Did you know that a scatter plot can show four dots even when a class has recorded eight measurements? Nothing has necessarily disappeared from the data. Several observations may sit in exactly the same position, leaving one printed mark on top of another. If a learner mistakes visible marks for individual people, a perfectly reasonable chart can produce the wrong answer.
Overplotting happens when two or more data observations share the same plotted coordinates or are drawn too close together to distinguish. To count participants, use the original rows or a frequency table, not just the number of visible dots. A larger symbol, a count label, a carefully jittered display or a transparent point style can help reveal overlapping values, but these are display choices; they must not quietly alter the actual measurements.
Here is a diagnosis-first guide to reading scatter plots, a foundational data-literacy skill for projects involving sleep, exercise, study habits, rainfall, height or any paired measurements. All pupils, scores and study-hour records below are invented for teaching, not measured outcomes, tuition testimonials or claims about what causes achievement.
First ask the question that catches the misunderstanding
Present a small scatter plot showing four separate marks and a caption stating that eight paired observations were collected. Ask, ‘Does this prove the caption is wrong?’ A learner who immediately says yes may be equating ink on the page with records in the source. A learner who says eight must be correct simply because the caption says so is trusting a label without checking evidence. Both children need a method that can test the claim.
Begin with a simpler diagnostic. Draw two identical dots at the same position, one in blue and one in black, on transparent paper. Put one sheet over the other. From above, there is one visible position but two observations. The missing information is not a mathematical error; it is a display limitation.
A complete worked example: four marks, eight students
Imagine a class survey recording rounded weekly study hours on a horizontal axis and an invented short quiz score out of ten on a vertical axis. Each row is one student’s paired answer. Assume there were eight distinct eligible students, one complete response each, and the study hours were rounded to whole hours. Here is the complete fictional dataset:
- Learner A: (1 hour, 5 marks); learner B: (1 hour, 5 marks).
- Learner C: (2 hours, 6 marks); learner D: (2 hours, 6 marks); learner E: (2 hours, 6 marks).
- Learner F: (3 hours, 7 marks).
- Learner G: (4 hours, 8 marks); learner H: (4 hours, 8 marks).
The eight rows occupy only four distinct coordinates: (1,5), (2,6), (3,7) and (4,8). A conventional plot that draws identical opaque dots at each coordinate may show only four visible symbols. Count the underlying responses instead: 2 + 3 + 1 + 2 = 8 learners. Nothing in the data implies that four learners were absent. The overlap is a property of how points were drawn.
An independent check is to count the row identifiers A through H: there are eight. A second check is to group identical (x,y) pairs and add their multiplicities. Each method asks a different question from visually counting the four ink marks.
Why the axes and denominators still matter
One coordinate is a pair, not two separate pupils. The point (2,6) means two reported hours paired with six quiz marks for one observation; it does not mean two pupils plus six pupils. Labels and units are necessary to interpret the position. A set of eight complete pairs supports a sample size of eight; it does not follow that all students in a school took part.
If a class plotted average scores by group rather than individual scores, then one point might represent an entire group for a different reason. Before declaring any hidden observations, ask what one row of the underlying data represents. Plot meaning begins with the record unit, not with its shape.
Four ways to make overlapping observations visible
1. Count from the table, then label a dot
The safest beginner method is a frequency table with the coordinate and its count. Our example would read ‘(1,5): 2’, ‘(2,6): 3’, ‘(3,7): 1’, ‘(4,8): 2’. The plot can display 2, 3, 1 and 2 beside the relevant locations, with a legend that says each number represents observations, not marks scored. This provides an auditable bridge from data to picture.
2. Use bubble size carefully
One option is to make each dot’s area or size respond to the number of identical observations. The ggplot2 geom_count reference describes a plotting method that sums observations at the same position and maps count to point size. Readers still need a legend and should avoid assuming that a dot twice as wide represents only twice as many records: width, diameter and area are not the same thing.
3. Add transparency or modest jitter
If points are drawn with transparency, multiple overlapping marks can sometimes appear darker. Another option, called jittering, places the symbols slightly apart to disclose density. The ggplot2 jitter reference explains this small display displacement. Always say that a jittered symbol’s new visible position is not a new measured value. In our example, learner A and learner B both truly reported one hour and five marks even if their displayed symbols appear a few millimetres apart.
4. Report the count in the caption
A caption such as ‘Eight student responses; identical pairs overlap’ prevents a reasonable reader from being misled. Add the sampling date or the fact that the values were rounded if this matters. A caption is a useful safeguard but cannot replace access to the frequency table when the task requires exact counting.
How learners move from guided reading to independent checking
Guided stage: A teacher gives six labelled row cards and asks students to place them on a coordinate grid. When several cards land on the same square, students stack them and write the count beside the stack. Before showing a graph, ask them to predict how many visible squares will be occupied and how many total cards exist.
Partner stage: One learner constructs a frequency table, while a partner creates a plot using the same rows. Each writes a separate total. The pair must explain any difference before comparing answers. Encourage the question, ‘Are we counting records, unique values or visible marks?’ The words matter as much as the arithmetic.
Independent stage: Give students a new data table and a plotted chart without a stated sample size. Ask them to calculate the number of source observations, count distinct coordinates, explain any difference and recommend one accessibility change. Then ask them to find an independent check, such as counting unique record IDs or summing a frequency column.
Try these problems, then check the answers
Practice A: A fictitious sports club records 12 paired observations. The coordinate counts are 5, 2, 4 and 1. How many separate plotted locations may appear, and how many observations exist? Answer: four locations and 5 + 2 + 4 + 1 = 12 observations. A plot can show four dots without losing eight entries.
Practice B: A science group has ten complete records, but three share the same coordinate; all others are at different coordinates. How many distinct coordinates are there? Answer: 1 for the shared triple and 7 for the remaining individual records, giving 8 distinct coordinates.
Practice C: Two rounded responses sit at (2,6). A graph jitters them visually to (1.9,6.1) and (2.1,5.9). Have the pupils’ original records changed? Answer: no. The displayed positions were shifted for readability. A graph must make the distinction explicit, and analysis should use the original pair (2,6) twice.
Stretch: If 20 data rows came from ten students measured on two days, is ’20 students’ an accurate caption? No. There are twenty observations but ten individuals, assuming each attended both days. The unit of analysis needs to be stated. Overplotting and repeated-measure design are two different issues.
Misconceptions that cause wrong conclusions
- ‘One visible dot equals one learner.’ It equals one visible position unless the plotting rule explicitly represents individuals with no overlap.
- ‘Every repeated response must be a duplicate form error.’ Different people can legitimately report the same rounded numbers. Investigate record identity before deleting anything.
- ‘Jitter increases precision.’ It changes display positions and may suggest false decimal accuracy if unexplained.
- ‘A larger bubble always represents more successful pupils.’ The bubble might represent frequency, class size or another variable; inspect the legend.
- ‘A positive-looking pattern proves more study caused a higher score.’ Our data were invented, small, rounded and not a causal experiment.
- ‘A sample of eight can speak for every school.’ Generalisation requires much more than counting visible symbols correctly.
A usable checklist for parents, teachers and student researchers
Parents helping with a school graph can ask, ‘What does one row mean?’, ‘What does one dot mean?’ and ‘Could more than one row sit there?’ These three questions expose many graph-reading mistakes without turning the conversation into a lecture. If the learner counts correctly but cannot explain the meaning of the plotted pair, go back to axes and units before increasing complexity.
Teachers can assess observable progress across three attempts: whether the child preserves every original record, makes a correct frequency summary, and writes a cautious one-sentence conclusion. A useful performance indicator is that the learner can explain why four visible points represent eight observations without needing a prompt. Check whether the skill transfers to a different context with duplicate rounded values, rather than reusing identical questions.
This connects with the Core Aim guide to preventing double-counting in frequency classes. In one case, the danger is counting a row twice; in the other, it is failing to count stacked rows at all. The mathematics diagrams guide offers a broader pathway through data representation.
Frequently asked questions
Can I count dots if I know there was no rounding?
You can count visible positions, but distinct exact values still might coincide. Without knowing how the graph was generated or viewing source records, do not assume each dot is one record.
Should I always jitter a scatter plot?
No. Jitter can make duplicates apparent, but it can also obscure meaningful exact coordinates or ranges. A count label or a linked frequency table may be more honest for small educational datasets.
Can one point mean multiple students in a group average?
Yes. In that case the point represents an aggregate by design, not necessarily an accidental overlap. Read the caption and identify what the dataset calls an observation.
What if the source data are unavailable?
State the limitation. You may say ‘four distinct visible positions’ but should not claim ‘four pupils’ or silently infer a total sample size.
The larger educational idea: a chart is evidence, not the whole dataset
An excellent learner learns to travel back and forth between a representation and the observations it claims to represent. Counting records, checking assumptions and explaining uncertainty are more valuable than making an attractive chart alone. Explore further through the eduKate Education learning hub and the diagnosis-first Secondary 1 Mathematics learning reference. The most useful sentence a student can write is wonderfully simple: ‘I checked the data rows before I counted the dots.’