Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Biostatistics / Data and description

Graphs that tell the truth

12 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Choose the right graph for categorical and numerical data

  • Read histograms, box plots, scatter plots and line graphs correctly

  • Spot and avoid misleading graphs (truncated axes, 3-D effects, hidden denominators)

A graph is a statement. A good one lets the reader see the pattern in seconds; a bad one misleads, sometimes on purpose. The rule is simple: show the data, make comparison easy, and never let design hide the truth.

Match the graph to the question

You want to showUseAvoid
Counts or percentages across categoriesBar chart (bars start at zero)3-D bars, pie charts with many slices
Distribution of one numerical variableHistogram, box plotBar chart of the mean only
Compare a numerical variable across groupsSide-by-side box plots, dot plotsBars with only mean ± error
Relationship between two numerical variablesScatter plotConnecting points with a line
Change over timeLine graphBars for continuous time
Time to an eventKaplan-Meier curveBar chart of proportions at one time
Effect estimates from several studiesForest plotTable alone
020406080Children fully immunised (%)62%North74%East81%Central69%West
A bar chart done right: the axis starts at zero, so the bar heights are proportional to the values.

The histogram: see the shape

A histogram cuts a numerical variable into equal-width intervals ("bins") and draws a bar whose height is the count (or percentage) in each bin. Bars touch, because the variable is continuous. It reveals the shape (symmetric, skewed, two peaks), the spread and outliers. Bin width matters: too wide hides detail, too narrow looks like noise. Try several.

The box plot: compare groups

A box plot shows five numbers: the minimum, Q1, the median, Q3 and the maximum (often with points beyond 1.5 × IQR from the box drawn separately as potential outliers). The box holds the middle half of the data; the thick line is the median. Placing box plots side by side compares groups in seconds.

100125150175Systolic blood pressure (mmHg)ControlIntervention
Box plots: the intervention group has a lower median and a narrower spread; one control participant is a possible outlier.

The scatter plot: two numerical variables

Plot one variable on each axis, one point per person. Look for a pattern (straight, curved, none), its direction and strength, and outliers. A scatter plot is always the first step before correlation or regression (chapter 6).

The line graph: change over time

Join points in time order. Use the same scale for groups you want to compare, label the axes with units and mark the intervention or event date.

05010015020182019202020212022YearCases per 100,000District A (intervention)District B (comparison)
A line graph with a common scale and a zero baseline lets the reader compare trends honestly.

How graphs mislead

Most misleading graphs are not frauds; they are careless. Learn these traps so you neither fall for them nor commit them.

  1. Truncated axis. A bar chart whose axis starts at 60 instead of 0 makes a difference between 62% and 69% look like a threefold gap. Bars encode length, so they must start at zero. (Line graphs may use a non-zero baseline if it is clearly labelled.)
  2. 3-D and decorative effects. Perspective distorts the visual size of bars and slices. Flat is honest.
  3. Pie charts. Humans compare angles poorly. Use a pie only for 2 to 3 slices adding to a whole; otherwise use a bar chart sorted by size.
  4. Dual axes. Two vertical scales on one graph let you make any two series look correlated by choosing the scales. Prefer two panels.
  5. Hidden denominators. "Cases doubled" means little if the population tripled. Show rates, and show n.
  6. Cherry-picked time window. Starting a line graph at a peak makes any decline look like success. Show enough history.
  7. Means without spread. A bar of the mean with no spread, or "error bars" that are not defined, hide the overlap between groups. State whether bars show SD, SE or a 95% CI.

The ink test

Ask of every element: "Does this help the reader understand the data?" If not, remove it. Remove gridlines that do not help, 3-D, shadows and unneeded colours. Use colour to mean something (a group), and check it reads in grayscale and for colour-blind readers.

Fixing a poor graph

Before: a 3-D pie chart with seven slices of health-facility types, a rainbow legend and no counts.

After: a horizontal bar chart sorted from largest to smallest, one calm colour, counts and percentages printed at the end of each bar, and the title stating the finding: "Dispensaries make up half of the 214 facilities". The reader now gets the message in two seconds and can compare any two bars exactly.

A graph checklist

Title says what and where and when. Axes labelled with units. Zero baseline for bars. n shown. Spread defined (SD, IQR, CI). Source cited. Colours explained. Could someone understand it without the text?

Summary

  • Pick the graph from the question and the variable types: bars for categories, histograms and box plots for distributions, scatter plots for relationships, lines for time.

  • Bar charts must start at zero; line graphs may not, but must be labelled.

  • Avoid 3-D effects, many-slice pies, dual axes and cherry-picked time windows.

  • Always show spread (and n), and define error bars.

  • Make the title a finding, not a label.

Check your understanding

Question 1
1

You want to compare the distribution of haemoglobin between three treatment groups. The best choice is:

Question 2
2

A bar chart's vertical axis starts at 60% instead of 0%. The main problem is that it:

Question 3
3

In a box plot, the box contains:

Question 4
4

Which is the best way to show how one numerical variable changes with another across individuals?

Question 5
5

'Cases doubled this year' may mislead because:

Question 6
6

Error bars on a graph should always be:

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

Summarising data: centre, spread and shapeProbability rules and why a positive test can mislead