Reading research and graphs: conclusions the evidence can support
Start with a clear question, then examine study design, axes, individual observations and uncertainty—with an interactive graph and a worked conclusion.
The article’s charts and datasets are fictional examples created for learning.
Read & put into practice
A taller bar can make us think, “This method is better.” Pause to ask: better by how much, on which measure, compared with what, and what has been left out? Those short questions are the beginning of a careful reading.
You do not need to understand every technical term before you begin. Move from the question to the method and evidence, then write a conclusion proportionate to what is known. This connects with inquiry evaluation and evidence use in the PISA 2025 framework; the exercises below are our own teaching examples. [1]
Key ideas to take away
- Read how the data were obtained, not only the result.
- Check units, the comparison baseline and individual observations.
- A useful conclusion says what was found and what remains unresolved.
Change the view. Read it again.
Fictional data: seedling heights from four pots in each group. Switch between three views. What changes, and what stays the same?
Mean height in each group
Compare the means in proportion
A averages 10 cm and B averages 12 cm. With a zero baseline, bar length reflects magnitude. The bars still hide how much the individual heights differ.
| Group | Observations | Mean |
|---|---|---|
| A | 9, 10, 10, 11 | 10 |
| B | 6, 8, 16, 18 | 12 |
Difference 2 cm · B’s mean is higher than A’s by 20%
Each value represents one pot, not repeated measurements of one plant. Allocation and starting heights are unknown. Use this dataset to practice reading graphs; it cannot establish that a treatment caused better growth.Five questions to guide your reading
Before reading the entire conclusion, write down what the researchers want to find out. Then locate the evidence they use to answer that question. This makes gaps between the question, method and claim easier to notice. Your first pass does not have to follow every line in order.
What is the question?
Is the study describing, comparing, predicting or testing causation? “Do the groups differ?” is not the same question as “Did this cause the difference?”
What was studied?
Identify the population or material, setting, time period and unit counted as one observation. Mark missing information instead of guessing.
How was the comparison made?
Look for a comparison group, allocation method, controlled conditions and starting differences. Two measurements of the same person are linked observations.
What was actually observed?
Record values, units, sample size and uncertainty. Separate what was measured from an explanation of why it happened.
How far does it apply?
Ask how new people, materials or settings differ from those studied. A result in one setting can provide a starting point for another question.
Read the axes before the height
Bar length represents magnitude, so the UK Government Analysis Function recommends a zero baseline for bars. A clearly labeled dot plot is an alternative for examining small differences. The lab above deliberately includes truncated bars so you can see how an impression changes without the means changing. [2]
In the fictional dataset, mean seedling height is 10 cm for A and 12 cm for B. The difference is 2 cm, or 20% relative to A: (12−10) ÷ 10 × 100. A percentage needs a stated baseline. It does not mean every seedling is 20% taller.
| Check | Question | Possible mistake |
|---|---|---|
| Variable and unit | Height, mass, concentration or a score? | Comparing incompatible units |
| Range and scale | Where does it start? Are steps equal or logarithmic? | Overstating or understating a difference |
| Denominator | A percentage of how many, with whom excluded? | Comparing rates without their base counts |
| Time | Are comparable time windows shown? | Inferring a trend from a selected window |
The mean is only part of the story
Weissgerber and colleagues examined data presentation in 703 physiology papers and showed how different distributions can produce similar summary graphs. Individual observations can make a small dataset easier to inspect. Their review concerns selected papers from 2014, not the quality of all research today. [3]
Our invented values are A: 9, 10, 10, 11 and B: 6, 8, 16, 18 cm. B has the higher mean, yet two B seedlings are shorter than every A seedling, and B is more spread out. Seeing all observations invites questions about conditions and initial differences.
The number of measurements is not always the number of experimental units. Measuring one plant four times still involves one plant. If treatments are assigned to pots, report the pots. Here each value represents a different pot, but allocation details are absent, so the example does not establish causation.
What do the small lines mean?
Read the caption before interpreting error bars. Standard deviation (SD) describes the spread of observations; standard error (SE) concerns uncertainty in the estimated mean. They answer different questions, so short bars do not necessarily mean the observations are tightly clustered. [3]
A confidence interval for a mean expresses uncertainty under a method and its assumptions. Its width relates to sample size and variability. A 95% confidence interval for the mean is not an interval containing 95% of individual observations. [4]
If you encounter a p-value, start here: it is not the probability that a hypothesis is true, and it does not measure effect size or importance. The ASA calls for context and complete reporting. Crossing 0.05 alone cannot establish practical value or universal applicability. [5]
What is needed to move from association to causation?
Suppose science-club members interpret graphs better. The two features are associated, but volunteers may already have stronger interest or preparation. Several factors could explain the difference. Write at least two alternative explanations before making a causal claim.
A before-and-after comparison shows change, but practice, classroom learning or missing follow-up can also matter. An appropriate comparison and allocation method can strengthen the design; implementation and missing data still need checking. Evaluating the method is part of evaluating the evidence. [1]
Write a conclusion that fits the data
| Part | Example wording | Why include it? |
|---|---|---|
| Observation | In this dataset, B averages 12 cm and A 10 cm: a 2 cm difference. | States magnitude and comparison |
| Individual variation | B ranges from 6 to 18 cm, so not every B plant exceeds every A plant. | Retains information hidden by the mean |
| Scope | There are four pots per group; allocation and initial heights are not supplied. | Identifies missing information |
| Next step | Check starting conditions and allocation before attributing the difference to B’s treatment. | Turns a limitation into an actionable question |
Practice with one figure first
Choose a figure from a paper permitted by your course or program rules. Write three sentences: what it measures, what it supports and what is missing. Check the methods afterward. If your interpretation changes, record which new detail changed it. [6]
Follow a citation to the original work and check its date, study type, authors and any correction notice. An abstract helps screen relevance. If you have not read the methods, say that your interpretation is still based on the abstract.
From reading to your next small step
Before using a conclusion
Check off what you have reviewed. Use the rest as reading questions, not as a research-quality score.
An unanswered question is a useful starting point for your next reading.
Selections stay on this page, are not submitted and reset when the page is reopened.Evidence-based reading starts with a clear question and ends by stating what you know, what supports it and what you still need to learn.
How this article was developed
This explanatory synthesis draws on the published 2026 OECD framework, the PLOS Biology article, UK chart guidance, the NIST handbook and the ASA release, checked on 2 October 2026. Sources were selected to teach evidence reading; this is not a systematic review.
The seedling dataset, charts, prompts and conclusion are original fictional examples. No TechEd participant experiment was conducted. No p-value or confidence interval is calculated for this dataset, and the checklist is not a validated competency measure.
TECH EDUCATION operates the programs discussed and publishes this article, creating an interest in the subject. This is a source-based editorial analysis, not an independent evaluation, and has not undergone external peer review.
Explore the original sources
- 01OECD · PISA 2025 Assessment and Analytical Framework (2026)
Chapter 2 · Evaluating inquiry and interpreting evidence; an assessment framework, not a trial of our activities
- 02UK Government Analysis Function · Bar charts (n.d.)
Guidance on zero baselines for bars and alternatives for highlighting small differences
- 03Weissgerber, T. L. et al. · Beyond Bar and Line Graphs: Time for a New Data Presentation Paradigm (2015)
PLOS Biology 13(4), e1002128 · Article and figures on the distributions hidden by summary graphs
- 04NIST/SEMATECH · e-Handbook: Confidence Limits for the Mean (n.d.)
Section 1.3.5.2 · Confidence limits for a mean and their assumptions
- 05American Statistical Association · Statement on Statistical Significance and P-Values (2016)
7 March 2016 release · Six principles, especially the limits of a p-value
- 06TECH EDUCATION · Exam scope (2026)
Check program scope and rules before applying a practice method
Sources checked October 2, 2026 · Please contact the team with a source if you identify a correction. Contact the team
Explore the related programs
Review the aims, activities and edition-specific conditions before choosing your next experience.
K-XCELK-XCEL | Korea eXcellence through Competency, Experience & Learning IMEDDATMedical Science Aptitude Development and Testing SCOPE KOREASCOPE KOREA | Student Conference on Presentation, Innovation & Evaluation