Common Misconceptions in Statistics and How to Correct Them | 统计常见误区与纠正方法

📚 Common Misconceptions in Statistics and How to Correct Them | 统计常见误区与纠正方法

Statistics is the science of making sense of data, yet even the most capable students fall into predictable traps. From muddling averages to misreading graphs, many errors stem from a handful of deeply rooted misunderstandings. The good news is that once you see where the pitfalls lie, you can avoid them with a small set of reliable habits. This article walks through the most frequent misconceptions encountered in Year 10 Eduqas Statistics, explains why they happen, and shows you step-by-step correction strategies. Think of it as your personal error-spotting guide.

统计学是让数据变得有意义的科学，但即使是最有能力的学生也常常掉入一些可预测的陷阱。从混淆平均数到误读图表，许多错误都源于少数根深蒂固的误解。好消息是，一旦你弄清楚这些陷阱在哪里，就能用一套可靠的思维习惯避开它们。这篇文章将带你梳理 Year 10 Eduqas 统计课程中最常见的误区，解释它们为何发生，并一步一步展示纠正的方法。把它当成你的个人错题导航器。

1. Confusing Mean, Median and Mode | 混淆平均数、中位数和众数

A student calculates the mean and says, “The average is 56, so most people scored around 56.” That sounds reasonable until you realise the data contains an extreme value of 200. The mean was dragged upwards, giving a poor picture of a typical score. Many learners use the word ‘average’ to mean any measure of central tendency, but mean, median and mode answer different questions.

一名学生算出平均数是 56，于是说“平均分是 56，所以大多数人的分数在 56 左右。”听起来挺合理，直到你发现数据中有一个极端值 200。均值被向上拉动了，因此对一个典型分数的描述很不准确。许多学习者用“平均”这个词来指代任何一种集中趋势度量，但平均数、中位数和众数回答的是不同的问题。

The mean is the sum divided by the count; it is sensitive to every data point. The median is the middle value when data are sorted; it resists the pull of outliers and reflects the centre for skewed distributions. The mode is the most frequent value; it can be used for categorical data and is useful when the most common observation matters. Treat them as three different tools: use the median when outliers are present or when the distribution is skewed; use the mean for symmetric data without extreme values; use the mode when dealing with non-numeric data or when you need the most typical category.

平均数是总和除以个数；它对每一个数据点都很敏感。中位数是排序后位于中间的值；它能抵抗异常值的拉动，对于偏态分布能更好地反映中心。众数是出现频率最高的值；它可以用于分类数据，当需要找出最常见的观测值时非常有用。把它们看作三种不同的工具：当存在异常值或分布偏斜时用中位数；当数据对称且没有极端值时用平均数；处理非数值数据或需要找出最典型的类别时用众数。

A quick check: if someone tells you the mean salary in a company is £200 000, ask whether a few very high earners are hiding a low typical income. A small histogram or box plot can instantly reveal whether the mean and median are far apart, signalling that the distribution is skewed.

快速检查：如果有人告诉你一家公司的平均薪资是 20 万英镑，想一想是不是少数高薪者掩盖了大多数人的低薪。一张小小的直方图或箱线图可以瞬间揭示平均数与中位数是否相距甚远，从而表明分布是否偏斜。

2. Misunderstanding the Range and Measures of Spread | 误解极差与离散程度

The range is often the first measure of spread students learn, and it invites two classic mistakes. One is assuming that a larger range always means data are more spread out around the centre; the other is ignoring the fact that the range relies only on the minimum and maximum, throwing away all information about the middle 80% of the data.

极差通常是学生最先接触的离散度量，它很容易引发两个经典错误。一是认为较大的极差就一定意味着数据在中心周围更分散；二是忽略了极差仅仅依赖最小值与最大值，丢弃了数据中间 80% 的全部信息。

Correct the first error by comparing two datasets: the set 1, 50, 50, 99 has a range of 98, but most values cluster tightly around 50. Another set of 40 values from 0 to 100 evenly spaced would also have a range near 100 yet be completely uniform. The range alone tells you nothing about clustering. Use the interquartile range (IQR) or standard deviation alongside the range. The IQR, Q₃ – Q₁, describes the spread of the middle half of the data and is robust to outliers. The standard deviation measures average distance from the mean and is sensitive to all values.

纠正第一个错误可以对比两组数据：数据集 1, 50, 50, 99 的极差是 98，但大多数值紧密围绕在 50 附近。另一组 40 个从 0 到 100 均匀分布的数据，极差也接近 100，但却是完全均匀的。极差本身无法告诉你数据的聚集程度。应当将极差与四分位距 (IQR) 或标准差一起使用。四分位距 Q₃ – Q₁ 描述中间一半数据的散布范围，对异常值稳健。标准差衡量数据与平均值的平均距离，对所有值都敏感。

For the second error, practise reading a box plot: the whiskers show the range, but the box shows the IQR where 50% of the data lie. A small box with long whiskers signals that the extreme values are far from the bulk of the data. Annotate your box plots with the actual range and IQR values to build the habit of interpreting spread in layers.

对于第二个错误，要练习读懂箱线图：须线显示极差，而箱体显示四分位距，那里包含了 50% 的数据。一个箱体很小但须线很长的箱线图，说明极端值离数据主体很远。在箱线图上标注实际的极差和 IQR 数值，养成分层解读离散程度的习惯。

3. Probability Pitfalls: ‘At Least One’ and Independence | 概率误区：“至少一个”与独立性的误解

The phrase ‘at least one’ regularly trips up Year 10 students. When asked, “What is the probability of getting at least one head in two coin tosses?” many quickly answer ½ + ½ = 1 or something similarly flawed. This happens because they add probabilities for overlapping events without subtracting the intersection, or they mistake ‘at least one’ for a simple OR situation.

“至少一个”这个表述经常绊倒 Year 10 学生。当问及“抛两枚硬币至少出现一个正面的概率是多少”时，许多人会迅速回答 ½ + ½ = 1 或类似的错误式子。发生这种情况是因为他们把重叠事件的概率相加却没有减去交集部分，或者把“至少一个”误解为简单的“或”的情形。

The safest route to ‘at least one’ is to use the complement rule: P(at least one) = 1 – P(none). For two fair coins, P(none) = P(both tails) = ½ × ½ = ¼, so the correct answer is 1 – ¼ = ¾. This method avoids the need to enumerate every combination, though listing the sample space {HH, HT, TH, TT} and counting the three favourable outcomes is also powerful at GCSE level.

处理“至少一个”最稳妥的路线是使用补集规则：P(至少一个) = 1 – P(一个都没有)。对于两枚公平硬币，P(一个都没有) = P(两个都是反面) = ½ × ½ = ¼，因此正确答案是 1 – ¼ = ¾。这个方法避免了逐一列举所有组合的需要，不过在 GCSE 阶段，列出样本空间 {正正, 正反, 反正, 反反} 并数出三个有利结果也是一种很有效的方法。

Independence is another common blind spot. Students learn that for independent events P(A ∩ B) = P(A) × P(B), and then apply it indiscriminately. They might multiply the probability of rain and the probability of a bus being late as if weather and traffic have no connection. Always ask: “Does knowing the outcome of the first event change the probability of the second?” If yes, they are dependent, and you need conditional probability. Draw a tree diagram with branching probabilities that change after the first stage to visualise dependence.

独立性是另一个常见的盲区。学生学到对于独立事件有 P(A ∩ B) = P(A) × P(B) 这个公式，然后就不加区分地使用。他们可能会把下雨的概率和公交车晚点的概率直接相乘，仿佛天气和交通状况毫无关联。要总是问自己：“知道了第一个事件的结果，是否会改变第二个事件的概率？”如果是，那它们就是相依的，这时就需要使用条件概率。画一个在第一阶段后分支概率发生变化的树形图，可以把相依性直观地展现出来。

4. Sampling Bias: Not All Samples Are Equal | 抽样偏差：不是所有样本都生而平等

Sampling feels straightforward—pick some people, ask questions—but invisible bias creeps in easily. A survey about sports participation conducted outside a gym will over-represent active individuals. An online poll on screen time will exclude those who use the internet less. Students often judge a sample as “big enough” without checking whether it represents the population fairly.

抽样这件事感觉很简单——找一些人问问就行——但看不见的偏差很容易渗透进来。在健身房外进行的运动参与调查会过度代表活跃人群。一个关于屏幕使用时间的在线调查则会排除那些较少使用互联网的人。学生们常常只因为样本“足够大”就认为它没问题，却没有检查样本是否公平地代表了总体。

The core concept to reinforce is random sampling. A simple random sample gives every member of the population an equal chance of being selected. This is hard to achieve in practice, so stratified sampling is often introduced: divide the population into groups (strata) and sample proportionally from each group. The mistake most learners make is confusing quota sampling with stratified sampling—quota sampling lets the interviewer choose people within a quota, introducing selection bias. Stratified random sampling still requires random selection within each stratum.

需要强化的核心概念是随机抽样。简单随机样本让总体中的每个成员被选中的概率都相等。这在实践中很难实现，因此常常引入分层抽样：将总体分成若干组（层），然后按比例从每一层中抽样。大多数学习者容易犯的错误是将配额抽样和分层抽样混为一谈——配额抽样允许调查员在配额范围内主观选择受访者，这就引入了选择偏差。分层随机抽样仍然需要在每一层内部进行随机选择。

When critiquing a sampling method on an Eduqas exam, always ask: Who might be left out? Is the sample frame complete? How was the sample drawn? A large sample that is biased is no more useful than a small biased sample—volume does not cure bias.

在 Eduqas 考试中评点一种抽样方法时，一定要问：谁可能被遗漏？抽样框完整吗？样本是如何抽取的？一个有偏差的大样本并不比一个有偏差的小样本更好——样本量无法治愈偏差。

5. Misinterpreting Graphs and Charts | 误读图表与统计图

A bar chart with a truncated vertical axis can make a three-percentage-point difference look like a crisis. A pictogram where larger categories are represented by wider icons confuses area with frequency. Students often skip reading the scale, the labels, and the source before jumping to a conclusion. This leads to half-formed arguments and lost marks on interpretation questions.

一张纵轴被截断的条形图能把三个百分点的差异渲染得像一场危机。一张象形图中如果用一个更宽的图标来代表更大的类别，就会让人们混淆面积与频数。学生常常不读刻度、标签和数据来源就匆忙下结论，这会导致不成熟的论证，并在解读类题目中丢分。

Build a three-step reading routine for any graph: (1) Read the title and axis labels—what is being measured? (2) Check the scale—does it start at zero? Are the intervals consistent? (3) Look for distortions—exploded pie slices, 3D effects that tilt area, or images that vary in size in two dimensions. A pie chart showing proportion must have slices that sum to 100%. If a pie chart is meant to show parts of a whole but has slices that do not add to 100%, something is wrong—perhaps it is showing counts rather than percentages, or the data is from a multiple-response question.

建立针对任何图表的“三步阅读法”：(1) 读标题和坐标轴标签——到底在测量什么？(2) 检查刻度——从零开始了吗？间隔一致吗？(3) 寻找扭曲之处——被炸开的饼图扇区、导致面积倾斜的三维效果，或者二维尺寸变化不一的图像。展示比例的饼图，其所有扇区之和必须为 100%。如果一张饼图明明是在表现部分与整体的关系，可各扇区加起来却不是 100%，那一定哪里出了问题——也许它展示的是计数而不是百分比，又或者数据来自一道多选题。

A common mistake with line graphs is to interpret crossing lines as evidence that one rate overtook another on a specific date, without checking whether the data points are monthly, yearly, or interpolated. Always inspect the time intervals and the type of data (discrete, continuous) before declaring a trend.

折线图中一个常见的错误是，把两条线的交叉直接当作在某个具体日期一项速率超过了另一项的证据，而没有检查数据点是每月、每年还是插值得到的。在宣布一个趋势之前，一定要检查时间间隔和数据类型（离散还是连续）。

6. Correlation Does Not Imply Causation | 相关不意味因果

Scatter graphs showing a strong positive correlation often provoke statements like “More ice cream sales cause more drownings.” While the two variables rise together in summer, the lurking variable—hot weather—drives both. Year 10 students frequently treat a high correlation coefficient (r close to 1 or –1) as proof of a causal link, which is a fundamental statistical mistake.

显示强正相关的散点图，常常会引发诸如“冰淇淋销量增加导致溺水人数增加”这样的陈述。虽然这两个变量在夏天确实会同步上升，但背后的潜伏变量——炎热的天气——同时驱动了这两者。Year 10 学生经常把一个很高的相关系数（r 接近 1 或 –1）当作因果关系的证据，这是一个根本性的统计错误。

Teach your brain to ask: Could there be a third factor affecting both? Could the causation run the other way? Is the relationship merely coincidental? A disciplined answer on correlation will always separate the statistical pattern from the real-world explanation. Write: “The scatter graph shows a positive correlation between X and Y, meaning that higher values of X tend to occur with higher values of Y. However, correlation does not imply causation. There could be a third variable, such as temperature, influencing both.”

教会你的大脑去问：是否存在一个同时影响两者的第三因素？因果关系是否有可能反方向运行？这种关系仅仅是巧合吗？一个严谨的关于相关性的答案，总会把统计模式与现实世界中的解释分离开。要这样写：“散点图显示 X 和 Y 之间存在正相关，这意味着较高的 X 值往往与较高的 Y 值同时出现。然而，相关性并不意味着因果关系。可能存在第三个变量，例如温度，同时影响了两者。”

To deepen understanding, practise with examples like “number of firefighters and damage at a fire” (larger fires call more firefighters, not the reverse) and “hours of TV watched and test scores” (could be linked through socio-economic factors). Identifying possible confounding variables is a higher-order skill that earns top marks in the statistical enquiry cycle.

要加深理解，可以用一些例子来练习，比如“消防员的数量与火灾损失”（更大的火灾需要更多的消防员，而不是反过来）和“看电视的小时数与考试成绩”（可能通过社会经济因素相关联）。识别出可能的混杂变量是一项高阶技能，在统计探究周期中能帮你赢得最高分。

7. Mistakes with Grouped Data and Midpoints | 分组数据与组中值的错误使用

When data are given in intervals like 0–10, 10–20, students are taught to use the midpoint to estimate the mean. The misconception arises when they forget that this is an estimate, not an exact calculation, or when they treat the interval boundaries as precise values. Another frequent slip is misaligning the midpoint with unequal class widths: the midpoint of the 20–60 interval is 40, but some learners quickly average 20 and 60 as 40 without checking whether the interval is symmetric—thankfully it is, but they should do it deliberately.

当数据以分组区间呈现（如 0–10、10–20）时，学生被教导用组中值来估计平均数。误区在于他们会忘记这只是一个估计值而非精确计算，或者把组界当作精确值来使用。另一个常见疏漏是在区间宽度不等时错误地对齐组中值：20–60 区间的组中值是 40，有些学生会不假思索地用 20 和 60 算出 40 却不检查区间是否对称——好在这次是对称的，但他们应该有意识地做这一步。

The correct procedure for grouped frequency tables: First, find the midpoint (lower bound + upper bound) ÷ 2 for each class. Multiply each midpoint by its frequency to get an approximate total for that class. Sum these products and divide by the total frequency. This yields an estimated mean. Always state that it is an estimate because we do not know the exact distribution within each class. An advanced common error is trying to find the median from a grouped table by picking the midpoint of the interval containing the median—instead, you must use linear interpolation to find where within that interval the median lies. For Eduqas Year 10, understanding that the median lies within the median class and that simple interpolation with a formula (lower bound + (n/2 – CF_prev)/f_class * class width) gives a better estimate is enough to avoid the crude-midpoint trap.

处理分组频数表的正确步骤：首先，求出每一组的组中值 (下界 + 上界) ÷ 2。将每个组中值乘以该组的频数，得出该组总值的近似量。把所有这些乘积相加，再除以总频数，就得到了估计的平均数。一定要说明这是一个估计值，因为我们不知道每一组内部数据点的精确分布。一个更高阶的常见错误是，试图通过直接取中位数组的组中值来求分组数据的中位数——正确做法是，你必须使用线性插值法来找出中位数在该组区间内的具体位置。对于 Eduqas Year 10 来说，理解中位数位于中位数组内，并且用公式（下界 + (n/2 – 前累计频数)/组频数 × 组距）进行简单插值能得到更好的估计，就足以避免直接用组中值这个粗糙的陷阱了。

8. Misreading Probability Tree Diagrams | 错误解读概率树图

Tree diagrams are wonderfully visual, yet students frequently add probabilities along a branch incorrectly or multiply when they should add. The most common error is building a tree where the second-stage probabilities are written as if the events are independent when they are not—for example, picking two sweets from a bag without replacement and labelling the second pick with the original fractions.

树形图非常直观，但学生经常错误地沿一条分支相加概率，或者在本该加法的时候用了乘法。最常见的错误是，在构建树形图时，把第二阶段的概率当成了独立事件的概率来计算，但实际上事件并非独立——例如，从一个袋子里不放回地取出两颗糖果，却仍然用原来的分数标注第二次抽取的概率。

Adopt a systematic drill: (1) Label the first outcome branches with their probabilities. (2) For the second outcome, ask: “Have the conditions changed?” If without replacement, adjust the denominators and possibly the numerators accordingly. (3) To find the probability of a single combination of outcomes (e.g., red then blue), multiply along the branch. (4) To find the probability of an event that can happen in more than one way (e.g., getting exactly one red), add the probabilities of the mutually exclusive branches that match. Many students mistakenly multiply 0.6 × 0.4 and stop, instead of adding the equivalent result for the other branch (0.4 × 0.6) when order matters.

采用一套系统化的训练步骤：(1) 用概率标注第一级结果的分支。(2) 对于第二级结果，问：“条件改变了吗？” 如果是不放回情境，相应地调整分母，必要时也调整分子。(3) 要找出某种单一结果组合（例如先红后蓝）的概率，沿分支相乘。(4) 要找出一个可能通过不止一种路径发生的事件的概率（例如恰好拿到一颗红色），则把那些匹配的、互斥的分支的概率相加。许多学生会算出 0.6 × 0.4 就停下来了，而忘记了在顺序有关时，还要加上另一条分支的等效结果 (0.4 × 0.6)。

Practice with replacement and without replacement scenarios, and with events that are conditional on a previous outcome, such as weather on consecutive days. Drawing the tree in full before calculating anything helps avoid premature rounding and keeps the work tidy.

用放回与不放回两种情境进行练习，也针对以先前结果为条件的事件进行练习，比如连续几天的天气。在任何计算开始之前，先把完整的树形图画好，能帮助避免过早取整，并让卷面保持整洁。

9. Conditional Probability Confusion | 条件概率混淆

When faced with a statement like “Given that a person is a smoker, what is the probability they are male?”, students often reach for the raw numbers and divide the wrong quantities. They might calculate P(male AND smoker) instead of P(male | smoker). This confusion stems from not translating the word ‘given’ into a restricted sample space.

当面对像“已知某人是吸烟者，这个人是男性的概率是多少？”这样的表述时，学生往往去拿原始数字，却除了错误的数。他们可能会计算 P(男性且吸烟) 而不是 P(男性 | 吸烟)。这种混淆源于没能把“已知”这个词语翻译成对样本空间的限制。

The formula P(A|B) = P(A ∩ B) / P(B) becomes intuitive if you visualise a two-way table. Highlight the row or column representing the condition (B). The total of that row or column is the new denominator. The cell that satisfies both A and B is the numerator. For example, from a table of 200 people:

	Male	Female	Total
Smoker	30	20	50
Non-smoker	70	80	150
Total	100	100	200

The probability that a person is male given they are a smoker is 30/50 = 0.6. The denominator is the smoker total (50), not the overall total (200). Repeat this highlighting exercise until the reflex becomes automatic.

如果你能将双向表可视化，公式 P(A|B) = P(A ∩ B) / P(B) 就会变得直观易懂。把代表条件 (B) 的那一行或那一列高亮出来，这一行或一列的总数就是新的分母。同时满足 A 和 B 的那个单元格的数值就是分子。例如，从一张 200 人的表格来看：已知某人是吸烟者，这个人是男性的概率是 30/50 = 0.6。分母是吸烟者的总数 (50)，而不是总人数 (200)。反复做这种高亮练习，直到形成条件反射。

Another stumbled-upon area is confusing P(A|B) with P(B|A). The “given” order matters. “Probability that it rains given that the forecast was rain” is not the same as “probability the forecast was rain given that it rained.” The first tests the forecast’s predictive accuracy (rain happened when predicted), the second tests how often rain events were captured. Discussing real examples like medical testing (sensitivity vs. positive predictive value) cements this distinction.

另一个容易绊倒的地方是把 P(A|B) 和 P(B|A) 搞混。“给定预报有雨的情况下实际下雨的概率”与“给定实际下雨的情况下预报有雨的概率”是不同的。前者检验的是预报的命中精度（预报有雨时真的下雨了），后者检验的是降雨事件被捕捉到的比例。讨论诸如医学检测之类的真实例子（灵敏度 vs. 阳性预测值）能巩固这种区别。

10. Misapplying the Normal Distribution | 误用正态分布

In Year 10, students begin to encounter the bell-shaped normal distribution. A predictable error is to assume that any symmetric data are normal and that the empirical rule (68-95-99.7%) applies automatically. Another mistake is to think that the mean is always at the highest point of the curve for every dataset, or that the normal distribution can be used for small, discrete datasets without checking conditions.

在 Year 10 阶段，学生们开始接触钟形的正态分布。一个可预见的错误是，假设任何对称的数据都是正态的，并且经验法则（68-95-99.7%）会自动适用。另一个错误是，以为在任何数据集里平均数总是处于曲线的最高点，或者不经条件检查就把正态分布用在小型的、离散的数据集上。

A dataset must be approximately symmetric and mound-shaped, with a single peak and tails that die out smoothly. The empirical rule states that for a normal distribution, about 68% of data lie within 1 standard deviation of the mean, 95% within 2, and 99.7% within 3. This is not true for skewed data or data with heavy tails. When Eduqas questions provide a mean and standard deviation and say “assuming a normal distribution,” use the rule; otherwise, do not overextend it.

数据集必须大致对称且呈山丘状，有一个单峰以及平滑衰减的尾部。经验法则表明，对于正态分布，大约 68% 的数据落在距离平均数 1 个标准差的范围内，95% 落在 2 个标准差内，99.7% 落在 3 个标准差内。但这对于偏斜数据或有厚尾的数据并不成立。当 Eduqas 的题目给出平均数和标准差，并写明“假设服从正态分布”时，再使用该法则；否则不要随意套用。

Another nuance: the standard deviation is a ruler for the normal curve. A value of 0.7 standard deviations above the mean is unremarkable; 2.5 standard deviations above is unusual. Students sometimes treat any value above the mean as an outlier, forgetting that variability is normal. Plotting data on a histogram with the normal curve overlaid helps develop an eye for genuine anomalies versus natural variation.

另一个细微之处：标准差是正态曲线的一把尺子。比平均数高 0.7 个标准差的值平平无奇；高 2.5 个标准差的值则是不寻常的。学生有时会将任何高于平均数的值都视为异常值，忘记了变异性是正常的。在直方图上叠加正态曲线，有助于培养识别真正异常与自然波动之间区别的眼力。

11. Confusing Experimental and Theoretical Probability | 混淆实验概率与理论概率

After rolling a die 30 times and getting only two sixes, a student declares the die is biased. That might be true, but 30 rolls is a small sample. Another learner carries out a sophisticated simulation and then discards the simulated results because they do not match the theoretical probability they expected, over-trusting the theory. Both errors stem from not grasping the relationship between short-term variation and long-run stability.

掷了 30 次骰子只得到两个 6，一名学生就宣称这枚骰子有偏差。骰子可能确实有问题，但 30 次是一个小样本。另一位学习者完成了一项精密模拟，因为模拟结果和自己预期的理论概率不符就将其丢弃，过度相信理论。这两种错误都源于没有理解短期变异与长期稳定性之间的关系。

Experimental probability is an estimate; it becomes more accurate as the number of trials grows (the law of large numbers). In the short run, streaks and gaps are normal. A fair coin can easily produce seven heads in ten flips. The key is not to jump to conclusions. Instead, carry out a hypothesis test frame of mind: state a null hypothesis (the die is fair), collect data, and ask how surprising the result would be under that hypothesis. In Year 10, this might be informal—comparing relative frequency to theoretical probability and noting the sample size.

实验概率是一个估计值；随着试验次数的增加，它会越来越精确（大数定律）。在短期内，连续出现或长时间不出现某结果是正常的。一枚公平硬币抛十次出现七个正面是很可能发生的。关键是不能急于下结论，而是要用假设检验的思维框架：提出一个原假设（骰子是公平的），收集数据，然后问自己在这个假设下出现这样的结果有多令人惊讶。在 Year 10 阶段，这可以是非正式的——比较相对频率与理论概率，并注意样本量的大小。

A useful classroom mantra: “Probability describes what we expect to happen in the long run, not what must happen in any single experiment.” Write this at the top of every probability assignment to resist the temptation to see patterns that are not there.

一句有用的课堂箴言：“概率描述的是长期内我们期望会发生的事情，而不是在任何单独一次实验中一定会发生的事情。” 把这句话写在每次概率作业的最上方，以抵抗那种看见并不存在的模式的诱惑。

12. Misjudging Statistical Significance and Real-World Importance | 错判统计显著性与实际重要性

In the enquiry cycle, Year 10 students are expected to evaluate findings. A common misstep is to treat a small p-value or a large difference as automatically important. A weight-loss drug might produce a statistically significant reduction of 0.2 kg on average—a result that is unlikely by chance, yet clinically meaningless. Conversely, a noticeable difference in a small pilot study might be dismissed because the sample was tiny, even though it warrants further investigation.

在探究周期中，Year 10 学生需要对研究结果进行评价。一个常见的失误是，把很小的 p 值或很大的差异自动看作有重要意义。一种减肥药也许能产生平均 0.2 公斤的统计显著下降——这个结果靠偶然出现的可能性很小，但在临床上毫无意义。相反，一个小型试点研究中出现的显著差异，却可能因为样本量小而被轻视，即便它值得进一步探究。

Separate the statistical from the practical. Statistical significance tells you whether an observed effect is likely to be genuine (not just noise), but effect size tells you how big it is. In your write-ups, always report both where possible: “The difference between groups was 3.2 percentage points, and a hypothesis test suggested this is unlikely to be due to chance (p < 0.05). However, the effect size is small, so the practical impact may be limited." This nuanced language distinguishes high-level answers.

要把统计上的和实际上的分开。统计显著性告诉你观察到的效应是否是真实的（而不仅仅是噪声），但效应量告诉你它有多大。在写作报告中，尽可能两者都报告：“组间差异为 3.2 个百分点，假设检验表明这不太可能由偶然导致 (p < 0.05)。然而，效应量很小，因此实际影响可能有限。”这种细腻的表述能让高层次答案脱颖而出。

Also, remember that a hypothesis test depends on the null and alternative hypotheses correctly framed. A one-tailed test is only appropriate when there is a clear directional expectation before seeing the data. Using a one-tailed test to chase a significant result after looking at the data inflates the Type I error risk. Learn to articulate why two

Published by TutorHao | Year 10 统计 Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

Common Misconceptions in Statistics and How to Correct Them | 统计常见误区与纠正方法