Faragiri

Learn Charts Excel

A hospital ward in Aalst charts its handover times: which chart answers which question, the eight checks before sending, and the two defaults that mislead by themselves.

6Sections
4Min read
4Depth

Wietse charts nursing handover times for a ward in Aalst. Compliance moved from 94.1 to 95.3 per cent over a quarter, and the default chart Excel produced made it look like the ward had transformed itself, because the axis started at 94.

Nobody chose that. Excel chose it, and the chart was not wrong so much as it answered a question nobody had asked.

The chart follows the question

Four questions cover almost everything, and each has one obvious answer.

Visual encoding choice

Four analytical questions and their single honest chart type

Matching chart geometry directly to the underlying question prevents visual distortion and reading fatigue.

  1. Comparison: Horizontal bar chartProvides space for long category labels without tilting text, leveraging human skill at length comparison
  2. Trend over time: Line chartEncodes progression as slope across evenly spaced chronological intervals on the horizontal axis
  3. Composition: Bar chartReplaces confusing pie slices because readers judge linear lengths far more accurately than angles
  4. Relationship: Scatter plotThe most under-used chart type, showing correlation between two independent variables honestly

Selecting the correct chart family answers the user question directly before any custom formatting or axis adjustment is required.

How do these compare: a bar chart, with bars horizontal when the labels are words longer than about two of them, because sideways text is unreadable and everyone tilts it anyway.

How has this moved over time: a line chart, and only a line chart. Time on the horizontal axis, evenly spaced, with gaps left as gaps rather than closed up.

What is this made of: a bar chart again. A pie is readable at two or three slices and is guesswork at six, because people compare angles badly and lengths well.

Is there a relationship between two measurements: a scatter, which is the chart most under-used in offices and the only one that shows two variables against each other honestly.

The two defaults that mislead

A truncated axis on a bar chart. A bar's length is the message, so cutting the axis multiplies the apparent difference: Wietse's 1.2 percentage points became a bar four times the height of its neighbour. A bar chart's axis starts at zero, and if the interesting range is narrow the answer is a line chart or a different measure, not a cut axis.

Axis baseline rules

Zero baseline for bars versus flexible baselines for lines

Cutting a bar chart axis distorts proportions, while forcing a line chart to zero flattens critical trend slopes.

Bar chart axis baselineMust start at zero
Line chart axis baselineStarts where data requires
Visual messageEncoded in physical bar lengthEncoded in trajectory and slope
Effect of truncated axisMultiplies small 1.2 point differences into quadrupled bar heightsReveals subtle rate changes that a zero floor would flatten
Default category orderingSort categories by value to turn walls of bars into clear rankingsSort time points strictly chronologically with explicit interval spacing

A bar's physical length represents magnitude, so cutting its axis misleads; a line represents rate of change, making narrow vertical focusing appropriate.

A line chart may start where it likes, because a line's message is the slope rather than the height, and forcing it to zero flattens everything worth seeing. That difference between the two is the single most useful thing on this page.

The second default is the automatic category order. Excel plots rows in sheet order, which is usually alphabetical or arbitrary. Sorting by value before charting is what turns a wall of bars into a ranking, and it takes four seconds.

Eight checks before it goes out

Does the title say the finding rather than the subject: Handover compliance rose 1.2 points, not Handover compliance.

One ward, one quarter

Eight checks, and the first chart Excel offered failed three of them

The state beside each is how the ward's default chart scored before anybody touched it. Two of the three failures were choices Excel made rather than choices anybody took.

Before it leaves the sheet

  1. NoThe title says the findingHandover compliance rose 1.2 points, not Handover complianceNo
  2. NoThe bar axis starts at zeroIt started at 94, which quadrupled the apparent changeNo
  3. NoCategories sorted by valueExcel plots them in sheet order, which is usually arbitraryNo
  4. YesUnits are on the axisPer cent of handovers, not a bare numberYes
  5. YesSource and period stated on the chartBecause most charts are read in an emailYes
  6. YesNothing on it carries no informationGridlines, a legend for one series, a shadowYes
  7. YesIt survives being printed in greyOne accent colour and everything else grey passes by constructionYes
  8. Not checkedReadable by somebody not in the roomThe real test, and the one the other seven serveNot checked

The eighth check cannot be self-assessed, which is why it is marked unknown: the only way to run it is to send the chart to somebody with no context and ask what it says.

Is the axis honest for the chart type. Are the categories sorted. Are the units on the axis. Is the source and the period stated somewhere on it. Is there anything on it that carries no information: gridlines nobody reads, a legend for a single series, a border, a shadow. Would it survive being printed in grey. And can somebody who was not in the room read it alone.

The last one is the real test, because most charts are seen in an email rather than presented.

The formatting that actually helps

One accent colour, everything else grey. A single highlighted bar carries more than eight coloured ones, and the ward's chart uses grey for eleven months and one colour for the month being discussed.

Signal over noise

Three formatting practices that maximize chart legibility

Stripping decorative elements and directing visual focus helps charts communicate immediately inside emails.

Visual restraint
  • Single accent colourRender background series in muted grey and highlight only the active discussion metric in colour
  • Direct line labelsPlace series names directly at line ends to eliminate back-and-forth eye tracking to a legend
  • Selective data valuesLabel only key milestone numbers rather than dressing up an entire data table in chart clothing

Audience comprehension

Charts succeed by directing attention to a single finding rather than decorating every data point with saturated colours and redundant labels.

Direct labels instead of a legend, when there are three series or fewer: the name at the end of the line, in the line's colour, which removes the back-and-forth between a key and a chart.

And data labels only where the number matters. Every bar labelled is a table wearing a costume.

The chart to avoid entirely

The dual axis. Two different measures on two vertical scales, where the crossing point of the two lines is decided by whoever set the scales rather than by the data. It is the standard way to make two unrelated things look related, and it is usually produced accidentally by somebody who wanted both series visible.

Deceptive dual scales

The distortion of dual axes versus the honesty of stacked charts

Dual vertical scales create artificial crossing points and manufacture false relationships between unrelated metrics.

  • Dual-axis chartWhy it deceives
    • Crossing points change depending on arbitrary scale limits
    • Makes unrelated metrics appear causally linked
    • Forces readers to decipher which series aligns with which axis
  • Two stacked chartsWhy it communicates honestly
    • Shares a single horizontal time axis for honest timeline alignment
    • Keeps vertical scales independent and tamper-proof
    • Removes false visual intersections while presenting all data

Multi-metric comparison

Stacking two charts with a common horizontal axis provides clear side-by-side metric comparison without visual distortion.

Two charts stacked with a shared horizontal axis says the same thing and cannot be read into.

What to build them with, as of 2026

Excel's Recommended Charts, from Excel 2013, which is a reasonable starting point and defaults to a truncated axis on bar charts, so the first thing to do is open the axis options and set the minimum to zero.

Charting tools timeline

Key charting milestones from cell sparklines to AI advisory

Modern tooling assists in chart creation, but auditing axes and visual honesty remains a human task.

  1. 2010Sparklines in cellsPlaced micro-charts inside individual cells to show trends across 62 suppliers
  2. 2013Recommended ChartsProvided automated starting points that still require resetting bar axes to zero
  3. StandardCopy as PicturePrevents live chart formatting from distorting when viewed on recipient machines
  4. 2026AI chart reviewModels suggest chart types and verify logic, but cannot inspect rendered axes

While assistant models verify whether a chart type fits a question, verifying axis honesty against default spreadsheet options remains essential.

Sparklines, from Excel 2010, which put a small line inside a single cell. In a table of 62 suppliers they replace 62 separate charts and they read better in a row.

Copy as Picture for anything going into a document, which replaced pasting a live chart that reformats itself when the document is opened on another machine.

A model asked which chart answers a stated question gives a sound answer, free on every major assistant, and it will also tell you what a chart you describe implies. It cannot see the axis you actually ended up with, which is where the misleading happens, so the eight checks stay a human job.

The summary the chart is drawn from usually arrives as summarising thousands of rows in seconds, and highlighting exceptions inside the table instead of beside it is making the exceptions visible. Getting the numbers into one honest column first is text, dates and the imported mess, under the most used analysis tool in the world.

Alongside this

Above this