PSLE-SCI-REALITY-0457
Wait, What? The Wobbly Data Became a Beautiful Smooth Line
A science dashboard shows twelve years of monthly measurements. The grey raw line jumps up and down. Beside it, a thick blue line labelled 12-month running mean moves much more smoothly. A learner points to the blue line and says, “So the real measurements changed smoothly like that.”
Not necessarily. A running mean is a transformed summary built from several measurements at a time. It can reveal a slower pattern that is difficult to see in noisy data, but it can also hide short-lived peaks, dips and sudden changes. A point on the smoothed line is usually not one direct instrument reading.
This is a strong PSLE Science evidence habit: do not ask only whether a graph looks clear. Ask what operation turned the observations into the displayed line, what information survived that operation, and what information became harder to see.
Quick Answer
- A running mean or moving average combines neighbouring observations into averages.
- The smoothed point is therefore derived from a window of data, not usually a new direct observation.
- A longer window generally removes more short-term variation.
- Peaks and dips can become smaller, wider or delayed depending on the method.
- A centred window and a trailing window do not place the average at the same time position.
- To evaluate a claim, inspect the raw measurements as well as the smoothed series whenever short-term events matter.
The Exact Learner Job This Reality Lab Owns
Owned learner job: evaluating a real scientific graph, dashboard or news-style chart that uses a running mean, moving average or smoothing window, and deciding which parts of the displayed curve are direct observations and which parts are calculated summaries.
Not owned here: generic averaging, graph-reading rules, trend-line fitting, statistics lessons, climate science, sensor theory or exam-answer templates. Those jobs belong to their existing owners. This page is the transfer layer: it teaches what to do when a real communication object quietly replaces a jagged series with a smoother one.
Why This Belongs in PSLE Science Reasoning
The current 2026 PSLE Science assessment is based on Singapore’s 2023 Primary Science syllabus. SEAB states that Application of Knowledge and Scientific Inquiry includes interpreting and analysing information, evaluating observations, information and methods, and communicating explanations and reasoning. The syllabus also encourages healthy scepticism and attention to how scientific knowledge is communicated.
A running mean is an ideal evidence-transfer object because nothing about it is automatically dishonest. Scientists use smoothing for legitimate reasons. The learner’s task is subtler: understand exactly what the transformation does before using the picture as evidence.
Rebuild the Object: The School-Roof Temperature Logger
Imagine a fictional school science club records one noon temperature each day for twelve days. These numbers are constructed for learning:
| Day | Measured temperature (°C) |
|---|---|
| 1 | 30 |
| 2 | 31 |
| 3 | 30 |
| 4 | 36 |
| 5 | 33 |
| 6 | 31 |
| 7 | 30 |
| 8 | 29 |
| 9 | 30 |
| 10 | 31 |
| 11 | 30 |
| 12 | 30 |
Day 4 is a short hot spike. Now calculate a simple three-day centred mean for Day 4 using Days 3, 4 and 5:
(30 + 36 + 33) ÷ 3 = 33°C
The original Day 4 measurement was 36°C. The smoothed value placed at Day 4 is 33°C. Neither number is “fake”. They answer different questions. The raw value describes that particular observation. The running mean describes the local average across a chosen window.
Observed, Calculated, Displayed, Claimed
| Layer | Example | Evidence meaning |
|---|---|---|
| Observed | 36°C on Day 4 | A direct reading under the stated method |
| Calculated | 33°C three-day mean centred on Day 4 | A summary of Days 3–5 |
| Displayed | A smooth blue curve | A representation of many calculated summaries |
| Claimed | “Temperature never rose sharply” | Needs checking against raw data |
This four-layer split prevents a common mistake: treating the prettiest line as if it were the original measurement record.
The Window Is Part of the Evidence
A running mean needs a window length. A three-day window combines three neighbouring observations. A seven-day window combines seven. A 12-month climate running mean combines twelve monthly values. The chosen window changes the picture.
If the window is short, much of the original variation remains visible. If the window is long, the curve becomes smoother, but short events become easier to miss. This means the phrase “the trend looks stable” is incomplete unless the reader knows the smoothing window.
Worked Case 1: One Spike, Three Windows
Suppose a sensor records the constructed sequence 10, 10, 10, 22, 10, 10, 10. The raw series contains a single large spike.
- With no smoothing, the spike is obvious: 22.
- With a three-point mean centred on the spike, the central value becomes (10 + 22 + 10) ÷ 3 = 14.
- With a five-point mean, the central value becomes (10 + 10 + 22 + 10 + 10) ÷ 5 = 12.4.
The longer window did not prove the event was smaller. It spread the influence of one observation across more positions. If the scientific question is “Was there a short-lived spike?”, the raw data are essential. If the question is “What was the slower background level?”, smoothing may help.
Centred Versus Trailing: Where Does the Average Sit?
A centred three-point mean for Day 5 might use Days 4, 5 and 6. A trailing three-point mean shown at Day 5 might use Days 3, 4 and 5. Both are legitimate conventions when clearly defined, but they place information differently in time.
This matters when someone claims that a change “started” at a particular date. A trailing smoother can respond after the underlying observations begin to change because the window still contains older values. A centred smoother uses future and past neighbours around each displayed point, so it may be unsuitable for real-time interpretation even though it is useful for retrospective analysis.
Worked Case 2: The Sudden Step
An original dataset stays at 5 units for six days, then jumps to 15 units and remains there. The real process changed suddenly. A five-day moving average would not jump from 5 to 15 in one step. It would pass through intermediate averages as the window gradually replaces old 5s with new 15s.
If a news-style graphic shows only the smoothed curve, a reader may think the underlying system changed gradually. That conclusion would be about the shape of the smoother, not necessarily the shape of the raw process.
Representation Check: What Is Hidden?
- Are the raw points shown faintly behind the smoothed line?
- Is the smoothing method named?
- Is the window length stated?
- Is the mean centred, trailing or another form?
- Are missing values skipped, replaced or allowed to break the window?
- Are values weighted equally?
- Are end points omitted because a full window is unavailable?
A responsible graph need not explain every mathematical detail in the headline. But enough information should exist for a careful reader to know what transformation produced the displayed pattern.
The Edge Problem
Imagine a centred 12-month mean. Near the very beginning of the record, there may not be six earlier months available. Near the end, there may not be six later months. Different systems handle this in different ways: they may leave gaps, use shorter windows, extend the series with assumptions or use trailing calculations.
So if a smooth scientific line stops before the raw series stops, that does not automatically mean the dataset ended. It may simply mean the selected smoother needs neighbouring observations that are not yet available.
Missing Data Can Change a Running Mean
Suppose a five-day window contains four measurements and one missing observation. A system might refuse to calculate the mean. Another might average the four available values. A third might fill the missing value first. Those choices can produce different curves.
That is why a smoothed line should not be separated from its data-processing notes when the missingness matters. The calculation rule is part of the method.
Comparison Check: Two Smooth Lines Can Use Different Windows
Imagine Product A is advertised with a 30-day moving average of sensor noise while Product B is shown with a seven-day moving average. Product A will often look smoother even if the underlying observations are equally variable. Comparing line smoothness alone would be unfair.
Before comparing two smoothed datasets, check that the window, weighting, missing-data rule and time basis are comparable. If the transformations differ, first return to the underlying values or recompute the summaries on the same basis.
Worked Case 3: The Product Dashboard
A fictional air-monitoring product page shows a smooth line labelled “Average particle level” and says, “Our readings remain exceptionally stable.” The line varies only from 18 to 20 units across a month.
The small print reveals that each plotted point is a seven-day running mean. Raw hourly values ranged from 7 to 42 units. The smooth line can support a statement about seven-day averages. It cannot by itself prove that individual hourly measurements remained close to 19.
The repair is simple: rewrite the claim to match the object. “The seven-day running average remained between 18 and 20 units during this period.” That is more precise and scientifically defensible.
Worked Case 4: The Viral “No Spike” Claim
A social-style post shows a heavily smoothed graph and says, “There was no sudden spike at all.” The correct response is not to accuse the author of manipulation. Ask whether the chosen smoothing could suppress or spread a short event.
- Request or inspect the unsmoothed series.
- Identify the running-mean window.
- Check whether the event duration is shorter than the window.
- Compare the raw maximum with the smoothed maximum.
- Check whether the claimed absence concerns raw measurements or the longer-term average.
Healthy scepticism means testing the claim against the processing method, not assuming bad intent.
What Smoothing Can Legitimately Help You See
Smoothing can be extremely useful when short-term fluctuations make a slower pattern hard to see. NOAA scientific plotting tools explicitly offer running means for time series, and NASA climate tools display multi-month running means. These are not tricks. They are declared transformations used to answer a particular scale of question.
The key is matching the transformation to the question. If the question concerns a multi-month background tendency, smoothing can reduce distracting short-term variation. If the question concerns an extreme one-day event, the same smoother can hide exactly the evidence that matters.
What Evidence Strengthens a Smoothed-Graph Claim?
- The raw series is available or shown alongside the smoother.
- The window length is stated.
- The method explains centred versus trailing placement.
- Missing-data handling is described.
- The conclusion matches the timescale of the smoothing.
- Important short-lived events are checked separately rather than declared absent from the smooth line alone.
- Comparisons use equivalent processing.
What Weakens It?
- Only the smooth line is shown while the claim concerns short-term extremes.
- The smoothing window is hidden.
- Different series use different windows without disclosure.
- The author treats derived points as direct observations.
- The graph’s end behaviour is interpreted without considering incomplete windows.
- Missing values are filled silently.
Tempting Reasoning That Fails
| Tempting claim | Why it fails | Better question |
|---|---|---|
| “The blue line is the real data.” | It may be calculated from several raw values. | What operation produced each point? |
| “A smooth line proves the system was stable.” | Smoothing itself reduces visible variation. | How variable were the raw observations? |
| “No spike appears, so no spike occurred.” | A short spike may be diluted across a wide window. | What do unsmoothed values show? |
| “A longer moving average is more accurate.” | Longer and shorter windows answer different timescale questions. | Which window matches the scientific job? |
| “The curve changed gradually, so the process changed gradually.” | A sudden step can become gradual after smoothing. | Does the raw record show a step? |
How Far Can the Conclusion Travel?
A declared 12-month running mean can support statements about the average level across its 12-month windows. It can help reveal longer-term movement. It can help compare periods when the same method is used consistently.
It cannot automatically support claims about the exact value in any single month, the height of a short event, the precise starting date of a sudden change, or the absence of extremes. Those jobs require the underlying observations and a method suited to shorter timescales.
PSLE-Style Transfer Case
A fictional plant-growth monitoring system records the height change of a fast-growing shoot every day. The raw daily changes are:
| Day | Daily change (mm) |
|---|---|
| 1 | 2 |
| 2 | 2 |
| 3 | 8 |
| 4 | 2 |
| 5 | 2 |
A three-day running mean is plotted. A student says, “The plant never had an 8 mm growth day because the running-mean graph never reaches 8 mm.” Evaluate the statement.
Reasoned answer: The statement is not supported. The original Day 3 observation is 8 mm. The running mean combines neighbouring daily changes and therefore produces a lower averaged value around the spike. The smoothed graph describes local averages, not the maximum direct daily observation.
Practice Lab 1: Name the Object
A graph label reads “5-day moving average”. Is each plotted point necessarily one measurement made on that date?
Answer: No. Each point is normally calculated from a window of several observations. Check the stated method to know which observations contribute.
Practice Lab 2: Window Choice
Which is more likely to preserve a one-day peak: a three-day mean or a 30-day mean?
Answer: The shorter three-day window is more likely to preserve more of the peak’s size. A 30-day window spreads one day’s influence across many observations.
Practice Lab 3: Claim Repair
A report says, “Pollution was always below 20 units,” but the only graph shown is a 14-day running mean below 20. Improve the conclusion.
Answer: “The 14-day running mean remained below 20 units during the displayed period.” Individual daily readings need separate evidence.
Practice Lab 4: Same Data, Different Picture
Two websites use the same raw series. One uses a seven-day moving average; the other uses a 90-day moving average. Why might the second look calmer?
Answer: Each 90-day point combines far more observations, so short-term changes have less influence on the displayed value.
Practice Lab 5: End of the Graph
Raw monthly data continue through December, but a centred 12-month smoother ends earlier. Does that prove later data are missing?
Answer: Not necessarily. A centred smoother may require later neighbouring months that are not yet available for the final points.
Delayed Independent Return
Later you meet a dashboard showing a seven-reading rolling average of river level. You do not need to remember this article’s temperature example. Ask four fresh questions: What is directly measured? How many observations enter each displayed point? Where is the window placed in time? Does the claim concern a short event or a longer-term average?
If you can answer those questions, the skill has transferred. The important habit is not “moving averages are bad”. It is “processed representations must be read according to the operation that created them”.
Routes to Existing Canonical PSLE Science Owners
For raw-versus-fitted graph evidence, see PSLE Science Reality Lab Vol No.033 | “The Line Goes Up” — Do the Individual Results Actually Follow It?. For cumulative totals versus rate, see Reality Lab Vol No.082. For gaps between observations, route to How to Read PSLE Science Data With Gaps Without Inventing What Happened Between Measurements.
Parent and Tutor Teaching Guide
Write seven numbers on paper: 10, 10, 10, 22, 10, 10, 10. Ask the child to circle the largest direct observation. Then calculate three-number averages centred on the middle positions. Ask again: “Did the 22 disappear from history, or did our new representation answer a different question?”
Next make two versions of the same data: one three-point mean and one five-point mean. Do not teach “short is right, long is wrong”. Ask which representation would be more useful for finding a one-day event and which would be more useful for describing the background level.
Finally, give the learner a claim before showing the data: “There were no sharp spikes.” Let the child decide what evidence object is needed. The strongest answer should request the raw series or a short enough timescale rather than relying on a heavily smoothed curve.
Authoritative Sources
- Singapore Examinations and Assessment Board — 2026 PSLE Science syllabus
- Singapore Ministry of Education — 2023 Primary Science Teaching and Learning Syllabus
- NOAA Physical Sciences Laboratory — Running Mean Smoother guidance
- NOAA Physical Sciences Laboratory — Climate Indices time-series tool
- NASA GISS Surface Temperature Analysis — running-mean display options
Quiet Return
A smooth scientific line can be extremely useful. Its calm appearance is not the problem. The problem begins when we forget what was averaged to make it.
When you see 12-month running mean, translate the label into a question: Which twelve measurements built this point, and is my claim about those averages or about the raw observations underneath them? That small question keeps the evidence attached to the method that produced it.
