Did you know that the strangest point on a box plot may be the most important measurement in a school investigation? It is tempting to remove a dot beyond the whiskers because it looks wrong. But a chart can identify a surprising value without proving that the value was recorded incorrectly. This distinction turns graph reading into genuine scientific judgement.
The immediate answer: a box plot outlier is an unusually distant observation under the chart’s chosen rule, not automatically a mistake. Investigate the original record, measurement method, units, conditions and comparison group before correcting or excluding it. Retain a valid observation, explain its effect on the summary, and label any justified exclusion transparently. The aim is trustworthy evidence, not a prettier graph.
- A box plot shows a central distribution using a median, quartiles and whiskers, sometimes with separate outlier markers.
- Check which quartile and whisker convention the worksheet or software uses.
- A common rule flags values below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR.
- Flagged does not mean false: verify the raw record and the reason for the unusual result.
- Compare mean and median, record any correction, and keep the original unaltered dataset.
- Progress from a guided eight-value example to a fresh independent check.
What an outlier marker actually means
A box plot compresses many values into a visual summary. The line inside the box is the median; the box ends show the first and third quartiles, often written Q1 and Q3. Their difference, Q3 − Q1, is the interquartile range, or IQR. In a familiar Tukey-style convention, points farther than 1.5 times the IQR outside either quartile are shown separately. Not every school textbook or software package calculates quartiles and whiskers in precisely the same way, so the legend matters.
The rule identifies an observation for closer attention. It cannot tell whether someone typed 30 instead of 13, whether a stopwatch malfunctioned, or whether a pupil truly needed longer to finish. To answer those questions, you need the evidence behind the graph. The NIST Engineering Statistics Handbook on outlier detection distinguishes potential data errors from genuine unusual behaviour and cautions against simply deleting uncertain points.
A carefully checked example: eight reading times
Imagine a fictional school reading task in which eight students each complete the same short exercise once. The times, recorded to the nearest whole minute and sorted from shortest to longest, are 8, 9, 10, 10, 11, 12, 13, 30 minutes. These are invented teaching values, not observations of real students. Assume no pupil was interrupted and that the recorded figures are correct until a record check suggests otherwise.
Using the median-of-halves quartile convention for this even-sized dataset, the median is (10 + 11) ÷ 2 = 10.5 minutes. The lower four values are 8, 9, 10 and 10, so Q1 = (9 + 10) ÷ 2 = 9.5 minutes. The upper four values are 11, 12, 13 and 30, so Q3 = (12 + 13) ÷ 2 = 12.5 minutes. Thus IQR = 12.5 − 9.5 = 3 minutes.
Calculate the usual fences: lower = 9.5 − 1.5 × 3 = 5 minutes; upper = 12.5 + 1.5 × 3 = 17 minutes. The observation of 30 minutes lies above the upper fence and would be flagged as an outlier. Under this convention, the whisker normally reaches the largest non-outlying observation, 13 minutes, while 30 minutes is marked separately. A different software implementation may use a different quartile definition, so always match the stated method.
Why deletion changes the story
The total time for all eight observations is 8 + 9 + 10 + 10 + 11 + 12 + 13 + 30 = 103 minutes, giving a mean of 103 ÷ 8 = 12.875 minutes. If 30 minutes were deleted without evidence, the remaining total would be 73 minutes across seven observations, giving 73 ÷ 7 ≈ 10.43 minutes. The mean falls by about 2.45 minutes. The graph and the conclusion would now conceal a real difficulty if that final student genuinely needed the longer time.
The median of all eight values is 10.5 minutes. Without the final observation, the seven-value median is 10 minutes. That smaller movement illustrates why the median can be less sensitive to an extreme observation than the mean, not why one statistic is universally superior. Choose a summary based on the question being asked and show the distribution when it matters.
Four questions that diagnose the mistake
- Was the value transcribed from the original sheet correctly, or is there a mismatch between 30 and 13?
- Were all times measured using the same start and stop rules, the same units and comparable task conditions?
- Was 30 minutes a real completion time, a timeout limit, an interruption or a missing value disguised as a number?
- Would retaining the value, alongside a median and a full-range note, produce a more truthful explanation than removing it?
A tutor can ask a learner to point to the exact record that would justify changing the dot. If the learner can only say, ‘it is too far away’, the diagnosis is clear: the learner is mistaking a statistical alert for a fact about data quality. Explain the difference, then practise with a documented error and a genuine unusual observation side by side.
Guided practice: verify before choosing
Suppose the source worksheet contains ’13 min’ for the final learner, but the spreadsheet contains 30. This is evidence of a transcription mistake. Correct the spreadsheet to 13, preserve a note of the change and recalculate the chart. In a different fictional case, the worksheet and digital timer both confirm 30 minutes and the conditions were comparable. Keep 30. It may represent a pupil who needs more time, a genuine response to the task or an unmeasured factor that deserves discussion.
These two decisions differ not because one chart looks less attractive, but because the provenance differs. In a classroom investigation, write an audit note: ‘Original record checked; 30 minutes confirmed; retained as a valid long completion time.’ For a documented typo, write: ‘Spreadsheet value corrected from 30 to 13 minutes after checking dated paper record.’ Do not silently overwrite the original file.
Independent practice: a new set of values
Try the same method with seven fictional times: 6, 7, 8, 9, 9, 10, 23 minutes. Under the median-of-halves convention excluding the middle value when n is odd, the median is 9, Q1 is 7, Q3 is 10 and IQR is 3 minutes. The upper fence is 10 + 1.5 × 3 = 14.5 minutes; 23 minutes is flagged. Now write two different conclusions: one for a confirmed transcription error, and one for a valid 23-minute observation. If both conclusions simply say ‘delete 23’, the reasoning is not finished.
Progress you can actually observe
A useful progress check is whether the learner can explain, without being prompted, the difference between flagging an outlier and excluding it. On the first attempt, they might identify the unusual number but have no evidence rule. On the next, they calculate the fence, check the raw value, state their inclusion decision and describe what happens to the mean. This is observable improvement in method, not a claim about future examination scores.
Parents can ask, ‘Where did that number come from?’ rather than, ‘Can you make the average better?’ Teachers can give two almost identical graphs with different supporting record notes. Students should reach different decisions because the evidence differs. The teaching follows the diagnosis, explanation, supported practice and independent checking sequence emphasised in eduKate’s Clementi Secondary 1 Mathematics tutorial reference.
Common misconceptions and limits
A point outside a whisker is not automatically an invalid measurement. A point within the whiskers is not guaranteed to be accurate. The 1.5 × IQR convention is a useful screening method, not a universal test of data truth, causation or fairness. With small samples, quartile definitions can noticeably change the display. With repeated measures from the same individual, independence also matters: a set of observations is not always a set of independent people.
For very small or skewed datasets, show individual values in addition to a box plot. For serious research, use a pre-declared data-quality policy, check instrument logs and consider robust analysis when warranted. A classroom box plot should never stand in for an investigation of the measurement process.
Frequently asked questions
Does the dot beyond the whisker mean that the student was careless?
No. It marks an unusual value according to a plotting convention. Carelessness is only one possible explanation, and it cannot be inferred from the dot alone.
Should I calculate the mean with the outlier?
Normally begin with the complete valid dataset. If the task asks you to compare a sensitivity analysis, show results both with and without the point and say exactly why. Never pretend that a filtered result is the unfiltered mean.
What if two graphing programs produce slightly different box plots?
Check their quartile, whisker and outlier rules. Software may use different conventions. For an assigned exercise follow the convention specified in the question; for a report disclose the one used.
Helpful next steps and sources
For the distinction between whiskers and individually drawn points, read How to Read Box Plot Whiskers When Outliers Are Marked Separately. To compare two classes fairly, continue with How to Compare Box Plots When Groups Have Different Numbers of Pupils. The NIST outliers guide provides the technical rationale. The eduKate Education Hub connects this skill with other learning routes.
The lesson to keep: identify the unusual value, find the supporting evidence, then decide what it means. The core aim is not to force data into an expected pattern. It is to learn when a surprising point should change our understanding.