Series ID: EDKS-ADV-VOC-ZH-0038 · Advanced English Vocabulary (Chinese Edition) · Lesson No.038 · C1 → C2 · 简体中文辅助
An average is a summary, not a person. A mean is not always typical. A median is not automatically better. And one central value cannot tell you how the rest of the distribution is arranged.
average 是摘要，不是“一个典型的人”；mean 不一定典型；median 也不是永远更好；而任何一个中心值，都不能单独告诉你整个分布长什么样。
Advanced learners often use average as if it were one simple concept. In everyday English, average can mean an arithmetic mean, a usual level, or a standard considered typical. In statistics, mean, median and mode are different measures of centre. In evaluation, average can even become an adjective meaning ordinary or neither very good nor very bad. The same familiar word therefore carries mathematical, descriptive and evaluative jobs.
OpenStax describes mean, median and mode as widely used measures of centre and shows why the mean is sensitive to extreme values while the median is less affected. NIST similarly emphasises that mean, median and mode capture different aspects of “centerness,” particularly when a distribution is skewed. This is the conceptual foundation for the lesson: the correct central-value noun depends on what structure you want the summary to preserve.
A centre summarises a distribution; it does not replace the distribution.
中心值是在总结分布，不是在取代整个分布。
This lesson owns the Mandarin-supported C1–C2 lexical route for average, arithmetic mean, median, mode, central tendency, typical value, representative value, weighted mean, trimmed mean, benchmark and “above/below average” language. It coordinates with Lesson No.037 on denominator metrics and Lesson No.039 on distribution/spread, but does not replace either.
Part I — Build the centre map | 第一部分：建立“中心值地图”
1. Average is an umbrella word in ordinary English | average 是广义词
Cambridge records several uses of average: a statistical average, a level considered typical for a group or sector, and an adjective meaning ordinary or typical. Therefore “the average was 70” may refer to an arithmetic mean, while “the average customer” refers to an imagined typical member, and “an average performance” evaluates quality. The surrounding noun and register decide which sense is active.
2. Arithmetic mean is the formal sum-divided-by-count measure | arithmetic mean
The arithmetic mean is calculated by adding values and dividing by the number of values. In many everyday contexts, average means this mean. In technical writing, naming mean removes ambiguity because “average” can refer broadly to other central summaries.
3. Mean is mathematical but polysemous in English | mean has many senses
Mean can be a noun meaning arithmetic average, a verb meaning signify/intend, and an adjective meaning unkind or poor in some contexts. In data prose, “the mean score” is unambiguous; in ordinary speech, “mean” without a noun may trigger other senses. Domain context controls interpretation.
4. Median is the middle position after ordering | median
The median is the middle value when observations are ordered, or the midpoint determined by the two central values when the number of observations is even. Its conceptual job is positional: half the observations lie on one side and half on the other under the usual definition. Unlike the mean, it does not directly depend on how far extreme observations sit from the centre.
5. Mode is the most frequent value/category | mode
The mode is the value or category that occurs most frequently. It is especially useful for categorical data where calculating a mean may make no sense. A dataset can have one mode, several modes or no uniquely useful mode when all values occur with similar frequency.
6. Mean, median and mode answer different centre questions | three centres
The mean balances all numerical values; the median identifies the middle ordered position; the mode identifies the most common value/category. In a symmetrical unimodal distribution they may be close or identical. In skewed or multimodal data they can differ sharply. The difference is information, not inconvenience.
7. Central tendency is the statistical family name | central tendency
Measures of central tendency is a formal statistical phrase for measures describing a central or typical location of data. It commonly includes mean, median and mode. General readers may prefer measures of centre or simply named statistics.
8. Central value is a plain analytical phrase | central value
Central value can be used more loosely for a value near the middle of a distribution. It is less technical than a formally defined statistic. Do not use it where the exact calculation—mean or median—matters.
9. Typical value is a meaning claim, not a formula | typical
A typical value is intended to represent what is usual or characteristic. The mean may be typical in some distributions, the median in others, the mode in categorical data, and sometimes no single value is genuinely typical. “Typical” therefore requires a judgment about the distribution, not just a calculator.
10. Representative value asks whether a summary reflects the group | representative
A representative value should capture something important about the group without being badly distorted by unusual observations or hidden subgroups. This is a substantive judgment. A mathematically correct mean can be unrepresentative of most members when a distribution is highly skewed.
11. “Average person” is often an abstraction | average person
If the average household has 2.4 people, no household needs to contain 2.4 people. The arithmetic mean can describe a population without corresponding to any actual member. “Average” as a summary should not be imagined automatically as a real typical individual.
12. Mean is sensitive to extreme values | outliers and mean
For 1, 2, 3, 4, 5, the mean is 3. Replace 5 with 100 and the mean rises dramatically. The extreme value contributes its full magnitude to the sum. This sensitivity can be desirable when extremes genuinely matter, but it can make the mean a poor description of what most observations look like.
13. Median is much less affected by extreme magnitude | robust centre
In the same ordered dataset, making the largest value 100, 1,000 or 1,000,000 does not change the median if the ordering around the middle stays the same. This robustness is why median income, house price and waiting time are often informative in skewed settings. It does not mean median is always “better”; it means it answers a different centre question.
14. Mode ignores distance entirely | frequency centre
The mode cares about frequency, not numerical distance. In the values 1, 1, 1, 100, 200, the mode is 1 even though the mean is much higher. This can be useful for the most common category or size, but it may not summarise the numerical centre well.
15. Symmetry often brings mean and median together | symmetric distribution
In a roughly symmetric distribution, mean and median are often close. OpenStax and NIST use this relationship to explain distribution shape. When they diverge substantially, the difference can signal skew, extremes or complex structure that deserves inspection.
16. Right skew often pulls the mean upward | right skew
When a distribution has a long upper tail, very high values pull the mean toward that tail. Median is less moved by their magnitude. Therefore mean income can sit well above median income in a right-skewed population. The difference does not mean one statistic is wrong.
17. Left skew can pull the mean downward | left skew
A long lower tail can make the mean smaller than the median. The relationship among mean, median and mode is a clue to shape, but not an infallible rule for every dataset. Technical interpretation belongs to statistics; the language lesson is to avoid treating one centre as self-evidently representative.
18. Multimodal distributions can make one centre misleading | multiple peaks
If half a population clusters around 40 and another half around 90, the mean may be 65 even though few people are near 65. A single centre falls into the valley between two subgroups. In such cases, reporting subgroup centres or the distribution is more informative than presenting one “average person.”
19. Bimodal means two prominent modes | bimodal
Bimodal is technical distribution vocabulary for two modes/peaks. It belongs primarily to Lesson No.039’s distribution owner, but it matters here because a bimodal dataset demonstrates why the mean can be mathematically central yet not typical of either group.
20. Average can hide polarisation | same mean, different people
Four survey responses 3,3,3,3 and four responses 1,1,5,5 can have the same mean under numerical coding, yet the first group is uniformly neutral while the second is polarised. The average alone cannot reveal disagreement structure. Distribution matters.
21. Average can hide inequality | same mean, different spread
Income distributions 50,50,50,50 and 0,0,0,200 share the same mean. They describe radically different societies. A centre is one coordinate, not a complete description of allocation.
22. Average can hide subgroups | aggregation
An overall average may combine groups with different centres: age groups, ability groups, regions or products. “Average score = 70” may conceal one group averaging 90 and another 50. Before calling 70 typical, inspect whether the population is heterogeneous.
23. Weighted mean gives observations/groups unequal contribution | weighted mean
A weighted mean multiplies values by weights before combining them. Weights may reflect group size, importance, probability, sampling design or credit value. “Weighted average” therefore requires the weighting rule. A course grade with 60% exam and 40% coursework is not the simple mean of two marks unless weights happen to be equal.
24. Simple mean of subgroup means can be wrong | average of averages
If one class has 10 students averaging 90 and another has 100 students averaging 70, the combined mean is not 80. Group sizes must weight the subgroup means. This is the central-value counterpart to Lesson No.037’s denominator architecture.
25. Trimmed mean reduces influence of extremes | trimmed mean
A trimmed mean removes a specified fraction of the lowest and highest observations before calculating the mean. OpenStax notes this as one method for creating a centre less dominated by outliers. The exact trimming rule must be reported; “adjusted average” is too vague if method matters.
26. Winsorized mean is specialist vocabulary | winsorized mean
Some statistical procedures replace extreme values rather than remove them. Winsorized mean is specialist terminology and belongs in recognition vocabulary unless the learner works with robust statistics. Do not use the term as a fancy synonym for trimmed mean.
27. Geometric mean is not arithmetic mean | geometric mean
The geometric mean is a different mathematical mean used in contexts involving multiplicative change, ratios or growth factors. Technical applications require the formal definition. A report that says “average annual growth” may use arithmetic or geometric methods depending context; verify rather than assume.
28. Harmonic mean is another specialist centre | harmonic mean
The harmonic mean appears in rate and ratio contexts. General C1–C2 learners mainly need to recognise that mean is a family, not automatically arithmetic mean. When the type matters, the adjective is part of the metric.
29. Benchmark is a reference, not necessarily an average | benchmark
A benchmark may be an industry average, best-practice standard, target or reference case. “Above benchmark” does not necessarily mean “above average.” If the benchmark is a target of 90 while the group mean is 75, the two references differ.
30. Part I checkpoint: centre first, representativeness second | 第一部分检查点
Before calling a number “average” or “typical,” ask: Which measure of centre? How is it calculated? Is the distribution symmetric or skewed? Are there extreme values? Are there multiple clusters? Are groups weighted? Does the central number correspond to a plausible member? A centre can be correct mathematically and still poor as a story about the group.
Part II — When one centre tells the wrong story | 第二部分：一个中心值什么时候会讲错故事？
31. Same mean, different spread | 同均值，不同离散程度
Datasets 49,50,50,51 and 0,0,0,200 can have similar or identical means under suitable scaling while looking completely different. A mean says where the balance point lies; it does not reveal how tightly observations cluster around it. Lesson No.039 will own spread, but Lesson No.038 must teach why centre alone is insufficient.
32. Same mean, different shape | 同均值，不同形状
Two distributions can share the same mean while one is symmetric and another heavily skewed. Calling both “average 50” is numerically correct but descriptively incomplete. The central value does not encode tail length, clustering or asymmetry.
33. Same median, different extremes | 同中位数，不同极端值
1,2,3,4,5 and -100,2,3,4,1,000 can have the same median around 3 after ordering, yet radically different ranges and means. The median protects the middle position but intentionally ignores how far extremes are from the centre. Robustness is a feature, but also a limitation if extreme magnitudes matter.
34. Same mode, different distributions | 同众数，不同分布
Two datasets can both have mode 10 while one contains mostly values near 10 and another contains a broad range with only a small cluster at 10. Mode identifies the most frequent value, not how dominant or representative that value is.
35. Outlier is a data point, not an insult | outlier
An outlier is an observation markedly different from others under some analytical rule. It may be error, rare genuine case or important signal. Calling a value an outlier does not authorise deletion. For centre selection, the key question is whether extreme magnitude should influence the summary.
36. Mean can be appropriate precisely because extremes matter | extremes may matter
If a hospital must plan average cost, very expensive cases affect total resource needs and should influence the mean. Choosing the median simply because it is “robust” could understate budget burden. Measure choice should follow the decision question, not a universal rule that median is better for skewed data.
37. Median can be better for a typical transaction | median transaction
If most house prices cluster below a few luxury properties, the mean may be much higher than what a typical buyer sees. Median house price answers a positional question: half transactions are below and half above. For “what does the middle transaction look like?”, median is often intuitive.
38. Mean can answer budget questions; median can answer typical-person questions | question determines centre
A company may need mean expense per employee to calculate total spending, yet median expense to understand the typical employee’s claim. Both centres are valid because they support different decisions. Reporting both can be more informative than arguing which is “the real average.”
39. Mode is useful for categories | categorical data
If customers choose red, blue or green, mean colour is meaningless. Mode identifies the most common category. This is an important reminder that not every variable supports arithmetic operations. The type of data determines which centre words make sense.
40. Mode can be useful for sizes and choices | modal category
A shoe retailer may care about the most commonly sold size. Median shoe size is mathematically possible under coding but may be less operationally useful. “Most popular size” and “modal size” describe frequency leadership, not balance point.
41. There can be multiple modes | multimodal
If sizes 38 and 42 are equally most common, the data are bimodal in a simple sense. Saying “the mode is 40” because 40 lies between them would be wrong. Mode is about observed frequency, not midpoint.
42. No unique mode can exist | no clear mode
If all values occur once, no value is more frequent than another. Some conventions say there is no mode; others may describe all values as equally modal. For general communication, “there is no clear most common value” is often safer.
43. “Typical” can refer to frequency, centre or characteristic pattern | typical is polysemous
A typical customer can mean someone near the statistical centre, someone from the largest subgroup, or a representative profile constructed from common features. Because the word is flexible, define what makes the example typical when decisions depend on it.
44. “Representative” asks about the group, not the number alone | representative
A value is representative if it captures an important property of the population for the task. A mean income may be representative for total-income calculations but unrepresentative of the lived income of most residents. Representation is task-relative.
45. Typical case and representative sample are different | case vs sample
A typical case is one example judged characteristic. A representative sample is a set of observations that adequately reflects a target population under a sampling/design criterion. Do not use “representative” casually when discussing samples; it carries methodological weight.
46. Average can become an evaluative adjective | average performance
“An average student” can mean near the group’s usual level; “an average meal” can mean mediocre. This evaluative use can sound dismissive when describing people. In educational communication, prefer specific performance descriptors rather than labelling a learner “average.”
47. Above average and below average require a reference group | reference class
“Above average” is incomplete unless the reader knows average of what: class, age group, national cohort, industry, historical period. Changing the reference group can change the label. Benchmark and average are not universal standards.
48. Sector average is a benchmark-like comparison | sector average
Cambridge includes examples such as sector/industry averages. These are empirical group summaries used as reference points. A company above sector average may still miss its strategic target. Average describes peers; target describes desired performance.
49. Baseline average differs from current average | time reference
“Above average” across the last decade can differ from “above this year’s average.” The temporal reference belongs to meaning. Long-term historical averages can be poor expectations during structural change.
50. Moving average is not the same as overall mean | moving average
A moving average recalculates a mean over rolling windows to smooth time-series data. It is a derived trend tool, not one mean across the entire period. Lesson No.036 owns trend use; Lesson No.038 owns the central-value concept behind the term.
51. Running average is context-dependent | running average
A running average may mean cumulative average up to the current point or a moving average depending domain. If method matters, define the window/calculation. Casual “average so far” can be clearer than ambiguous technical shorthand.
52. Rolling average requires a window | rolling average
“Seven-day rolling average” tells readers each point summarises a seven-day window. Omitting the window can make trend interpretation difficult because different windows smooth differently.
53. Weighted average requires weights that sum meaningfully | weight architecture
Weights can be percentages summing to 100%, counts, probabilities or other coefficients. “Weighted average = 75” says little without knowing whether weights represent group sizes, assessment importance or sampling adjustment.
54. Course grade is often weighted mean | educational weighting
If coursework is 40% and exam is 60%, a 90 coursework mark and 60 exam mark produce 72 overall, not simple average 75. “Average mark” can be ambiguous when assessment components carry unequal weights. Use weighted overall mark if clarity matters.
55. Portfolio return is often weighted by allocation | finance weighting
A portfolio’s overall return depends on asset weights, not simple average of each asset’s return unless all allocations are equal. Technical finance has additional complexities; the lexical lesson is that “average return” needs a weighting method.
56. Population mean and sample mean have different roles | population vs sample
The mean calculated from a sample estimates or describes the sample, depending the task. Calling it “the population average” requires inference. Lesson No.033’s evidence boundary applies: sample statistic does not become population truth through paraphrase.
57. Expected value is not always a typical observed value | expected value
Lesson No.035 noted that expected value is technical probability vocabulary. An expected value can be a mathematical mean of possible outcomes and may not itself be likely or even possible as one outcome. For a fair die, expected value 3.5 is not a face you can roll. “Expected” does not mean “typical observed result” automatically.
58. Average rate needs denominator weighting | combining rates
If one branch has 10 cases at 90% success and another 1,000 cases at 50%, simple averaging their success rates gives the wrong combined rate. To find the aggregate success rate, combine successes and opportunities or weight by denominators. Lesson No.037 supplies the denominator logic.
59. Mean percentage can be meaningless across incompatible bases | average of percentages
Averaging “20% of revenue,” “30% of students” and “10% probability” produces a number with no coherent meaning because the quantities belong to different metric families. Same numerical format does not make values commensurable.
60. Mean of ratios may need weights or another summary | ratios
Average student–teacher ratio across schools can be calculated in several ways: simple mean of school ratios or overall students divided by overall teachers. These answer different questions because school size changes weighting. Name the method.
61. Average class size differs from student-weighted class size | weighting perspective
Average across classes gives each class equal weight; average class size experienced by a randomly chosen student effectively weights large classes more. A single phrase “average class size” may hide perspective. Quantitative language should match the unit of analysis.
62. Unit of analysis determines what is averaged | unit of analysis
Are we averaging students, classes, schools, days or regions? The mean can change because the observational unit changes. Before reporting “average”, identify what each observation represents.
63. Missing values can change the mean | missing data
If only students who submitted an assignment are included, the mean score describes submitters, not all enrolled students. Missingness changes the set being averaged. “Average among respondents/completers” may be necessary.
64. Zero and missing are different in averages | zero vs missing
Treating missing observations as zero can drastically lower a mean; excluding true zeros can raise it. Quantitative English should reflect the data rule: “non-response treated as zero,” “missing values excluded,” or another documented method.
65. Censoring and caps can distort averages | capped values
If a survey records income only as “$200,000 or more,” the exact mean may be hard to estimate because top values are censored. Median may be easier to report accurately. The methodological issue matters because the central value depends on what data are actually observed.
66. Top-coding limits extreme-value information | top-coded
Top-coded data replace values above a threshold with a common maximum category/value. This protects privacy or simplifies reporting but affects means. It is specialist recognition vocabulary, useful for reading official statistics.
67. Winsorisation and trimming are not ordinary cleaning | robust procedures
Removing or modifying extremes changes the centre by design. Reports should not say “the average after cleaning” when a formal trimmed or winsorized procedure was used and that method affects interpretation.
68. Arithmetic mean requires meaningful addition | data type
Numbers used as labels do not necessarily support a meaningful mean. Averaging postal codes or arbitrary category numbers is meaningless. Even ordinal rating scales require care because equal numerical gaps may not correspond to equal conceptual distances.
69. Likert-scale averages need interpretive caution | ordinal scales
Survey responses such as strongly disagree → strongly agree are often coded 1–5 and averaged in practice, but the interpretation depends on methodological assumptions. A distribution showing polarisation can be lost in the mean. Reporting category shares or median/mode may complement the average.
70. Part II checkpoint: “average” is a method choice | 第二部分检查点
Whenever “average” appears, ask: arithmetic mean, weighted mean, median, mode, moving average, expected value or informal typical standard? Then ask what observations enter, how they are weighted, whether extremes matter, and whether one centre actually represents the group. The word average is easy; the underlying decision is not.
Part III — Choose the centre for the reader job | 第三部分：根据读者任务选择中心值
71. Income: mean and median answer different questions | 收入
Mean income answers something like “total income divided across people.” Median income answers “what is the midpoint person/household in the ordered distribution?” In a right-skewed income distribution, mean may be much higher than median. Both are correct, but using mean to describe the “typical person” can create a misleading mental picture.
72. House prices: median often communicates the middle transaction | 房价
A small number of luxury properties can pull the mean upward. Median transaction price is often used to describe the middle sale. However, mean remains useful for total-market revenue calculations or average transaction value. “Median is better” is too broad; the decision objective decides.
73. Waiting time: median can protect the typical experience | waiting time
If most customers wait five minutes but a few wait two hours, mean waiting time can rise substantially. Median tells what the middle customer experienced. But operations managers may still need the mean or high percentiles to plan total resource use and tail problems.
74. Exam scores: mean may summarise class level, median checks skew | exam scores
Mean score is common in school reporting because every mark contributes. Median can reveal whether a few extremely high/low scores are moving the mean. Mode may identify a common score band but is often less informative with many possible marks. Reporting mean plus spread/subgroups is usually stronger than naming one “average student.”
75. Pass mark changes the decision question | centre vs threshold
Mean score 70 does not tell us how many students passed a 50-point threshold. A class with mean 70 can have very different pass rates depending on distribution. Lesson No.037 owns the pass-rate proportion; Lesson No.038 owns the central score. Do not use one as proxy for the other.
76. Classroom grouping can make the mean imaginary | heterogeneous groups
If one group scores around 40 and another around 90, overall mean near 65 may describe almost nobody. A tutor planning instruction should inspect subgroup centres rather than designing a lesson for the “average 65-point student” who may not exist.
77. Customer spending: mean matters for revenue, median for typical basket | business
Average order value often uses the mean because total revenue equals mean order value times number of orders under a simple calculation. Median order value may better describe the middle purchase when a few huge orders distort the mean. A product team and finance team may legitimately prefer different centres.
78. SaaS users: average usage can hide power users | user behaviour
A small group of power users may generate most events, making mean activity far above median activity. “Average user performs 100 actions” can be misleading if median user performs 12. Segmenting heavy, medium and light users can be more representative than one centre.
79. Server latency: median can hide tail pain | technical performance
Median latency may look excellent while a small share of users experience extreme delays. Mean may also hide tail structure. Technical teams therefore often report percentiles such as p95/p99. Lesson No.039 will own distribution/tails; the central lesson is that a centre is not a tail metric.
80. Reliability: average uptime can hide outage clustering | time aggregation
Two services can have the same average uptime but different outage patterns: one has many short interruptions; another one long outage. The mean rate cannot show temporal distribution. Centre summaries need complementary structure metrics when user experience differs.
81. Ratings: mean star rating can hide polarisation | review scores
A product with all 3-star reviews and a product with half 1-star, half 5-star reviews can both average 3. The first produces consensus; the second produces polarisation. “Average rating 3” alone hides the customer-experience distribution.
82. Mode can reveal the dominant rating category | modal rating
If 5 stars is the most common category but many low ratings pull mean down, reporting both mean and mode may explain why “most common rating” and “average rating” differ. Use mode when the most frequent response matters operationally.
83. Survey scales: median/mode can preserve ordinal meaning | ordinal response
For ordered categories such as satisfaction levels, median and mode can be intuitive because they rely on order/frequency rather than assuming equal intervals. Some analyses still report means under accepted practices. The vocabulary should not pretend methodological controversy disappears; name the measure used.
84. Salary offers: market median and market average differ | compensation
Salary data often skew upward because a few senior roles earn far more. Median salary can represent the midpoint role; mean salary reflects total payroll burden. Recruiters, workers and finance teams may ask different questions from the same dataset.
85. Insurance claims: mean cost matters for expected total cost | claims
Claims are often right-skewed because most are modest but some are enormous. Median claim may describe typical claim, but mean cost is important for expected total payout across many claims. Extreme claims are not noise when they genuinely drive financial risk.
86. Travel time: median trip vs mean network burden | transport
Median commute can represent the middle traveller; mean commute times the number of travellers approximates total person-time under conditions. A planning question about typical experience differs from a question about aggregate time cost.
87. Energy consumption: mean can be driven by high-use households | energy
If a few large homes consume much more energy, mean household consumption exceeds median. Utility capacity planning may need mean/total; affordability or typical-household communication may use median or subgroup profiles.
88. Medical measurements: “normal average” is not a universal healthy value | domain caution
Clinical reference values are domain-specific and may depend on age, sex, measurement method and population. A population mean should not be interpreted as a personal medical target. This lesson teaches language architecture, not medical guidance.
89. Manufacturing: mean dimension can hide tolerance failures | quality control
A batch can have perfect mean diameter while half items are too large and half too small. Quality depends on spread and tolerance, not centre alone. Lesson No.034 owns tolerance/threshold; Lesson No.039 owns spread. The mean cannot certify conformity.
90. Finance: average return needs time-weighting method | returns
Arithmetic and geometric average returns answer different questions when compounding matters. Technical finance uses formal conventions; do not label one “average return” without method when the distinction affects interpretation.
91. Growth rates: arithmetic mean can misstate compounded experience | growth
A +50% year followed by -50% does not return to the original value; it leaves 75% of the starting amount. Simple arithmetic mean of growth rates is 0%, yet compound outcome is negative. Geometric methods are relevant in multiplicative processes.
92. Speed: average speed depends on time and distance weighting | rates
Average of 30 km/h and 60 km/h is not always 45 km/h for a journey; it depends on whether equal times or equal distances are spent at each speed. Harmonic mean may arise in equal-distance problems. The broader language lesson: average of rates depends on weighting.
93. Sports: batting/scoring averages encode domain rules | domain averages
Sports metrics called “average” may use specific denominators and formulas. Do not assume a familiar word means arithmetic mean of raw scores. Domain labels can package specialised calculations.
94. Weather: average temperature depends on averaging convention | weather averages
Daily mean temperature may be computed using a specified observation procedure. Climate normals average across defined multi-year periods. “Average weather” should not be treated as a day that literally occurs; it summarises a distribution over time.
95. Population age: mean age and median age differ | demography
Mean age weights extreme ages by distance; median age splits the population into younger and older halves. Demographic reporting commonly uses median age because it describes the midpoint population age, but mean may serve other analyses.
96. Typical student is often a dangerous fiction | education
Curriculum design for an “average learner” can fail when ability, language background and pace vary widely. Use central summaries for planning, but preserve variation and subgroups. A mean score should not become a psychological profile.
97. Representative example should not be cherry-picked | example selection
A representative example is intended to illustrate a broader pattern. Choosing an unusually dramatic case and calling it representative exaggerates the group. The adjective should be justified by distribution or selection logic, not rhetorical convenience.
98. Typical case can be central on some features and not others | multidimensional typicality
A customer may have median spend but unusual age; average session length but rare product choices. “Typical customer” compresses many dimensions into one label. Persona design should identify which dimensions the case represents.
99. Representative sample does not mean every person looks average | sample representation
A representative sample should reproduce relevant population structure, including variation. It may intentionally include extremes and minorities. “Representative” refers to the sample’s relationship to the population, not to every individual being near the mean.
100. Prototype is not average person | prototype
In cognitive/UX language, a prototype may be an abstract bundle of characteristic features rather than an actual arithmetic average. Do not use statistical vocabulary to describe conceptual prototypes unless calculation is genuinely involved.
101. Archetype is not representative statistic | archetype
An archetype is an idealised or characteristic model/type. It may intentionally exaggerate features. Calling an archetype “the average user” confuses narrative/persona construction with empirical centre.
102. Norm can mean expected standard, not mean | norm
Norm can mean usual behaviour, accepted standard or statistical reference. “Above the norm” may not mean above the arithmetic mean. Check the domain and reference procedure.
103. Normal can mean common, standard or statistical distribution | normal
Normal is highly polysemous: ordinary, expected, healthy, or relating to a normal statistical distribution. “Normal value” can be risky because it may imply desirable/healthy rather than merely common. Prefer specific reference ranges or population descriptions where possible.
104. Usual, common and typical overlap but differ | usual/common/typical
Usual often describes what normally happens for a person/system; common describes frequency across cases; typical describes characteristic pattern. None is a formal mean unless context explicitly links them to statistics.
105. “Average of X” and “average X” differ grammatically | grammar
“The average of the five scores was 70.” “The average score was 70.” Both are natural. “On average, students scored 70” packages the mean as an adverbial summary. Learn these sentence frames for clear reporting.
106. On average is not every time | on average
“Users spend 20 minutes on average” does not mean every user spends about 20 minutes. Some may spend one minute and others hours. On average summarises the group or repeated observations; it does not guarantee low variation.
107. Average per person can be redundant or useful | average per
“Average expenditure per person” is understandable but sometimes redundant if per-capita already implies averaging across population. In technical contexts, keep the conventional metric name rather than piling synonyms.
108. Mean value and average value often overlap | mean value
In mathematics/statistics, mean value can be precise. In plain English, average value is more accessible. Choose register, but do not switch if a field uses “mean” with a specific definition.
109. Median value should not be called average casually when distinction matters | naming
Some people use “average” loosely to include median. In a report comparing mean and median, this becomes confusing. Use the exact term when more than one centre appears.
110. “Typical average” is often redundant or vague | avoid noun piles
If you mean arithmetic mean, say mean/average. If you mean a value chosen to represent usual experience, say typical value and explain the criterion. Combining terms can hide rather than clarify.
111. Centre can shift because composition changes | compositional effects
Overall average can rise even if every subgroup’s average is unchanged, simply because more observations come from a high-valued subgroup. This is another reason Lesson No.037’s denominator/composition logic and Lesson No.038’s centre logic must work together.
112. Centre can stay unchanged while everyone changes | offsetting changes
If half the group rises ten points and half falls ten points, the mean may remain unchanged despite large individual movement. “Average remained stable” does not imply individuals stayed stable.
113. Mean improvement can hide widening gaps | equity
Average score can rise while low-performing students decline and high-performing students improve sharply. A centre can improve while inequality worsens. Use subgroup/distribution language before making broad claims about “students improving.”
114. Median improvement can hide tail deterioration | tails
Median can improve while the worst cases deteriorate if changes occur below the middle without moving the median position. No centre protects every decision. Tail-sensitive outcomes require distribution or percentile reporting.
115. Part III checkpoint: choose centre by decision, not habit | 第三部分检查点
Mean is strong when totals, balance or every magnitude matters. Median is strong for ordered midpoint and resistance to extremes. Mode is strong for most frequent category/value. Weighted/trimmed means solve different problems. “Typical” and “representative” are interpretive labels that require distribution knowledge. No central measure wins every task.
Part IV — Mandarin-to-English centre control | 第四部分：中文母语学习者的平均值与典型值转换
Chinese words such as 平均、均值、一般、典型、代表性 and 正常 overlap more freely than precise English statistical reporting allows. The repair is to ask what job the Chinese expression performs: arithmetic calculation, midpoint, most frequent value, usual level, reference standard, or characteristic example. Then choose the English term.
116. 平均 can be average, mean or on average | 平均不是一个固定词
“平均分为 70” → The average/mean score was 70. “平均每人 20 元” → 20 dollars per person on average. “平均来说” → on average. The English noun/adverb structure changes with the sentence. Do not force every 平均 into average as a noun.
117. 平均数 / 算术平均数 = mean / arithmetic mean | 统计学明确化
When the calculation is sum divided by count, arithmetic mean is exact. In many reports mean is sufficient. Use average for a general audience if no other kind of average is in play, but switch to mean when the distinction from median or mode matters.
118. 均值 usually maps to mean | 均值 = mean in technical prose
“样本均值” → sample mean. “总体均值” → population mean. “加权均值” → weighted mean. In technical writing, this family is more precise than repeatedly using average.
119. 中位数 = median | middle-position statistic
“收入中位数” → median income. “房价中位数” → median house price. Do not translate 中位数 as “middle average.” Median is a defined statistical noun and should remain intact.
120. 众数 = mode | most frequent value/category
“最常见的鞋码是众数” → The most common shoe size is the mode. In non-technical communication, most common value/category may be clearer than mode. Do not confuse 众数 with “majority,” which refers to a share exceeding half.
121. 中心趋势 = central tendency | family name
“集中趋势指标” → measures of central tendency. This is formal statistics vocabulary. For general readers, “measures of centre” or simply “mean, median and mode” can be more accessible.
122. 典型值 = typical value, but define what makes it typical | 典型是解释，不只是计算
“典型值” may refer to a median, mode, selected representative case or domain-specific reference. English typical value does not tell the reader how it was calculated. If calculation matters, name the statistic; if representativeness matters, explain the criterion.
123. 代表值 can be representative value / central estimate | 代表值不是固定 mean
“选一个代表值” → choose a representative value if the goal is to summarise. In modelling/forecasting, central estimate may be conventional. Do not translate it automatically as average unless an averaging method is actually used.
124. 代表性 = representativeness | sample vs value
“样本具有代表性” → the sample is representative. This is a methodological claim about relation to the population, not a statement that sample values are near the mean. “代表性强” can be awkward if translated literally; specify more representative of the target population.
125. 一般水平 can be typical/usual level, not mean | 一般水平
“行业一般水平” may be typical industry level, industry norm or industry average depending whether the source actually calculated a mean. Do not manufacture an arithmetic statistic from a qualitative phrase.
126. 平均水平 usually means average level, but ask which average | 平均水平
In public prose, “below the national average” is natural. In technical reports, specify below the national mean or below the national median if the reference statistic matters. “Average level” alone may hide which centre is used.
127. 正常值 is not automatically average value | 正常 ≠ 平均
“正常值” may mean a reference range, expected operating value or clinically defined interval. Translating it as average value can be wrong and potentially misleading. Use normal/reference range only when the domain defines it, and avoid treating normal as synonymous with healthy or desirable.
128. 常态 can be norm / usual state / normal condition | 常态
“恢复常态” → return to normal/usual conditions. “社会常态” may be the norm. This is not a central-value statistic. The word family around “normal” often describes convention or usual state, not arithmetic centre.
129. 平均每人 = per person on average / average per person | per-person centre
“平均每人使用 12 GB” → Users consumed 12 GB per person on average. If the metric is conventionally per capita, use that. Remember that the per-person average may not describe any individual user when usage is skewed.
130. 人均 = per capita, not necessarily “typical person” | 人均
Per-capita income is mathematically total income divided by population. It is a mean-like normalised quantity but may be far above median income in an unequal distribution. “人均收入” should not be paraphrased “what the typical person earns.”
131. 户均 = average/per household | 户均
“户均收入” may be average household income. If the source specifically uses median household income, do not replace it with average. Household size/composition can also affect interpretation; unit of analysis is the household, not person.
132. 加权平均 = weighted average / weighted mean | 加权
Use weighted mean in technical statistics, weighted average in general/business contexts. The weights must be named when they affect interpretation: weighted by enrolment, sampling weight, portfolio allocation or assessment percentage.
133. 简单平均 = simple average / unweighted mean | 简单平均
“简单平均各地区比例” may be take the unweighted mean of regional percentages. This gives each region equal influence regardless of population. If population-weighted national percentage is intended, the calculation is different.
134. 截尾平均 = trimmed mean | robust centre
“去掉最高最低各 10% 后求平均” → 10% trimmed mean under an appropriate convention. This is not the same as median or winsorized mean. The trimming fraction is part of the method.
135. 滚动平均 = moving/rolling average | time-window centre
“七日滚动平均” → seven-day moving/rolling average. Name the window. This is a time-series smoothing measure, not the same as average over the full dataset.
136. 移动平均 = moving average | moving average
Moving average is the standard term in many technical/business contexts. A “moving mean” is possible in specialised usage but less common generally. Follow domain collocation.
137. 平均增长率 may require arithmetic or geometric mean | 平均增长率
“平均年增长率” is ambiguous unless the method is specified. CAGR/compound annual growth rate uses compounding logic; arithmetic average of annual rates is different. Translate the method, not only the phrase.
138. 平均速度 can hide harmonic-mean cases | 平均速度
“平均速度” → average speed, but calculation depends on total distance divided by total time, not simple mean of speed values unless time intervals align appropriately. Language learners should not assume “average of the numbers shown” equals the physical average.
139. 平均价格 can be mean transaction price or index-like measure | 平均价格
“平均房价” may be mean sale price; “房价中位数” is median price; a housing price index is another derived measure entirely. Do not translate all three as “average house price.”
140. 平均工资 and 工资中位数 must stay distinct | 工资
Mean wage/salary divides total wages by workers; median wage/salary locates the midpoint worker. In skewed earnings distributions, these can differ substantially. News translation should preserve which statistic the source gives.
141. 高于平均 = above average, but name reference if needed | 高于平均
“高于全国平均水平” → above the national average. If the national reference is specifically a median, say above the national median. “Average” should not erase a formally named statistic.
142. 低于平均 = below average | 低于平均
The phrase can be purely descriptive or evaluative. “Below-average cost” may be desirable; “below-average accuracy” may be undesirable. Do not let direction automatically become value judgment.
143. 平均偏高 / 偏低 require reference wording | slightly above/below average
“平均偏高” may be slightly above the reference average or “the mean was somewhat higher.” Avoid vague average is high when the comparison group is unclear.
144. 平均人 / 普通人 are not the same | average person vs ordinary person
The average person can refer to statistical/typical traits. An ordinary person means non-special/expert in many contexts. Chinese 普通人 should not automatically become average person, which can imply a quantitative comparison.
145. 普通 = ordinary/common, not average automatically | 普通
“普通消费者” → ordinary/general consumer may be better than “average consumer” unless a data-defined persona is intended. Average can sound evaluative or statistical.
146. 常见 = common, not mean | 常见
“最常见值” → most common value, which may correspond to the mode. “常见做法” → common practice. Frequency language should not be rewritten as average without a calculation.
147. 典型 = typical / characteristic, not arithmetic average | 典型
“典型案例” → typical/representative case depending selection logic. It is usually not “average case” unless the case is deliberately near a numerical centre.
148. 有代表性 = representative of, not representative for every purpose | 有代表性
Say what it represents: representative of the target population, representative of typical classroom conditions. A sample can be representative on age but not income; an example can be representative of process but not outcome distribution.
149. 最常见 = most common / modal | 最常见
“最常见答案” → most common response. In technical prose, modal response is possible. Do not translate it as “average response,” which may refer to a numerical mean.
150. 中间水平 = middle / median-like, but not automatically median | 中间水平
Chinese informal “处于中间水平” may mean neither high nor low, not a mathematically calculated median. English mid-range / around the middle may be better unless the actual median is intended.
151. 中等 = moderate / medium, not median | 中等 ≠ median
Median is a statistical position. Medium/moderate are scale categories. “中等难度” → moderate difficulty, never “median difficulty” unless a formal median is calculated.
152. 平均分布 is dangerous phrasing | 平均分布
If Chinese means “evenly distributed,” use evenly distributed, not “averagely distributed.” If it means distribution around an average, say that. Average distribution is usually unclear without a technical context.
153. 均匀分布 = uniform distribution / evenly distributed | 均匀
Uniform distribution can be a formal probability/statistical term. Evenly distributed is general descriptive language. Do not use the technical noun when you simply mean “spread fairly evenly.”
154. 平均值受极端值影响 = mean is sensitive to extremes | 极端值
A natural translation is The mean is sensitive to extreme values/outliers. Avoid saying “the average is inaccurate because of outliers.” The mean may be exactly correct mathematically; the issue is representativeness for the reader job.
155. 中位数较稳健 = median is more robust to extremes | 稳健
In statistical language, robust can describe resistance to extreme values or assumption violations. Do not translate 稳健 simply as “stable.” “Median is robust to outliers” is different from “median is stable over time.”
156. 平均值并不代表大多数人 | no “average person” assumption
A strong English sentence: The mean does not necessarily represent what most individuals experience. This is often more precise than “the average is misleading,” because it explains what type of representation is missing.
157. 分组平均 = subgroup mean/average | 分组
“分组平均值” → subgroup means. If group sizes differ, overall mean needs weighting by observations. “Average of group averages” should not be assumed to equal the overall average.
158. 总体平均 = overall mean / population mean, depending scope | 总体
“总体” in Chinese can mean overall/aggregate or statistical population. Translate carefully. Overall mean may combine subgroups in the observed dataset; population mean is a formal parameter of a target population.
159. 样本平均 = sample mean | sample mean
Use sample mean when the values come from sampled observations. Do not drop “sample” when readers may otherwise think the number describes the whole population exactly.
160. Part IV checkpoint: translate the centre job, not the Chinese surface word | 第四部分检查点
When Mandarin uses 平均、一般、典型、代表性、常见、正常 or 中间, ask: Is this a calculated mean? a midpoint? the most common value? a usual standard? a representative example? a benchmark? or simply ordinary? English separates these jobs more sharply. Choose the centre word only after the job is clear.
Part V — Central-value failure laboratory | 第五部分：中心值失误实验室
The examples below deliberately hold one summary statistic constant while changing everything around it. The goal is to train a reflex: whenever a centre appears, ask what the rest of the distribution could look like. A centre is useful because it compresses; it becomes dangerous when compression is mistaken for complete description.
161. Case 1 — same mean, consensus vs polarisation | 同均值：一致 vs 两极化
Group A ratings: 3,3,3,3. Group B: 1,1,5,5. Both have mean 3, yet the first group is unanimous and the second is split. “Both groups had the same average opinion” is numerically true but socially misleading. Better: “Both groups had the same mean rating, but Group B was strongly polarised.”
162. Case 2 — same mean, equal vs unequal income | 同均值：平等 vs 不平等
Households 50,50,50,50 and 0,0,0,200 both have mean 50. “Average income is identical” is correct; “income conditions are identical” is not. Distribution and inequality differ dramatically. A centre cannot substitute for spread/allocation.
163. Case 3 — CEO salary changes the mean | 极端高值拉高均值
Nine employees earn 50 and one earns 550. Mean pay is 100; median pay is 50. Saying “the average employee earns 100” can create a false mental picture because nine of ten earn half that. “Mean pay is 100, while median pay is 50” reveals the skew more honestly.
164. Case 4 — median hides extreme top burden | 中位数隐藏总成本
Nine medical claims cost 100 and one costs 100,000. Median claim describes a typical claim well; it is useless for estimating total payout by itself. Mean cost is heavily influenced by the expensive case precisely because total financial burden is. The right centre depends on whether the reader asks “typical claim?” or “expected total cost?”
165. Case 5 — “average household = 2.4 people” | 不存在的平均家庭
No household contains 2.4 people. The mean is still perfectly valid as a population summary. This case teaches that a central statistic need not correspond to an actual observation. “Average” should not automatically be personified.
166. Case 6 — bimodal class makes mean learner fictional | 双峰班级
Half the students score around 40 and half around 90. Mean near 65 describes almost nobody. A tutor teaching “the average student” at 65-level difficulty risks underserving both groups. Subgroup centres and the distribution are pedagogically more useful.
167. Case 7 — mean improves while struggling subgroup worsens | 平均改善掩盖弱势组下降
Top half rises from 80 to 100; bottom half falls from 60 to 55. Overall mean rises, yet struggling students deteriorate. “Students improved” overgeneralises an aggregate centre. Better: “Overall mean increased, driven by gains among higher-performing students, while lower-performing students declined.”
168. Case 8 — unchanged mean while everyone moves | 均值不变但人人变化
Half the students gain ten points; half lose ten points. Mean is unchanged. “Performance was stable” is false at individual level. Better: “The class mean was unchanged, although individual scores shifted substantially in opposite directions.”
169. Case 9 — average of averages without weights | 平均数的平均数陷阱
Class A has 10 students with mean 90; Class B has 100 students with mean 70. Simple mean of class means is 80, but student-level combined mean is about 71.8. “Average of the two class averages” answers an equal-class question; “overall student mean” answers an equal-student question. Weighting changes the unit of analysis.
170. Case 10 — weighted grade | 加权成绩
Coursework 90, exam 60. Simple mean = 75. If coursework weight = 40% and exam = 60%, final = 72. Calling 75 “the average grade” ignores assessment design. Use simple mean versus weighted overall mark.
171. Case 11 — mode for categorical choice | 类别数据不能乱求均值
Favourite colours: red, red, blue, green. Mode is red. Coding red=1, blue=2, green=3 and taking a mean would create a numeric result with no meaningful colour interpretation. Arithmetic centre requires meaningful numeric structure.
172. Case 12 — ordinal scale and fake precision | 顺序量表的虚假精度
Satisfaction levels coded 1–5 produce a mean 3.67. The decimal may look precise, but the scale’s conceptual intervals may not be equally spaced. Depending on the method, reporting median category and response distribution can complement the mean. Precision in digits does not guarantee precision in meaning.
173. Case 13 — average user vs power users | 平均用户被重度用户拉高
Ninety users perform 5 actions; ten power users perform 500. Mean usage is far above what most users do. “The average user performs 54.5 actions” is mathematically valid but behaviourally odd. Median, segments and percentile distribution tell a more useful product story.
174. Case 14 — mean latency vs tail latency | 平均延迟隐藏尾部痛苦
Most requests complete in 100 ms; a small fraction take 10 seconds. Mean may rise, median may remain excellent, yet tail users suffer badly. System teams often report p95/p99 for this reason. Centre does not describe tails.
175. Case 15 — median can hide bottom deterioration | 中位数也会隐藏尾部
If only the lowest 20% deteriorate, median may not move at all. “Median stable” is not “everyone stable.” A robust statistic can deliberately ignore changes in tails. That is useful for one question and dangerous for another.
176. Case 16 — mode can be rare in a broad distribution | 众数不一定“代表大多数”
If value 10 appears 5 times and every other value appears 4 times, 10 is the mode but represents only a small fraction of observations. “Most common” does not mean “most observations have this value.” Mode gives rank by frequency, not majority status.
177. Case 17 — no actual observation near the mean | 均值落在空白区
Values cluster around 10 and 90; mean = 50. Few or none near 50. “Typical value = 50” is unjustified. A central balance point can sit in a low-density region. Lesson No.039 will formalise clustering and multimodality.
178. Case 18 — median income vs per-capita income | 中位收入 vs 人均收入
Per-capita income is a mean-like total/population measure. Median income identifies midpoint. In unequal populations they diverge. Translating both Chinese phrases as “average income” destroys an important economic distinction.
179. Case 19 — mean household size vs individual experience | unit-of-analysis trap
Average household size across households gives each household equal weight. Average household size experienced by a randomly chosen person gives large households more weight. “Average household size” can mean different perspectives unless the observation unit is defined.
180. Case 20 — expected value 3.5 on a die | 期望值不是会出现的值
A fair six-sided die has expected value 3.5, but no roll can equal 3.5. Expected value summarises repeated probabilistic outcomes mathematically. It is not a “most likely roll” or typical single outcome.
181. Case 21 — arithmetic average growth says 0%, capital loses 25% | 复合增长陷阱
Value rises 50% then falls 50%: 100 → 150 → 75. Arithmetic mean of rates is 0%, but total return is -25%. Multiplicative processes need compounding-aware summaries. “Average growth rate = 0” can be mathematically one type of average yet economically misleading.
182. Case 22 — average speed 30 and 60 | weighting time vs distance
A journey at 30 km/h outbound and 60 km/h return over equal distances does not have simple average speed 45 km/h because more time is spent at the slower speed. This case teaches that rate averages require correct weighting, not automatic arithmetic mean.
183. Case 23 — moving average hides a shock | smoothing
A seven-day moving average smooths a one-day spike. This may reveal underlying trend but hide the acute event. “Average remained moderate” can coexist with a dangerous one-day peak. Smoothing is a lens, not the raw history.
184. Case 24 — average wait across sites without volume weights | branch weighting
Site A serves 10 people with mean wait 60 minutes; Site B serves 1,000 people with mean wait 5 minutes. Simple average of site means is 32.5 minutes; customer-weighted overall mean is close to 5.5. The correct average depends on whether the unit is site or customer.
185. Case 25 — industry average vs target | benchmark confusion
Industry mean defect rate = 4%; company target = 1%; actual = 3%. “Better than average” is true, but “meets target” is false. A benchmark and target serve different reference jobs.
186. Case 26 — above average depends on group | reference-class trap
A student score of 80 may be above class mean 70 but below national mean 85. “Above average” without reference group is incomplete. The adjective needs an explicit or well-established comparison population.
187. Case 27 — representative example chosen for drama | cherry-picking
A case study selects the highest-performing learner and calls them “representative.” This is rhetorically useful but methodologically misleading unless the selection reflects typical distribution. Representative is a claim about relation to the population, not storytelling power.
188. Case 28 — representative sample contains unusual people | variation belongs
A truly representative sample of a diverse population may include rare extremes. Removing everyone unusual to create an “average-looking” sample can make it less representative. Representation includes the population’s real heterogeneity.
189. Case 29 — median house price rises because low-end sales disappear | composition effect
If fewer inexpensive homes sell while prices of individual homes stay similar, observed median sale price can rise due to transaction mix. “House prices increased” may overstate underlying price change. Centre can move because the composition of observations changes.
190. Case 30 — mean grade rises because weaker students drop out | survivor composition
Average grade among remaining students rises after several low-performing students leave. “Students improved” is not established. The population changed. Report “mean among remaining students rose” and investigate composition.
191. Case 31 — zero treated as missing | coding choice
If zero sales is a valid value but the system treats zero as missing, average sales among recorded values will be too high. Conversely, coding non-response as zero can pull the mean down. The centre depends on inclusion rules.
192. Case 32 — top-coded income lowers mean | censored extremes
If all incomes above 200,000 are recorded as 200,000, mean calculated directly from top-coded values underestimates actual mean. Median may be unaffected if the median lies well below the cap. Data collection rules interact differently with different centres.
193. Case 33 — trimming removes real business risk | robust summary trade-off
A 10% trimmed mean of claims may describe typical claim costs, but if catastrophic claims drive solvency risk, trimming them out is inappropriate for total-risk planning. Robustness is not automatically truth; it is resistance to extremes for a particular analytical purpose.
194. Case 34 — mode changes with binning | grouped data
When continuous values are grouped into ranges, the modal class can depend on bin boundaries. “Most common range” is a property of the grouping scheme as well as raw data. Do not confuse modal class with exact modal value.
195. Case 35 — median of categories requires order | ordinal categories
Satisfaction categories have natural order; favourite colours do not. A median category can make sense for ordered responses but not unordered nominal categories. Data type precedes centre selection.
196. Case 36 — average of rankings | ranking caution
Mean rank may summarise positions across judges, but rank differences are ordinal and interpretation can be subtle. “Average rank 2.4” is a summary of ordered positions, not a score with equal measurement intervals. Use the conventional method but avoid over-reading decimals.
197. Case 37 — median age rises without anyone ageing unusually | population composition
Median age can rise because birth rates fall, migration changes or older cohorts become larger. “People are ageing faster” is not implied by a rising median age. The centre moves as population composition changes.
198. Case 38 — average temperature vs daily experience | time aggregation
A monthly mean of 25°C can combine cool nights and hot afternoons. “Temperature was around 25 every day” does not follow. Temporal averaging can create a central value rarely observed at any specific time.
199. Case 39 — average order value rises because one whale customer | business skew
One huge enterprise purchase raises mean order value while median and most customer baskets remain unchanged. “Customers are spending more” overgeneralises. Better: “Mean order value rose, driven by a small number of large orders.”
200. Case 40 — mean survey score hides response shift | distribution shift
Mean satisfaction remains 4.0, but responses move from mostly 4s to half 3s and half 5s. Average is unchanged while polarisation increases. A centre cannot reveal distributional rearrangement.
201. Case 41 — mean and median both move, but for different reasons | decomposition
A few extreme high values rise sharply, pulling mean upward; simultaneously the middle group improves slightly, nudging median upward. Saying “average rose” hides two mechanisms. Reporting both centres can reveal broad improvement plus top-tail acceleration.
202. Case 42 — mode changes while mean barely moves | categorical shift
Ratings move from many 3s to many 4s while a few low scores offset the mean. Mode shifts from 3 to 4, mean barely changes. The most common response changed even though arithmetic centre did not. Choose the centre matching the question.
203. Case 43 — no single centre is representative | multimodal population
One cluster around 20, another around 80, no observations around 50. Mean 50, median maybe 50 depending counts, modes near 20 and 80. Any single central value hides the two-group structure. The right report describes clusters/subgroups.
204. Case 44 — typical profile combines incompatible averages | Frankenstein average
A persona takes average age, average income, modal device and median session time. The resulting “typical user” may not correspond to any real segment because each feature’s centre comes from different people. Multivariate typicality requires more than combining marginal averages.
205. Part V checkpoint: central-value failure is usually a missing distribution question | 第五部分检查点
When an average sounds surprising, ask what it hides: extremes, skew, subgroup composition, polarisation, weighting, missing values, tail behaviour or incompatible units. The centre may be calculated correctly. The failure is often the claim made from it.
Part VI — FENCE, drills and mastery | 第六部分：FENCE、练习与真正掌握
206. Build the centre FENCE | 中心值 FENCE
| Fence | Question | Language job |
|---|---|---|
| F0 Variable | What is measured? | score, income, category, wait time |
| F1 Data type | Numeric, ordinal or categorical? | decide which centres make sense |
| F2 Centre | Mean, median, mode or another? | name exact statistic |
| F3 Weight | Equal or weighted? | state weighting rule |
| F4 Shape | Symmetric, skewed, multimodal? | judge representativeness |
| F5 Extremes | Do tails/outliers matter? | mean vs robust centre |
| F6 Reader job | Total burden, midpoint, most common? | choose centre by purpose |
| F7 Companion | What distribution info is needed? | spread, subgroup, percentile |
| F8 Transfer | Can I paraphrase without changing centre? | meaning-preserving output |
207. Build a centre-control card | 中心值控制卡
| Field | Example |
|---|---|
| Variable | household income |
| Unit | dollars/year |
| Population | households in region |
| Centre | median |
| Reason | skewed upper tail |
| Companion | mean + quartiles |
| Reference | 2026 households |
| Interpretation | midpoint household, not “typical every household” |
208. Practice A — choose mean, median or mode | 练习 A
- Most common shoe size.
- Middle house transaction in a highly skewed market.
- Total expected claims cost divided by number of claims.
- Most frequent survey category.
- Middle waiting time.
- Overall payroll divided by employees.
Choose a centre and explain the reader job. More than one statistic can be useful, but identify which question each answers.
209. Practice B — same mean, different story | 练习 B
- 50,50,50,50
- 0,0,0,200
- 25,25,75,75
- 10,40,60,90
All can be constructed around similar centres. Describe what the mean fails to show in each: equality, polarisation, spread or clustering.
210. Practice C — weighted average | 练习 C
- Coursework 80 at 30%, exam 70 at 70%.
- Branch A mean 90 for 10 users; Branch B mean 60 for 100 users.
- Portfolio A return 10% at 20% weight; B return 4% at 80% weight.
Calculate conceptually or numerically, then write a sentence that names the weights. Do not say “simple average” unless weights are equal.
211. Practice D — typical or merely central? | 练习 D
- Values cluster at 20 and 80; mean 50.
- Income mean 100, median 50.
- Ratings all equal 3.
- Most values 10, one value 1,000.
Decide whether the mean, median or mode can reasonably be called typical, and explain why “typical” is an interpretive claim rather than a formula.
212. Practice E — reference group | 练习 E
A score of 80 is above class mean 70, below national mean 85 and equal to school median 80. Write three accurate “above/below/at average” sentences without losing the reference group or statistic.
213. Practice F — Mandarin repair | 练习 F
- 这个班的平均分是 72。
- 工资中位数明显低于平均工资。
- 最常见的购买金额是 20 元。
- 平均用户每天使用产品 12 分钟，但中位用户只有 3 分钟。
- 这个“平均学生”其实并不存在。
- 七日滚动平均值下降，但单日峰值上升。
- 加权平均成绩为 76 分。
214. Practice G — paraphrase without changing centre | 练习 G
Original: “Median household income was $60,000.” Unsafe: “The average household earned $60,000.” Safe alternatives must preserve median: “The midpoint household income was $60,000” for a general reader, or retain the technical term directly.
215. Practice H — centre + companion | 练习 H
- Mean latency = 200 ms, p99 = 5 s.
- Median income = 50, mean = 100.
- Mean rating = 3, distribution = half 1 and half 5.
- Mean score = 70, subgroup means 50 and 90.
Write one two-sentence summary for each that uses a centre plus the minimum companion information needed to prevent a misleading story.
216. Practice I — average-of-averages trap | 练习 I
Three schools have means 90, 70 and 60 with 10, 100 and 1,000 students respectively. Explain why the simple average of school means answers a different question from the student-weighted overall mean. Then write a sentence naming the unit of analysis explicitly.
217. Practice J — data type check | 练习 J
- Favourite colour.
- Satisfaction category: low/medium/high.
- Income in dollars.
- Rank in race.
- Temperature.
For each, state which centres are meaningful and which would be questionable. Do not calculate merely because a value can be encoded numerically.
218. A 15-minute centre session | 15 分钟训练
| Time | Task |
|---|---|
| 0–3 min | Choose one real dataset/report with an average. |
| 3–6 min | Identify exact centre and unit of analysis. |
| 6–9 min | Ask what extremes/skew/subgroups might hide. |
| 9–12 min | Find one companion statistic/distribution fact. |
| 12–15 min | Explain why this centre fits the reader job. |
219. A 30-minute centre laboratory | 30 分钟中心值实验室
| Time | Task |
|---|---|
| 0–5 min | Create two datasets with same mean/different spread. |
| 5–10 min | Add an outlier and watch mean/median change. |
| 10–15 min | Create a bimodal dataset where mean is untypical. |
| 15–20 min | Combine unequal groups and test weighting. |
| 20–25 min | Translate Mandarin centre terms into precise English. |
| 25–30 min | Give a spoken recommendation: which centre and why? |
220. Seven-day centre challenge | 七天训练
| Day | Focus |
|---|---|
| 1 | mean / average |
| 2 | median / midpoint |
| 3 | mode / most common |
| 4 | weighted / trimmed / moving averages |
| 5 | typical / representative / norm |
| 6 | Mandarin false-equivalence repair |
| 7 | full FENCE audit |
221. Thirty-day centre-control challenge | 30 天系统升级
- Collect 20 reports using “average” and identify the exact statistic.
- Find 10 examples where mean and median tell different stories.
- Create 10 same-mean/different-distribution examples.
- Repair 15 “average person” overgeneralisations.
- Calculate/interpret 10 weighted averages with explicit weights.
- Identify 10 cases where mode is useful for categorical data.
- Translate 20 Mandarin centre expressions precisely.
- Pair 20 centres with a companion spread/subgroup statistic.
- Explain 10 centre choices orally under one minute each.
222. Twelve-week progression | 12 周路线
| Weeks | Focus | Evidence of mastery |
|---|---|---|
| 1–2 | mean / median / mode definitions | exact centre naming |
| 3–4 | skew / extremes / robustness | centre choice justified |
| 5–6 | weighted / moving / trimmed means | method labels preserved |
| 7–8 | typical / representative / norm | no “average person” shortcut |
| 9–10 | cross-domain centres | reader-job matching |
| 11–12 | centre + distribution synthesis | independent reporting |
223. First weak-link diagnostic | 第一个弱点诊断
| Symptom | Likely weakness | Repair |
|---|---|---|
| I call median “average”. | centre naming | mean/median contrast |
| I say median is always better. | reader-job control | budget vs typical case examples |
| I call mean “typical” automatically. | distribution blindness | skew/bimodal drills |
| I average group averages equally. | weighting | unit-of-analysis practice |
| I average category codes. | data-type control | nominal/ordinal/numeric audit |
| I use normal and average interchangeably. | polysemy | norm/reference/mean distinctions |
| I describe every representative case as average. | typicality semantics | representative vs central examples |
| I report centre without spread/subgroups. | compression overreach | centre + companion habit |
224. How a strong tutor teaches centre vocabulary | 高水平导师怎样教？
A weak lesson teaches three formulas. A strong lesson keeps the mean fixed while changing the distribution, then keeps the median fixed while changing the tails, then changes the question: “What does a typical person experience?” versus “What total resource burden should we budget for?” The statistic becomes a tool chosen for a job rather than a ritual calculation.
225. The learner should become their own centre editor | 自我编辑
Before writing “average,” ask: Which centre? Which observations? Which weights? Which reference group? Is the distribution skewed or multimodal? Do extreme values matter for the decision? Would median or mode answer a different question? What companion information is needed so readers do not imagine an “average person” who does not exist?
226. Reading harvest: circle every centre word | 阅读提取
Mark average, mean, median, typical, representative, benchmark, on average. Ask whether the source defines the statistic and whether another centre/spread measure appears nearby. This builds sensitivity to how expert writers frame central values.
227. Listening harvest: listen for the qualifier after average | 听力提取
Speakers often say “on average, across all branches,” “median income,” “weighted average,” “industry average.” The small qualifier determines scope/method. Missing it can turn a precise statistic into a vague one.
228. Writing activation: centre + reason + companion | 写作激活
Use a three-part frame: centre + why it is appropriate + companion distribution fact. Example: “Median waiting time was 6 minutes, a useful midpoint because waits were strongly right-skewed; the 95th percentile was 40 minutes.” This prevents centre-only reporting.
229. Speaking activation: 45-second centre choice | 口语激活
Given a dataset scenario, explain: “I would use the median because…; I would also report the mean/percentile because…”. The exercise trains real-time selection rather than formula recall.
230. Paraphrase test: preserve exact centre | 改述测试
Original: “The median wait was nine minutes.” Unsafe: “Customers waited nine minutes on average.” Safe: “Half of observed waits were nine minutes or less and half nine or more, with a median of nine minutes” if that explanation fits the dataset convention. Paraphrase should not convert positional centre into arithmetic mean.
231. Summary test: preserve centre plus distortion warning | 摘要测试
If a report says mean 100, median 50 because of a long upper tail, a summary “average = 100” hides the key distribution fact. A safe compression: “Mean income was 100, twice the median, reflecting a strongly skewed upper tail.”
232. Translation test: 平均 does not erase median | 翻译测试
If the Chinese source explicitly says 中位数, English must say median, even if ordinary readers loosely call it an average. Translation should preserve the statistic before simplifying the explanation.
233. Centre-selection checklist | 中心值选择清单
- Do I need total/balance information? → mean often matters.
- Do I need midpoint typicality in skewed data? → median may matter.
- Do I need the most common category/value? → mode.
- Are observations weighted unequally? → weighted centre.
- Are extremes errors/noise or real consequential cases?
- Is the distribution multimodal so no single centre is representative?
- Will readers mistake a central value for an actual typical person?
234. Canonical ownership boundary | 本课所有权边界
Lesson 038 owns central values and typicality: average, mean, median, mode, weighted/trimmed/moving means, typical and representative values, benchmarks and above/below-average language. Lesson 037 owns denominators/rates/shares; Lesson 039 owns distribution/spread/tails; Lesson 040 owns frequency/prevalence/incidence.
235. Recommended reference floor | 推荐参考资源
- OpenStax — Measures of Center — mean, median, mode and resistance to outliers.
- OpenStax — Mean, Median, and Mode — core definitions and trimmed mean.
- OpenStax — Skewness and Mean/Median/Mode — centre relationships under skew.
- NIST/SEMATECH — Skewed Right Distribution — centre measures under skew.
- Cambridge Dictionary — average — statistical, typical and evaluative senses.
- Cambridge Dictionary — mean — statistical and general senses.
- Cambridge Dictionary — median — median definitions and usage.
236. SEO language map | 本课关键词范围
This lesson naturally serves mean vs median, average vs median, mean median mode English, typical value vs average, representative value, weighted average English, trimmed mean, moving average, average of averages, skewed distribution mean median, average person myth, median income vs average income, C1 C2 quantitative English, 英语平均值, 均值中位数众数区别, 平均数英语, 加权平均英语, 典型值英语, 代表性英语 and 中文母语高级英语.
237. Your assignment | 本课作业
Choose one real dataset/report with at least two possible measures of centre. Build the FENCE and write a 300-word explanation for a general reader.
- Name the variable and unit.
- Identify data type.
- Calculate/record mean, median and mode where meaningful.
- Inspect skew, clusters and extremes.
- State which centre best answers your main reader question.
- Explain why another centre could answer a different question.
- Add at least one companion distribution/spread statistic.
- Translate into Mandarin.
- Translate back and verify centre terms survived.
- Two days later, reconstruct the FENCE from memory.
238. Why this owner earns longform depth | 为什么这一课需要长文深度？
The learning job is not teaching three formulas. It is teaching when a mathematically correct centre becomes a misleading description: skewed income, multimodal classes, weighted grades, power-user products, tail latency, survivor composition, ordinal surveys, moving averages and bilingual confusion between 平均、一般、典型 and 正常. Each failure needs a different repair, which is why a durable owner requires depth.
239. Final rules | 最后的规则
A centre summarises a distribution; it does not replace the distribution.
中心值是在总结分布，不是在取代整个分布。
Choose the centre that answers the question, not the one you calculate by habit.
选择能回答问题的中心值，而不是习惯性计算的那个平均值。
Series: EDKS-ADV-VOC-ZH · Lesson 038.
