EDUKATE SINGAPORE · GOLD STANDARD PRACTICAL CASE STUDY 188
The mean can rise while most people’s pay falls.
One large executive raise lifts the average pay figure of a fictional ten-person publisher, despite reductions for nine workers. This worked example separates mean, median, payroll cost and individual experience, explaining why an accurate statistic may still support a misleading headline.
All amounts and persons are invented; no real employer or labour-law outcome is assessed.
Quick figures
- Old payroll: $47,000 for ten workers.
- New payroll: $55,200 for ten workers.
- Mean: $4,700 rises to $5,520.
- Median: $3,000 falls to $2,800.
- Employee experience: nine of ten receive less.
- Decision: report both levels instead of a universal improvement claim.
Chapter guide
- 1. The headline and nine workers’ pay packets tell different stories
- 2. Calculate the original mean from the actual payroll population
- 3. The updated mean rises by 17.45%
- 4. The median and individual count reveal the other trend
- 5. OECD and MOM say averages need distributional context
- 6. The report’s implication needs a separate evidence test
- 7. Comparable figures require comparable people and pay definitions
- 8. A correction should not expose private salary records
- 9. A useful dashboard keeps multiple measures visible
- 10. Practice: a rising mean and a falling median
1. The headline and nine workers’ pay packets tell different stories
A fictional ten-person education publisher reports that average monthly salary rose by about 17%. Finance verifies the total payroll and finds the arithmetic correct. But individual records show that nine employees experienced a fall from $3,000 to $2,800, while one executive’s monthly amount rose from $20,000 to $30,000.
An arithmetic mean can increase when a large rise for one high earner outweighs declines experienced by many other people. The Gold Standard issue is not the accuracy of the mean; it is the unsupported leap from an aggregate number to the statement that ordinary staff became better off.
2. Calculate the original mean from the actual payroll population
Nine employees at $3,000 cost $27,000, and one at $20,000 brings total monthly pay to $47,000. Dividing the total by the ten employees gives a mean of $4,700. Nobody in this fictional workforce actually receives $4,700, even though the average is calculated correctly.
The mean answers a legitimate question about total compensation divided by a defined headcount. It does not automatically tell a reader what the middle worker earns. A payroll report should identify the amount included, the period and the population, because a bare average may be easily misunderstood.
3. The updated mean rises by 17.45%
The nine employees’ later pay totals 9 × $2,800 = $25,200. With the executive’s later $30,000, total monthly pay is $55,200. Divide by ten to obtain $5,520. The difference from the original mean is $820, or approximately 17.45% of $4,700.
Calling that a rounded 17% mean increase is not mathematically false. But it is not evidence that most employees enjoyed a raise. A report describing an average must not borrow the everyday meaning of ‘typical person’ when the actual distribution strongly differs.
4. The median and individual count reveal the other trend
With ten ordered pay figures, the median lies halfway between the fifth and sixth. Both middle amounts were $3,000 before the change, so the old median was $3,000. After the change both are $2,800, making the new median $2,800, a fall of about 6.67%.
Nine of the ten workers earn less in the example, so 90% have lower nominal pay. The mean goes up, the median goes down and most individual amounts fall. All three conclusions can be simultaneously correct because they answer different questions.
5. OECD and MOM say averages need distributional context
The OECD wage indicator explains that means can be influenced by very high or low earnings and may not reflect the experience of many workers. Singapore’s Ministry of Manpower income guidance distinguishes the middle worker’s median income from mean income measures.
Those official sources do not describe the fictional publisher’s pay records. They explain why choosing a statistic requires a question: payroll budget per employee, the typical worker’s pay, or the fraction whose pay rose or fell. One indicator cannot accurately answer them all.
6. The report’s implication needs a separate evidence test
The publicity slide reads ‘Average staff pay improved 17%; our employees are better rewarded.’ Its first clause is narrowly supportable with the selected figures; the second suggests a broad benefit to staff. The latter is unsupported because nine of ten people receive lower nominal monthly amounts.
A bounded correction can preserve the legitimate aggregate finding while explaining the median and individual trend. It must not infer that any wage change was lawful, unlawful or agreed to without the relevant contracts, applicable employment rules and actual circumstances.
7. Comparable figures require comparable people and pay definitions
This invented example holds the population constant at ten and uses the same pay component in both months. Real employment data can change with headcount, hours, part-time versus full-time status, bonuses, allowances and the entry or exit of senior staff. Those differences may influence what a mean comparison can establish.
Analysts should preserve the actual pay definition, population, dates and any legitimate adjustments before making a claim. GS-R02 defines consistent acceptance criteria; GS-R08 asks whether the calculation supports the exact statement being published.
8. A correction should not expose private salary records
Finance can preserve and verify individual records under proper access controls while the public report uses suitable aggregate information. A small organisation should consider whether an overly detailed breakdown could identify people whose remuneration is confidential.
Management should not turn a statistical correction into an accusation against the single executive or the nine other employees. The reporting issue is the meaning of the aggregate claim. Questions about employment arrangements require their own authorised governance review.
9. A useful dashboard keeps multiple measures visible
The corrected view can report total monthly payroll, mean pay, median pay and the fraction of employees whose amounts increased or decreased. Each remains labelled with its precise definition. A changing employee population should be separately disclosed rather than compared as though it contains exactly the same people.
GS-R07 connects finance data with communication, and GS-R12 requires review after material changes in workforce or payroll rules. A credible system allows the numbers to tell different but compatible truths instead of forcing every indicator to show improvement.
10. Practice: a rising mean and a falling median
Exercise. Nine people earn $3,000 each and one earns $20,000. Later the nine earn $2,800 and the executive earns $30,000. Calculate the mean, median, total cost and direction of individual pay changes in each period.
The mean rises from $4,700 to $5,520, about 17.45%; the median falls from $3,000 to $2,800, about 6.67%. Total pay rises from $47,000 to $55,200 while nine of ten people earn less. The Gold Standard lesson is to preserve the distributional context before claiming a workforce benefit.
FAQs
How can the mean rise while most workers earn less?
One large high-earner increase can outweigh multiple smaller reductions in the payroll total.
Which measure describes the middle worker?
The median, subject to a clear and consistent population definition.
Is the mean calculation an error?
No. It is correctly calculated in this fictional example; the broad conclusion is unsupported.
Does a lower salary establish an employment-law breach?
Not without contractual, legal and procedural evidence beyond these illustrative amounts.
Would inflation change the interpretation?
It affects real purchasing power, but the current example explicitly compares nominal amounts.
What should an auditor check?
Payroll total, number of workers, pay definition, compared dates, the ordered median and the exact claim made to readers.
Source-to-claim checklist
- Verify total payroll and headcount.
- Use one pay definition across compared periods.
- Calculate the mean and median separately.
- Count workers whose pay rose or fell.
- Identify changes in hours and workforce composition.
- Preserve personal confidentiality.
- Approve conclusions that follow from the measured indicator.
Why nobody in the example actually earned the mean
Before the salary changes, the mean of $4,700 is not an amount paid to anyone. Nine people earn $3,000, and one earns $20,000. The mean is a property of the whole ten-person payroll, calculated by spreading the total cost equally for analytical purposes. A reader who hears ‘the average employee earns $4,700’ could imagine a representative person receiving that amount. No such person exists in the fictional distribution. That does not make the average wrong; it makes the implied individual portrait misleading. Good economic writing specifies whether ‘average’ refers to an arithmetic total, a middle person or a particular role, and it avoids switching meanings halfway through a sentence.
Why the single executive raise dominates the change
Nine reductions of $200 each lower total monthly pay by $1,800. The senior executive’s $10,000 increase more than offsets them. The net change is $10,000 minus $1,800 = $8,200, exactly the difference between $55,200 and $47,000. Divide the net increase by ten people and the mean gains $820. The figures explain the apparent paradox rather than merely asserting that extremes can matter. One concentrated change affects the sum strongly, while the middle two values in the ordered distribution still decrease. The Gold Standard requires the explanatory bridge from arithmetic to human meaning whenever a headline suggests shared improvement.
The median is not another kind of arithmetic mean
For a ten-person population, the median is found by ordering all values and averaging the fifth and sixth positions, rather than dividing the full payroll by ten. The two middle salaries are $3,000 before the change and $2,800 afterward. The unusually high salary is still in the distribution, but its numerical distance from the middle values does not directly pull the median upward. The median answers a different question: what amount sits in the middle of the ordered values? It can be useful for understanding a typical worker while still omitting details about the tails, hours and contractual arrangements.
What happens when the workforce itself changes
The fictional example deliberately holds the workforce constant. Imagine instead that a company hired two senior executives or moved many workers to part-time employment between the periods. Its average might change simply because the people and hours counted changed. Comparing total remuneration across different populations then requires additional explanation. A consistent payroll definition can still produce a number that answers a changing-composition question rather than the wage growth experienced by the same employees. A careful report identifies whether it is comparing all employees in each period or following a stable cohort, and it should not claim one interpretation from evidence collected for the other.
An hourly measure can answer another legitimate question
A worker paid $2,800 for fewer monthly hours may have experienced a different hourly-rate movement from someone paid $2,800 for unchanged full-time hours. Conversely, a person whose monthly amount rose may have worked substantially more hours. None of those circumstances is established for the fictional publisher; the case fixes the monthly salary figures to make the mean–median contrast understandable. In real payroll analysis, appropriate hours and employment-status information may be essential. A Gold Standard comparison states what the numerator and denominator include, rather than assume that any monthly increase necessarily reflects improved rates or working conditions.
Nominal earnings and purchasing power are different
The figures in this case describe nominal dollars. Inflation-adjusted income asks another question: how much can a worker purchase at the later period’s prices? Even a nominal wage increase might fail to preserve purchasing power if relevant prices rise sufficiently, while a nominal decline need not fully describe changing household circumstances. A true real-income comparison needs a justified price measure and clear dates. We do not invent an inflation rate here or describe the company’s pay pattern as a national wage trend. A careful publication can say that the median nominal amount fell while reserving broader economic welfare judgments for better evidence.
How a truthful correction should sound
An effective statement could read: ‘For the same ten workers under our selected monthly pay definition, mean pay increased from $4,700 to $5,520, approximately 17.4%. Median pay decreased from $3,000 to $2,800, approximately 6.7%, while nine workers received lower nominal amounts.’ The sentence reports the apparently conflicting statistics together. It does not claim all staff were better rewarded, that the company broke employment law or that one individual deserved their salary. The Gold Standard supports plain language that preserves numbers, units, groups and uncertainty rather than a reassuring phrase chosen before the analysis.
Privacy is part of credible human-resources reporting
In a company with just ten employees, breaking salary statistics down into narrow roles can make an individual easy to identify. The publisher may need to disclose aggregated patterns to authorised reviewers without publishing personal payroll records to everyone. This case openly uses fictional individual amounts to teach the mathematics; an actual employer should apply its legitimate data-handling and confidentiality rules. Transparent reporting does not require indiscriminate release of identifiable records. The management system can retain a verifiable audit trail under appropriate access controls while explaining mean, median and their limits to readers through meaningful aggregate categories.
Distinguish the reporting defect from the wage decisions
The salary reductions themselves raise potentially important governance questions, but the statistical example does not contain employees’ agreements, contract terms, governing law or reasons for those changes. A reviewer should therefore avoid treating the misleading headline as proof that every underlying wage decision was unlawful or improper. The reporting decision is assessable on the available evidence: an aggregate statistic was used to imply broad worker benefit contrary to the distribution. The legality and fairness of any actual change are separate questions for qualified employment and governance review. The Gold Standard prevents one legitimate concern from being exaggerated into allegations for which evidence is absent.
An audit should reconstruct, not just repeat, the spreadsheet
The independent reviewer begins with the original ten salary records and confirms that each belongs to the accepted comparison population. They recompute old and new totals, means, medians, counts of rising and falling amounts and the percentage bases. The reviewer then reads the proposed public claim, rather than assuming that correct formulas guarantee truthful communication. A spreadsheet can return $5,520 without recognising that ‘all staff gained’ is unsupported. An audit reaches the actual claim-to-evidence connection, checks what remains uncertain and records the authorised correction without altering the original pay data merely to produce a tidier result.
A first-principles teaching activity with ten cards
Give students ten paper cards with the original pay figures and ten with the later figures. Ask them to sort each set, identify the fifth and sixth amounts, and independently add the totals. Before revealing the results, invite a prediction: must a higher mean mean that most workers were paid more? The cards show why the answer is no. Students can identify the one extreme increase that dominates the total while the middle of the distribution shifts downward. The exercise practises percentages, ordering, averages and logical inference without using confidential real payroll information.
A leadership dashboard should support more than one purpose
The finance team may legitimately use mean compensation per employee to assess labour cost and budgeting. HR may use median and pay-distribution measures to understand the middle of the workforce. A governance committee may examine contract terms or worker outcomes through separate appropriate evidence. A public report may explain the same measures to non-specialists. The dashboard should not force these different users into one generic green ‘salary growth’ box. It should label metrics, disclose what they omit and make significant changes in population or definition visible before people act on the figures.
The error can recur without any spreadsheet fault
Imagine the company refreshes its spreadsheet every month and every formula is correct. A communications template still automatically generates ‘Our people are better paid’ whenever the mean rises. The workflow would continue producing unsupported claims even though no calculator ever misbehaved. The corrective action therefore belongs at the reporting rule and approval stage, not only at arithmetic validation. The person approving the headline needs to understand which statement is authorised by which statistic. A repeat test should deliberately feed the reporting system a rising mean and falling median to ensure it flags the misleading stock phrase.
A second worked comparison shows why percentages need the right base
The new mean of $5,520 exceeds the old $4,700 by $820. Dividing the increase by the original $4,700 gives about 17.45%. If someone instead divides the change by the new $5,520, they obtain a different fraction answering a different question. Similarly, the median decrease of $200 is 6.67% of the original $3,000, not 6.67% of the new $2,800. A Gold Standard report names the starting point and the direction of change rather than publish a percentage with an invisible denominator. These details matter whenever an executive summary compresses a large dataset into one sentence.
How to test the repair with a second synthetic dataset
Create fictional salary records where one very high value rises and most ordinary values fall. The quality reviewer should be able to produce the mean, median and distribution correctly, and the communications workflow should refuse an unsupported universal-benefit headline. Then change headcount between periods and see whether the system requests a comparison-population note. Finally, introduce an example with bonuses in one period but not the other, which should trigger a definition check. The Gold Standard control passes when a new reviewer can recognise what the data establish without relying on the memory of the original analyst.
The final case-study conclusion can therefore retain two apparently contradictory observations: payroll arithmetic shows an increase, while the typical and most individual workers’ pay records show decreases under the same fictional comparison. Neither should be hidden merely because the other is less encouraging. The number is useful only when a reader can tell whose experience, what period and which definition it represents. That is the practical meaning of evidence-led management.
This method transfers to educational test scores, fundraising totals, service response times and any other aggregate where extreme values or changing populations can dominate a headline. A strong Gold Standard answer begins by naming the unit of analysis and the legitimate question before selecting a mean, median or percentage.
Methodological sources and casebook connections
- OECD · Average Annual Wages
- MOM · Income Summary Table
- MOM · Median Income Introduction
- Gold Standard · Evidence and Verification
- Gold Standard · Core Requirements
- Case 178 · Payroll Was Correct but Wages Arrived Late
- Gold Standard Casebook
MOM and OECD provide statistical principles, not these fictional salary amounts. GS-R01–GS-R12 is eduKate’s educational management reference.