Wait, What? A graph can make bad data look convincing.
A smooth line, labelled axes and a neat trend do not guarantee a strong conclusion. Data presentation is not decoration. It is part of the reasoning chain that connects what you measured to what you are justified in claiming.
Strong practical science follows a discipline: preserve the raw evidence, present it without distortion, look for structure, test alternative interpretations, then state only the conclusion the data can support.
Start with a table that protects the evidence
A good results table makes the structure of an investigation visible. The independent variable usually appears in the first column, followed by measured dependent-variable values and any calculated quantities. Units belong in headings, not repeated after every number.
Raw measurements should normally be preserved. If you measure 12.4 s, 12.6 s and 12.5 s and calculate a mean of 12.5 s, do not discard the original readings. The scatter itself is evidence about repeatability.
Decimal places tell a story
If the same instrument is used under the same conditions, measurements of the same type should generally be recorded consistently. Writing 4.2, 5.17 and 6 for readings from one instrument may hide how precisely the observations were actually made.
Calculated values should also avoid false precision. A calculator can create digits that were never present in the experiment. The number of reported figures should reflect the information content of the measurements, not the display capacity of the calculator.
Choose the graph for the scientific question
For two continuous quantitative variables, a scatter plot or line graph is often appropriate. For categories, a bar chart may be more appropriate. The graph type should reveal the relationship rather than merely satisfy a memorised classroom rule.
Axes should be labelled with quantity and unit. Scales should use the plotting area effectively and remain easy to interpret. A scale chosen to exaggerate tiny differences can mislead; a scale so compressed that meaningful variation disappears is equally unhelpful.
Best-fit does not mean connect-the-dots
When the purpose is to infer an underlying relationship, a best-fit line or curve represents the pattern supported by the collection of points. Joining every point creates a path through measurement noise and can imply changes for which there is no scientific basis.
At higher levels, the chosen model matters. Is the relationship linear? Proportional? Inverse? Exponential? Does a transformed graph produce a more informative test? A straight-looking segment over a narrow range is not proof that the mechanism is linear everywhere.
Anomalies are questions, not rubbish
An anomalous result is one that does not fit the broader pattern as expected. Students often learn the dangerous habit “remove anomalies.” Science requires more care.
First ask whether there is a reason to suspect a procedural failure, transcription mistake, equipment problem or genuinely unusual event. If time permits, repeat the relevant condition. If the value remains unusual, it may reveal something about the phenomenon rather than merely the experimenter.
The Royal Society of Chemistry’s practical guidance encourages learners to identify anomalous results and, where possible, repeat measurements rather than casually dismiss them. See the RSC evaluating-experiments guidance.
Correlation is evidence of association, not automatic causation
In a controlled experiment, changing one variable while controlling plausible alternatives strengthens a causal interpretation. In observational data, two quantities can vary together because one affects the other, because the direction is reversed, because a third factor affects both, or by coincidence.
Therefore “the graph goes up” is not enough. A conclusion should connect the trend to the design and acknowledge what the method can and cannot establish.
Interpolation and extrapolation
Interpolation estimates within the measured range. Extrapolation predicts beyond it. Extrapolation is usually riskier because the relationship may change outside the observed range.
Imagine a reaction rate rising approximately linearly from 20 °C to 40 °C. Extending that line blindly to 200 °C would ignore changes in state, reaction pathway, enzyme denaturation in biological systems, apparatus limits and other mechanisms. A mathematical line is not permission to ignore physics, chemistry or biology.
How to write a conclusion that deserves marks
A strong conclusion does three things: states the relationship, supports it with relevant evidence, and stays within the limits of the investigation.
Weak: “Temperature increases rate.” Better: “Within the tested range, increasing temperature increased the measured reaction rate; for example, the time to collect the fixed gas volume decreased as temperature rose.” Stronger still: explain whether the relationship appears linear, proportional or otherwise, if the data justify that claim.
The Singapore-Cambridge practical framework explicitly assesses analysis and interpretation, conclusions, predictions, significant sources of error and improvements. See the SEAB 2026 Chemistry practical assessment framework.
Observation, transformation, inference
It helps to separate three layers. Observation: the recorded reading. Transformation: a calculation such as mean, gradient, reciprocal or percentage change. Inference: what the pattern suggests about the scientific system.
Confusing these layers can make reasoning circular. You do not “observe” that resistance is proportional to length merely by looking at a wire. You measure quantities, transform or graph them, then infer a relationship from the evidence.
Secondary → JC progression
Secondary: construct clear tables and graphs; use units; calculate means; spot patterns; treat anomalous values cautiously; draw evidence-based conclusions; suggest improvements.
JC: use gradients, intercepts and transformations as model tests; reason about uncertainty and error bars where appropriate; compare competing functional relationships; distinguish interpolation from extrapolation; discuss whether deviations expose limitations of the model.
Checkpoint: the suspicious straight line
A student measures current through a component at five voltages. Four points lie close to a straight line through the origin. One point lies far above the line.
- Should the point be deleted immediately?
- What should the student check first?
- If a repeat falls close to the original four-point trend, what does that suggest?
- What would a straight line through the origin mean physically if the relationship is genuinely proportional?
Answer key and WHY reasoning
No. The point is evidence until there is reason to treat it otherwise. Check the raw record, apparatus connections, scale reading and whether conditions changed. If a repeat at the same voltage lies near the broader trend, the original point is more plausibly associated with a measurement or procedural disturbance. A straight line through the origin supports proportionality over the tested range: doubling one quantity would correspond approximately to doubling the other, within experimental uncertainty.
How to study practical data
Take old experimental datasets and practise without the original question. Ask yourself: What quantities are here? What graph would expose the relationship? Which point deserves investigation? What model might fit? What claim is justified? What claim would go too far?
This trains the skill that practical examinations are really testing: turning measurements into a defensible scientific argument.
Authoritative next steps
- SEAB 2026 O-Level syllabus directory
- SEAB 2026 A-Level syllabus directory
- Royal Society of Chemistry: Evaluating experiments
Teaching Guide
For parents and teachers: give students imperfect datasets, not only tidy textbook graphs. Ask what can be claimed, what cannot, what should be repeated and what evidence would change their mind. The target is not prettier graph paper. It is intellectual discipline: preserving the difference between data and the story we tell about the data.