that point, and show everything else relative to it.
is the index value in period . is the actual value in period , and is the value in the base period. An index of 120 means 20% above the base, and 90 means 10% below. Official inflation figures are published exactly this way. More in Foundations: indexing and base rates.
Indexing has one framing choice built in: the base. Pick a weak period as the base and everything after it looks like a recovery. Pick a strong one and everything looks like a decline.
Example: visits, orders and revenue over five quarters
Sam has five quarters of data for Shannon Outfitters (invented):
| Quarter | Web visits | Orders | Revenue |
|---|---|---|---|
| Q3 2025 | 300,000 | 5,600 | €400,000 |
| Q4 2025 | 360,000 | 7,000 | €520,000 |
| Q1 2026 | 270,000 | 4,800 | €350,000 |
| Q2 2026 | 330,000 | 6,000 | €410,000 |
| Q3 2026 | 380,000 | 6,900 | €430,000 |
On one axis, visits run from 270,000 to 380,000 and orders from 4,800 to 7,000. The orders line hugs the bottom of the chart and looks flat.
Sam indexes each series to Q3 2025. For visits in Q3 2026: . For orders: . For revenue: . The other quarters work the same way.
Now the story is visible. Visits and orders grew by about a quarter, revenue by 7.5%. The gap between the orders line and the revenue line is the fall in revenue per order that Cian found.
Then the base trap. Suppose Gráinne indexes to Q1 2026, the quiet winter quarter, instead. Revenue in Q3 2026 becomes , and her slide can say "revenue up 23% since the start of the year". True, and almost entirely the normal summer season. The honest choice is a base that means something for the question: the same quarter last year for a seasonal business, or the average of a full year.
Money now and money then are not the same money
A euro in 2025 buys less than a euro in 2021. Revenue in euro as it was at the time is called nominal. Revenue converted into the euro of one chosen year, so that prices are held still, is called real. A nominal chart of several years of revenue mixes two things: how much more the business sold, and how much prices went up. Only the first one is growth.
To convert, you use a price index such as the Consumer Price Index (CPI), which the Central Statistics Office (CSO) publishes every month.3
is revenue in year , expressed in the money of the base year. is the nominal revenue in year . is the price index in year and is the price index in the base year. If prices rose, is bigger than , so the fraction is below 1 and real revenue is smaller than nominal. The index logic is in Foundations: indexing and base rates.
For a growth rate over a period, there is a shortcut:
is the real growth rate, is the nominal growth rate and is the inflation rate over the same period, all as decimals (so 20% is 0.20). Subtracting from is a rough approximation that works only when both are small. See Foundations: growth rates and compounding.
Example: five years of Shannon Outfitters in real euro
Shannon Outfitters' annual revenue (invented) went from €1.60 million in 2021 to €1.88 million in 2025. The nominal growth is , so 17.5%. A good few owners would call that a grand five years.
The CSO's CPI (base December 2023 = 100) averaged 86.0 in 2021 and 102.9 in 2025.3 Prices rose by , so about 19.7%.
In 2021 euro, 2025 revenue is million. The real growth is , a real fall of about 1.8%. 17.5% more euro bought about 2% less.
The same sum for each year, with the yearly CPI averages of 92.7 for 2022, 98.6 for 2023 and 100.7 for 2024,3 gives the real line in the chart.
Over a single year the gap is smaller, but it is not zero. Irish consumer prices in September 2026 were 4.1% higher than a year earlier.4 That figure comes back in the board slide at the end.
Simpson's paradox: every part improves and the whole gets worse
Sometimes a figure for the whole business moves one way while every part of it moves the other. This is Simpson's paradox, and it happens whenever the mix between the parts changes.5 The best known case is graduate admissions at the University of California, Berkeley, in 1973. Across the university, men were admitted at a higher rate than women. Department by department, Peter Bickel and his colleagues found no bias against women. Women had applied more often to the departments that turned most people away.6
The arithmetic behind it is a weighted mean. A total rate is the rates of the parts, each weighted by its share of the whole.
is the overall figure, here the gross margin of the whole business as a percentage. runs over the channels. is channel 's share of revenue (the shares add up to 1), and is the margin inside channel . If the weights shift towards a low margin channel, can fall even when every rises. The weighted mean is in Foundations: the mean, and the article on why averages mislead goes much deeper into the paradox.
Example: both channels earn more, the business earns less per euro
Gross margin is the share of revenue left after paying for the goods sold. The physical shop sells at full price with no delivery cost. Online sells with discounts and free delivery. Both improved their margin by a point this year:
| Channel | Revenue Q3 2025 | Margin Q3 2025 | Revenue Q3 2026 | Margin Q3 2026 |
|---|---|---|---|---|
| Limerick shop | €220,000 | 52% | €196,000 | 53% |
| Online (web and social) | €180,000 | 28% | €234,000 | 29% |
In Q3 2025 the shop had of revenue and online . So .
In Q3 2026 the shop had and online . So .
Cian can show "margin down 1.3 points". Gráinne can show "margin up in every channel". Both are true. The honest chart shows the parts and the whole together, and the honest sentence names the cause: the mix moved online. One more framing hides in the same table. Gross profit in pounds went from €164,800 to about €171,700, up by roughly €6,900. A lower margin on a bigger total can still leave more money in the bank.
Survivorship bias: the data you see is the data that stayed
Every dataset is made of the things that were around to be counted. Customers who left do not fill in surveys. Products that were dropped are not in this year's sales report. Shops that closed are not in the industry average. When the missing ones differ from the ones that stayed, every chart built on the survivors is biased. This is called survivorship bias.
The classic story comes from the Second World War. The statistician Abraham Wald worked on how to protect aircraft, using the damage on planes that came back. The planes that were hit in the most vulnerable places mostly did not come back, so the damage you could see on returning planes pointed away from where armour was needed most. Marc Mangel and Francisco Samaniego later reconstructed his method from his wartime memoranda.7
Example: the loyalty scheme that everyone loves
Gráinne's next slide says "85% of members rate the loyalty scheme good or very good". Sam checks where the number comes from.
| Group | Members |
|---|---|
| Joined the scheme in the year | 2,000 |
| Left the scheme again | 800 |
| Still members, surveyed | 1,200 |
| Answered the survey | 600 |
| Answered good or very good | 510 |
The 85% is . It is the share of happy people among members who stayed and chose to answer. The 800 who left were never asked, and they are exactly the people most likely to say no. Measured against everyone who joined, the share with a confirmed good opinion is . That is not the true approval rate either, since 1,400 people gave no answer. The honest statement is a range with its denominator: "Of 2,000 people who joined, 510 told us they rate it good, 90 told us they do not, 800 left without being asked, and 600 members did not reply." A short exit question for people who leave would turn the biggest unknown into data.
The baseline decides the verdict
"Up" and "down" only mean something against a baseline, and a business has at least three: the same period last year, the target someone set, and the plan or budget the board approved. A single actual figure can beat one and miss another. Choosing which one to show is choosing the verdict.
Each baseline answers a different question. Against last year: are we growing? Against the target: did we reach what we aimed for? Against the plan: is the money we budgeted on the basis of this figure still safe? A board needs all three, in a fixed order, every quarter. If the comparison changes from report to report, a reader should ask why.
The size of a gap against a baseline is the same relative change as before, with the baseline as (see Foundations: percentages).
Example: €430,000 is a win, a miss and roughly on plan
Shannon Outfitters' Q3 2026 revenue was €430,000.
| Baseline | Baseline value | Gap in euro | Gap as a percentage |
|---|---|---|---|
| Same quarter last year | €400,000 | up €30,000 | up |
| Target set in January | €460,000 | down €30,000 | down |
| Budget plan | €420,000 | up €10,000 | up |
Gráinne shows the first row. Cian shows the second. The plan row is the one nobody chose, and it is the one Aoife signed off. All three go on the slide, always in the same order.
Framing on purpose, and saying so
Every chart has a framing, because every chart needs a unit, a baseline and a choice of what to divide by. You cannot avoid choosing. What you can avoid is choosing silently, or choosing a different framing for each number depending on which one flatters it.
Across the water, the UK's Office for Statistics Regulation calls the standard it expects from government "intelligent transparency". Its guidance asks for figures to come with their source, their definitions, their strengths and limitations, and the context needed to read them, and it warns against selective use of data without that context.8 That is a public sector standard for another jurisdiction, but it is a good one for any board report in Limerick or anywhere else.
In practice it comes down to a few habits:
- Say the framing in the chart. One line under the title: what is compared with what, in which unit, adjusted or not.
- Use the same framing for good and bad news. Points for every rate, or percentage change for every rate. Never one for the good and the other for the bad.
- Show the size and the speed. Absolute change and relative change together, at least for the headline numbers.
- Show the parts next to the whole whenever the mix can move.
- Ask who is missing before trusting a rate.
This is also why I start every analytics project from the decision the numbers should support rather than from a dashboard (the article start with the decision, not the dashboard explains the method). Once you know the decision, the right framing usually follows: a decision about stock needs units, a decision about marketing spend needs euro per customer, and a decision about prices over several years needs real money.
The flow chart below is the checklist Sam runs before any number goes onto a slide.
Example: the framing note under one chart
Sam's chart of channel revenue gets this line under its title: "Change against July to September 2025, in euro, not adjusted for inflation. Percentage change shown on each bar." It takes ten seconds to write. It removes the argument about which framing was used, so the meeting can argue about the business instead.
Worked example: one quarter, two stories, one honest slide
Here is the full quarter, put together. First the two stories as they arrived.
Gráinne's four charts:
- Social shop revenue up 150%.
- Web conversion up 20%.
- Customers up 24%.
- Margin up in every channel, and 85% of loyalty members happy.
Cian's four charts:
- Limerick shop revenue down €24,000.
- Returns up 25%.
- Revenue per customer down 13%.
- Business margin down 1.3 points, and revenue €30,000 below target.
Every one of these eight claims is correct. Read Gráinne's alone and the business is thriving. Read Cian's alone and it is in trouble. Aoife cannot decide anything from either.
Sam builds one slide instead, in four steps.
Step 1, the headline in euro and as a percentage. Revenue rose from €400,000 to €430,000, which is €30,000 or 7.5%.
Step 2, the real figure. Prices were 4.1% higher than a year earlier,4 so with the shortcut formula, real growth is , about 3.3%. Strictly the quarter's own inflation rate should be used, and a monthly figure is close enough for a board slide as long as the slide says so.
Step 3, where the change came from, as a bridge. A waterfall chart starts at last year's total, adds each channel's change, and lands on this year's total. Both totals are drawn from zero, so the size of every step is honest.
Step 4, the context everyone needs, one framing for each kind of number:
| Measure | Q3 2025 | Q3 2026 | Change |
|---|---|---|---|
| Revenue | €400,000 | €430,000 | up €30,000 (7.5%, about 3.3% real) |
| Against target (€460,000) | down €30,000 (6.5%) | ||
| Against plan (€420,000) | up €10,000 (2.4%) | ||
| Customers | 5,000 | 6,200 | up 1,200 (24%) |
| Revenue per customer | €80.00 | €69.35 | down €10.65 (13.3%) |
| Web conversion rate | 2.0% | 2.4% | up 0.4 points |
| Return rate | 8% | 10% | up 2 points |
| Gross margin | 41.2% | 39.9% | down 1.3 points, both channels up 1 point |
| Gross profit | €164,800 | €171,740 | up €6,940 |
The title of the slide states the finding, as the first chapter of this series recommends: "More customers and more profit, but each customer spends less and the Limerick shop is shrinking." Under it, the framing line: "All changes against July to September 2025. Rates in percentage points. Real growth uses the CPI rate for September 2026."
Notice what happened to the argument. Gráinne's best point (more customers, online growing) is on the slide. Cian's best point (smaller baskets, a shrinking shop, a missed target) is on the slide. The loyalty score is left off until the leavers have been asked. Aoife now has two real questions to decide on instead of two stories to choose between: why are new customers buying less, and what is the plan for the Limerick shop?
The point in one paragraph
The same data can support opposite conclusions, and usually neither side is lying. They are choosing different framings: relative or absolute, percentage change or points, total or per customer, a convenient base, nominal or real money, the whole or its parts, the survivors or everyone, last year or the target. You cannot present a number without a framing. You can present it with a framing that you chose on purpose, applied the same way to good news and bad, and written under the chart. When two framings tell different stories, the honest slide shows both and lets the decision makers see the tension, because the tension is usually where the decision is.