Wait, what? A scientific scatter plot shows 80 visible dots. The caption says the experiment contained 120 measurements. A student points at the screen: “That cannot be right. I can count only 80.”
The student has counted visible marks, not necessarily data records. In a scatter plot, two or more observations can have the same—or nearly the same—horizontal and vertical coordinates. Their markers may be drawn directly on top of one another. One visible dot can therefore hide several records.
This Reality Lab owns one evidence-transfer job: when a scatter plot looks as though it contains a certain number of measurements, check whether overlapping markers make the visible-dot count smaller than the record count. The plot can still be scientifically useful. The learner simply has to distinguish data from the way data are drawn.
Quick Answer
No. Eighty visible dots do not prove there were only eighty measurements. Scatter plots can suffer from overplotting or overdraw: multiple records occupy the same display area and their markers overlap. To know the number of records, use the dataset, caption, sample-size statement, interactive count, transparency, jitter, density display, or another representation designed to reveal overlap.
The durable habit is: a mark on a graph is a representation of data; it is not automatically a one-for-one inventory of records.
The Owned Learner Job — Not Generic Graph Reading
This article does not replace general graph-reading, correlation, axes, trend, sample-size or statistical reasoning. Those remain with existing canonical owners. It applies those skills to one specific communication object: a scatter plot whose overlapping points make record count and local density visually ambiguous.
For the broader distinction between observation and inference, use How to Tell Observation, Inference, Prediction and Explanation Apart in PSLE Science. For deciding which evidence actually answers a claim, use How to Choose the Decisive Evidence for a PSLE Science Answer Without Copying the Whole Table.
Composite Case: 120 Leaves, 80 Visible Dots
A fictional class measures 120 leaves. For each leaf it records:
- leaf length, rounded to the nearest centimetre; and
- leaf mass, rounded to the nearest gram.
The class plots mass on the vertical axis and length on the horizontal axis. Because the measurements are rounded, many leaves share exactly the same pair of displayed values. Four different leaves might all be 8 cm long and 3 g in mass. When the software draws four identical markers at coordinate (8, 3), the markers sit on top of one another.
| Raw record | Length shown | Mass shown |
|---|---|---|
| Leaf 17 | 8 cm | 3 g |
| Leaf 42 | 8 cm | 3 g |
| Leaf 81 | 8 cm | 3 g |
| Leaf 103 | 8 cm | 3 g |
If every marker uses the same size and solid colour, these four observations may look like one dot. The plot has not deleted the records from the dataset. The display has hidden their multiplicity.
Observed, Drawn and Inferred
| Layer | What it contains | What can go wrong? |
|---|---|---|
| Raw data | 120 measurement pairs | Nothing about visible overlap yet |
| Plotting rule | One marker is drawn at each pair of coordinates | Markers can share coordinates |
| Rendered image | Some markers cover others | Visible marks can be fewer than records |
| Viewer inference | “I see 80 dots, so n = 80” | Visible marks are mistaken for record count |
The scientific error occurs at the last step. The plot image is real. The records are real. The mistaken assumption is that every visible marker must correspond to exactly one record.
Why Overlap Happens
Point overlap becomes more likely when:
- many observations are plotted in a small area;
- values are rounded to coarse units;
- variables take only a few possible values;
- markers are large;
- the plot is displayed at small size;
- several groups are plotted on top of one another;
- the dataset is much larger than the number of pixels available to display it.
Peer-reviewed visualization research describes this as overdraw or overplotting. The important student lesson is not the specialist vocabulary. It is the evidence consequence: what you can see may under-represent how many records occupy a region.
Representation Check: What Does One Marker Mean?
Before counting dots, read the caption and plotting method. Ask:
- Does one marker represent one record?
- Can multiple records share a coordinate?
- Are values rounded or binned before plotting?
- Are markers transparent?
- Has jitter been added?
- Are point sizes proportional to count?
- Is density shown by colour or contour instead?
- Were records sampled or filtered before display?
Different plotting choices answer different visual questions. A scatter plot optimised for trend may not make exact record counts visually obvious.
Transparency: A Darker Dot Can Mean More Overlap
One common solution is to make markers partly transparent. If several points overlap, the region becomes darker. This can reveal dense clusters that were hidden by solid markers.
But transparency creates a new reading task. A darker area may indicate more overlapping points, but the exact count depends on opacity, colour blending and rendering. Unless the legend or software provides a direct mapping, “twice as dark” does not necessarily mean “twice as many records.”
This is a useful reminder: a visualization technique can solve one ambiguity while introducing another. Read the encoding rather than guessing from appearance.
Jitter: Revealing Hidden Points by Moving Them Slightly
Another common technique is jitter. The plotting software moves overlapping markers a small random or structured distance so that they can be seen separately.
Suppose four records all have the true displayed coordinate (8, 3). A jittered plot might place them at 7.9, 8.0, 8.1 and 8.1 horizontally. Those slight offsets are a display device. They are not four newly measured leaf lengths.
A learner who reads jittered coordinates as exact measurements makes the opposite error: a visualization added displacement for readability, and the reader mistakes the displacement for data.
Comparison Check: Same Data, Four Different Plots
| Display | What becomes easier? | What caution remains? |
|---|---|---|
| Solid scatter plot | Overall relationship and outliers | Dense overlap can hide records |
| Transparent scatter plot | Dense regions become darker | Darkness is not automatically an exact count |
| Jittered scatter plot | Coincident points become visible | Offsets are display changes, not measured values |
| Density or contour display | High-density regions become clear | Individual records may no longer be directly visible |
No representation is “the truth” in isolation. Each is a tool that preserves some features and compresses others. Scientific visual literacy means knowing which feature the current display makes easy to see.
Baseline Check: Count of What?
A caption says “n = 120.” What does n count?
It might count leaves, repeated measurements, samples, participants, days, test runs or filtered records. The count must be tied to an object. A scatter plot with 120 rows of data may contain fewer than 120 independent biological samples if some rows are repeated measurements. That is a separate ownership question from overplotting, but it shows why sample-size labels need definitions.
For this article, assume each row is one leaf. The only question is whether the visible marker count reveals all rows. It may not.
Method Check: Rounding Can Create Artificial Ties
If a balance measures to 0.01 g but the published plot rounds every mass to the nearest gram, many distinct raw values can collapse onto the same plotted coordinate.
Example:
| Raw mass | Plotted mass after rounding |
|---|---|
| 2.61 g | 3 g |
| 2.83 g | 3 g |
| 3.12 g | 3 g |
| 3.39 g | 3 g |
Four different measurements now share one vertical value. If their horizontal values also tie, the markers can completely overlap. A dense-looking or sparse-looking image therefore depends partly on measurement resolution and plotting choices.
Worked Case 1: The Hidden Stack
A chart contains 50 records but only 41 visible markers. The raw table shows ten records with coordinate (5, 7). One marker at (5, 7) is therefore hiding a stack of ten.
Correct conclusion: visible-marker count is not the record count in this plot. Use the data or a representation that exposes multiplicity.
Worked Case 2: Two Groups Occupy the Same Point
Blue and orange groups are plotted together. An orange point is drawn after a blue point at the same coordinate and covers it. A student concludes only the orange group has a record there.
Too strong. Drawing order can hide one group behind another. Transparency, outlines, interactive hover or separate panels may reveal the overlap.
Worked Case 3: Jittered Points Look Like Measurement Noise
A jittered plot shows five points spread between x = 9.8 and x = 10.2. The raw data say all five had x = 10. A student reports that the measured values ranged from 9.8 to 10.2.
Incorrect. The spread was added for display. The original x-values were tied at 10.
Worked Case 4: Tiny Screen, Same Dataset
The same scatter plot is shown on a desktop monitor and a small phone. More dots appear distinct on the larger display. Did the dataset change?
No. Available screen space changed. Overdraw can be a property of the rendering as well as the dataset.
Worked Case 5: Density Is Not Exact Count
A dark cluster appears in the upper right of a transparent plot. A student says it contains exactly 40 records because it looks twice as dark as a cluster known to contain 20.
Not justified. Darkness depends on marker opacity and overlap. Without a calibrated legend or direct count, visual darkness is qualitative evidence of density, not an exact record total.
Alternative Explanations for “There Are Fewer Dots Here”
- There may genuinely be fewer records.
- More records may overlap at identical coordinates.
- A filtering rule may have removed some records.
- One group may be hidden behind another group.
- The marker size may be too large.
- The chart may have sampled the dataset for display.
- Missing values may prevent some records from being plotted.
The picture alone may not distinguish these explanations. Check the caption, code, dataset or interactive metadata before deciding which is correct.
Evidence That Strengthens a Record-Count Claim
- The caption states the number of records plotted.
- The raw or downloadable data contain the same count.
- Filtering and missing-value rules are documented.
- Transparency or jitter reveals overlapping observations.
- Interactive hover reports multiplicity at tied coordinates.
- A separate frequency table confirms counts in dense regions.
- Group drawing order is documented or separate panels are available.
Evidence That Weakens “I Counted the Dots, So That Is n”
- The plot contains many tied or rounded values.
- Markers are opaque and large.
- The display is small.
- Several categories share one plotting area.
- The caption warns about jitter, sampling or transparency.
- The dataset is far larger than the number of visible marks.
- Dense regions look like solid blobs.
Tempting but Invalid Reasoning
- “One dot equals one record because scatter plots use dots.” Several records can share a dot location.
- “Jittered coordinates are the real measurements.” Jitter may be an intentional display displacement.
- “A darker region contains exactly proportionally more records.” Not without a defined encoding.
- “If a point is not visible, the record was deleted.” It may be covered by another marker.
- “The phone version has fewer data than the desktop version.” Screen resolution can change visible overlap.
How Far Can the Conclusion Travel?
| Claim | Visible dots alone enough? |
|---|---|
| There are at least this many visible marker locations. | Yes. |
| There are exactly this many data records. | Not if overlap is possible. |
| This region contains a dense concentration of records. | Possibly, especially with transparency or density encoding. |
| There are exactly twice as many records because it looks twice as dark. | No, unless the visual scale defines that relationship. |
| Every jittered position is a measured coordinate. | No. |
PSLE-Style Transfer Case: Seed Germination Times
This is an original practice case, not an examination question.
A class records germination day and final seedling height for 60 seedlings. Many seedlings germinate on the same day and finish at the same height rounded to the nearest centimetre. The scatter plot shows only 47 visible dots.
A student says, “Thirteen measurements are missing because 60 − 47 = 13.”
Evaluate the statement.
A strong answer says the visible-dot count does not show whether records are missing. Multiple seedlings may have the same plotted values and their markers can overlap. The student should check the raw table or a display that reveals repeated coordinates before concluding that measurements are missing.
Delayed Independent Return
Later, a map shows only 30 visible symbols but the legend says 45 monitoring stations. What habit should return? Do not immediately accuse the map of an error. Ask whether stations share locations at the display scale, whether symbols overlap, whether clustering is used, and whether the legend reports the underlying record count.
The surface object changed from a scatter plot to a map. The evidence habit survived: visible marks and underlying records are not always one-to-one.
Explained Practice
1. Raw table says n = 100; plot shows 75 marks. What should you check first? Whether multiple records have identical or near-identical coordinates and overlap.
2. Jitter reveals 12 marks around x = 5. Can you conclude their true x-values range from 4.8 to 5.2? No. The offsets may be visual only.
3. A cluster becomes darker when opacity is reduced. What does that suggest? More overlapping marks may occupy that region, but exact count requires a defined scale or data.
4. Why can rounding increase overplotting? Different raw values can collapse to the same displayed coordinate.
5. How do you prove the record count? Use the dataset or an explicit count supplied by the source, not visual dot counting alone.
Parent and Tutor Teaching Guide: Stack Coins on a Coordinate
Draw a simple coordinate grid. Give a learner ten small paper circles representing ten records. Place four circles at exactly the same coordinate. From above, they look like one circle. Ask, “How many marks can you see?” Then lift the stack and ask, “How many records are actually here?”
Next, spread the four circles slightly apart and label the movement “jitter.” Ask whether their measured coordinates changed. The answer is no: only the display changed.
Finally, use semi-transparent paper or light pencil strokes so overlap becomes darker. Ask what darkness tells you and what it does not. The learner should say it helps reveal density but does not automatically give an exact count.
Current PSLE Science Frame
The 2026 PSLE Science assessment is based on the 2023 Primary Science syllabus. Current assessment objectives include interpreting and analysing information, evaluating observations, information and methods, and communicating explanations and reasoning. MOE also frames science learning around healthy scepticism and evidence-based reasoning. Overplotting is a useful real-world transfer object because the learner must distinguish the underlying observations from the limitations of their visual representation.
Authoritative Sources
- NIST/SEMATECH Dataplot — Scatter Plot: defines the scatter plot as a display of paired variables and its use for relationships, variation and outliers.
- Mayorga & Gleicher — Splatterplots: Overcoming Overdraw in Scatter Plots: peer-reviewed visualization research explaining how overlapping glyphs can obscure data density and records.
- Park et al. — ATOM: A Grammar for Unit Visualizations: discusses overplotting when two or more marks partially or completely overlap.
- SEAB — 2026 PSLE Science syllabus.
- MOE — 2023 Primary Science Teaching and Learning Syllabus.
Quiet Return
Scatter plots compress many observations into a small visual field. That compression is useful because patterns become visible. It also means one mark can hide another.
Count records with the data. Read patterns with the plot. Do not assume those two jobs always produce the same number.
