A plant bends towards a window.
What did you observe?
What do you infer?
What can you conclude?
These questions sound similar.
They are not.
The plant bending is an observation.
Saying that light influenced the direction of growth is an inference or explanation built from the observation and scientific knowledge.
Saying that all plants always grow towards any light source is a much broader claim that the single observation does not justify.
Primary Science becomes more reliable when students separate what the world directly showed them from what their minds added to explain it.
This article focuses on the boundary between observation, inference, prediction, explanation and conclusion.
The central loop is:
observe → interpret → connect to model → make bounded claim → test again
Quick Answer: What Is an Observation?
An observation is information obtained through the senses or through measuring instruments.
Examples:
- the liquid level rose by 3 cm;
- the bulb lit up;
- the leaf turned yellow;
- the mass decreased from 50 g to 46 g;
- the ice cube became smaller;
- bubbles formed around the electrode.
An observation should describe what was detected, not automatically why it happened.
Quick Answer: What Is an Inference?
An inference is an interpretation that uses observations together with prior knowledge to suggest what may explain them.
Observation:
The outside of the cold cup became wet.
Inference:
Water vapour in the surrounding air condensed on the cold surface.
The inference is not directly seen.
It is a model-based interpretation of what was seen.
Observation Is Not “Fact” and Inference Is Not “Guess”
Students sometimes hear:
Observation = fact.
Inference = guess.
That is too crude.
Observations can contain measurement error.
Different observers can describe the same event differently.
Instruments can have limited precision.
Inferences can be strongly supported when several observations and established scientific models agree.
The useful distinction is:
observation = detected information;
inference = meaning constructed from that information.
The Observation-Inference Boundary
Consider a beaker of water heated on a hot plate.
Possible observations:
- the thermometer reading increased;
- bubbles appeared;
- steam or water vapour became visible near the top under certain conditions;
- the water level decreased after prolonged heating.
Possible inferences or explanations:
- the water gained thermal energy;
- some liquid water changed state;
- the lower water level resulted from water leaving the beaker as gas.
The scientific model connects the observations.
Measurement Is an Observation Through an Instrument
A thermometer, balance, ruler, stopwatch or sensor extends observation.
But the reading is still a representation.
A thermometer does not display “how hot something truly is” in a philosophical sense.
It provides a measurement according to the instrument’s scale, design and precision.
Students should therefore learn to record:
- value;
- unit;
- relevant time or condition;
- comparison point where needed.
“It got hotter” is lower resolution than “the temperature increased from 25°C to 43°C in five minutes”.
Observation Resolution Controls Claim Resolution
A low-resolution observation supports only a low-resolution claim.
Observation:
Plant A looked taller than Plant B.
This may support a rough qualitative comparison.
Measurement:
Plant A measured 18 cm and Plant B measured 12 cm after 14 days.
Now the comparison has higher resolution.
The quality of the claim cannot exceed the quality of the evidence that carries it.
Prediction Is Different Again
A prediction projects a model forward.
Observation:
When the amount of light decreased, the measured growth rate changed.
Inference:
Light availability may be affecting the plant’s growth under these conditions.
Prediction:
If a similar plant receives less light under otherwise similar conditions, its growth may also be reduced.
The prediction is a future-facing claim generated from the model.
A Conclusion Must Match the Investigation
Suppose two identical containers of water are placed in different locations.
Container A loses more water than Container B.
A student may conclude:
Heat causes evaporation.
But if several conditions differed—temperature, airflow, exposed surface area—the investigation may not isolate one cause.
A safer conclusion is bounded by the actual comparison.
Under the conditions tested, Container A lost water faster than Container B.
Then the student can discuss which variable may explain the difference if the experimental design supports that inference.
Fair Comparison Protects Inference
If two cases differ in many ways, the observation may be clear while the causal inference is weak.
Example:
- Plant A gets more water, more light and a larger pot.
- Plant B gets less water, less light and a smaller pot.
Plant A grows taller.
Observation:
Plant A was taller.
Invalid strong inference:
More water caused Plant A to grow taller.
Several variables changed.
The comparison does not isolate water.
Correlation Is Not Automatically Cause
Two things can change together without one directly causing the other.
Primary students do not need advanced statistics to learn this habit.
They can ask:
- What else changed?
- Was the comparison fair?
- Is there another possible explanation?
- Did we actually test the proposed cause?
This protects scientific reasoning from certainty inflation.
The Language of Claims Matters
Compare:
- the data show;
- the data suggest;
- the observation is consistent with;
- this may indicate;
- this proves.
These phrases carry different levels of commitment.
Scientific writing should not make a stronger claim than the evidence can support.
Claim strength should track evidence strength.
Worked Example 1: Condensation
A cold metal can is taken from a refrigerator and placed on a table. Water droplets later appear on the outside.
Observation: droplets appeared on the outside of the can.
Prior knowledge: air contains water vapour; cooling can cause water vapour to condense.
Inference: water vapour in the surrounding air cooled at the can’s surface and condensed into liquid droplets.
Common wrong inference:
Water leaked through the can.
The observation alone does not distinguish the two explanations.
Scientific knowledge and further testing help decide.
Worked Example 2: Shadows
A torch is moved closer to an object and the shadow on a screen becomes larger.
Observation: the shadow size increased when the torch-object arrangement changed.
Inference/model: changing the relative positions of light source, object and screen changes how light rays are blocked and therefore the shadow size.
Overclaim:
A brighter light always makes a bigger shadow.
The observation did not test brightness as the causal variable.
Worked Example 3: Magnets
A paper clip moves towards a magnet before touching it.
Observation: the paper clip moved towards the magnet without direct contact.
Inference: the magnet exerted a magnetic force over a distance.
The force is not directly seen.
Its effect is observed.
This distinction helps students understand how Science often infers invisible mechanisms from visible effects.
Worked Example 4: Plant Growth
Two similar plants are grown with the same soil, pot size and water, but one receives more light. After two weeks, their growth differs.
Observation: measured growth differed between the two plants.
Inference: the difference in light exposure may have contributed to the different growth under the conditions tested.
Stronger causal confidence requires careful control, repetition and consideration of other sources of variation.
Anomaly Is Information
Suppose four measurements follow a clear pattern and one does not.
Students sometimes erase the odd value because it “spoils” the conclusion.
That destroys information.
Ask:
- Was there measurement error?
- Was the procedure inconsistent?
- Is the model incomplete?
- Is the variation real?
- Should the experiment be repeated?
An anomaly is a return signal from the world.
Repeatability Changes Confidence
One observation can be interesting.
Repeated similar observations under controlled conditions increase confidence that the pattern is not a one-off accident.
This does not make every conclusion absolute.
It improves the evidence base.
Observation Can Be Qualitative or Quantitative
Qualitative:
The leaf became paler.
Quantitative:
The mass decreased by 4 g.
Both can be scientifically useful.
The choice depends on what the investigation needs to distinguish.
Tables and Graphs Are Observation Representations
Raw observations are often converted into tables and graphs.
This changes the representation, not the underlying data.
A graph can make a trend visible.
But the graph does not automatically explain the trend.
The student still needs a scientific model to connect pattern to mechanism.
The Observation → Model → Claim Ladder
A useful answering structure is:
- Observation: what changed or differed?
- Concept/model: which scientific idea explains the observation?
- Mechanism: how does the concept produce the observed result?
- Claim: what conclusion is justified?
This helps students avoid jumping directly from a picture to a memorised keyword.
The “Did I See It?” Drill
For each statement, ask:
Could I directly see or measure this, or did I infer it?
Examples:
- “The temperature rose.” → observable/measurable.
- “The particles moved faster.” → model-based inference at Primary level.
- “The bulb became brighter.” → observation if brightness comparison is visually evident.
- “More current flowed.” → inference unless current was measured directly.
This drill teaches epistemic boundaries.
The Claim-Strength Drill
Give one set of evidence and three possible claims.
Ask which is justified.
- Claim A: describes only the observed pattern.
- Claim B: gives a plausible mechanism supported by the design.
- Claim C: makes a universal statement far beyond the tested conditions.
Students learn that stronger-sounding language is not automatically better Science.
The Alternative Explanation Drill
After making an inference, ask:
What else could explain this observation?
Then ask what additional observation or experiment would distinguish between the explanations.
This converts uncertainty into an investigation plan.
Common Failure Mode 1: Explanation Written as Observation
The student writes an invisible mechanism when asked what was observed.
Repair: use sensory or measurement language only.
Common Failure Mode 2: Observation Repeated as Explanation
The question asks why the bulb is brighter.
The answer says:
Because it became brighter.
The observation has been repeated without mechanism.
Repair: connect the observed difference to the relevant scientific concept.
Common Failure Mode 3: Universal Claim From One Case
One successful test becomes “this always happens”.
Repair: bound the conclusion to the tested conditions.
Common Failure Mode 4: Fair-Test Weakness Ignored
Several variables changed but the student assigns the outcome to one cause.
Repair: identify alternative explanations and redesign the comparison.
Common Failure Mode 5: Graph Pattern Becomes Causal Claim
The student sees two variables rise together and immediately states that one caused the other.
Repair: separate pattern description from causal explanation.
Current PSLE Science Explicitly Requires Inquiry Reasoning
The 2026 PSLE Science syllabus assesses knowledge and understanding together with the application of scientific inquiry. Candidates are expected to make predictions and hypotheses, interpret and analyse information, evaluate observations, information and methods, and communicate explanations and reasoning using words, diagrams, tables and graphs.
That makes the observation-inference boundary central to examination performance.
The student must know not only a scientific concept, but what kind of claim the available evidence can legitimately support.
Parent-Friendly Science Questions
- What did you actually see or measure?
- What part of your answer is an inference?
- Which scientific idea connects the two?
- Could another explanation fit the observation?
- Was the comparison fair enough to support the conclusion?
- Did your conclusion go beyond what was tested?
Tutor-Friendly Science Questions
- Can the student separate observation from mechanism?
- Can they convert qualitative observations into measured comparisons when appropriate?
- Can they calibrate claim strength?
- Can they identify alternative explanations?
- Can they evaluate whether the method isolates the intended variable?
- Can they use anomalous results as diagnostic information rather than deleting them?
Official Singapore Science Reference
Final Principle: Separate the World From the Model, Then Reconnect Them Carefully
Science begins with contact with the world.
Observe.
Measure where useful.
Represent the information.
Then infer.
Use a model.
Make a bounded claim.
And return to the world with another test.
Reliable Science is not refusing to infer. It is knowing which parts came from observation, which parts came from the model, and how strongly the evidence allows the two to be connected.
That is observation versus inference.