Edexcel ALevel数学 统计 正态分布 假设检验

Edexcel ALevel数学 统计 正态分布 假设检验

统计是 Edexcel A-Level 数学第三张卷子的前半张，50 分。它不像纯数那样靠推导堆分，也不像力学那样靠受力图拿分，分数集中在几类固定动作上：把数据整理成能比较的形式、认出该用哪个分布、把检验结论写成一句能读懂的话。下面按考纲顺序拆成十三个部分，每部分配一段英文对照，方便连着准备双语术语。

Statistics is the first half of the third Edexcel A-Level Mathematics paper and carries 50 marks. It does not reward long derivations the way pure mathematics does, and it does not rely on force diagrams the way mechanics does. The marks sit in a handful of repeatable actions: turning raw data into something comparable, choosing the right distribution, and writing a conclusion a reader can act on. The thirteen parts below follow the syllabus order, each with one English paragraph for terminology practice.

Paper 3 的统计部分只有 50 分，时间是最大变量

Pearson Edexcel 的 A-Level Mathematics（9MA0）共三张卷：Pure Mathematics 1、Pure Mathematics 2、Statistics and Mechanics。每张 2 小时、100 分，各占 33.33%。第三张卷分两节，Section A 统计 50 分，Section B 力学 50 分，这是 Edexcel 9MA0 specification 里写明的结构。这意味着统计部分大约只有一小时，平均 1.2 分钟拿一分。很多学生以为统计难在公式，实际上时间大多耗在抄数据、按计算器和反复读题上，一半的失分不是不会，而是没算完。平时练习就按 1.2 分钟一分掐表，先拿下读图与描述题，把最后十五分钟整块留给两道假设检验。

Pearson Edexcel's A-Level Mathematics (9MA0) is examined in three two-hour papers of 100 marks each, contributing 33.33% apiece. Paper 3 is split into Section A on statistics and Section B on mechanics, 50 marks each, as set out in the Edexcel 9MA0 specification. The statistics half therefore has about an hour, which works out at roughly 1.2 minutes per mark. Students tend to describe statistics as a memory subject, yet the time actually disappears into copying data, keying values into a calculator and re-reading the question.

平均数、标准差与编码变换

Edexcel 公式册给出两类离散程度：分母为 n 的总体标准差和分母为 n-1 的样本标准差 s。题目说 from a sample of 20 observations 时用 s，说 the whole population 时用总体标准差，这一处判断错误会让后面所有检验统计量一起错。编码 y = (x - a)/b 的规则要背成一个动作：均值同步变成 (x̄ - a)/b，标准差只除以 |b|，加减常数 a 不改变离散程度。拿一组均值 42、标准差 5 的数据做例子，令 y = (x - 10)/2，新均值是 16，新标准差是 2.5，整体像被挪了位置又缩了一半。

The Edexcel formula booklet lists two measures of spread: the population standard deviation with divisor n and the sample standard deviation s with divisor n-1. If the question says a sample of twenty observations, you use s. Coding behaves predictably: under y = (x - a)/b the mean becomes (x̄ - a)/b while the standard deviation is simply divided by |b|, because shifting every value by a constant changes no distances.

直方图的频数密度与线性插值

不等组距的直方图画的是 frequency density，也就是频数除以组距，面积才代表频数；直接比柱高会把宽组看小、窄组看大。中位数和四分位数在累积频数曲线或累积频数表上做线性插值，取相邻两点的坐标按比例算，别拿组中值硬套。常见的错误是把 (n+1)/2 与 n/2 两种口径混着用，同一道题里要统一，并在答卷上写清你用的是哪一个。分组宽度也要事先想好，组距取 5 还是 10 会改变柱子的形状，自己分组时挑能整除数据的宽度，读图误差小一些。

Histograms with unequal class widths plot frequency density, frequency divided by class width, so it is the area of a bar that represents frequency. Comparing bar heights directly makes wide classes look small and narrow ones look large. Medians and quartiles come from linear interpolation on the cumulative frequency curve: take the two neighbouring points and interpolate proportionally. Mixing the (n+1)/2 and n/2 conventions inside one question is a familiar way to lose method marks.

离群值的 1.5 IQR 判定与箱线图

判定规则是固定的：小于 Q1 - 1.5 IQR 或大于 Q3 + 1.5 IQR 的观测值算离群值，画箱线图时用叉号或单独的点标出来。剔除与否要看题目问什么，问 describe the distribution 时保留并单独说明，问 cleaned mean 时才删。删完再算平均数，要写清删掉了几个值，否则过程分拿不到，后面的标准差也会跟着错。箱线图与直方图描述的是同一批数据，结论要能对上：箱线图看中位数位置与偏斜方向，直方图看峰的个数和尾巴长短，两张图的说法互相矛盾，说明中间某一步算错了。

The rule is fixed: a value below Q1 minus 1.5 IQR or above Q3 plus 1.5 IQR counts as an outlier and is marked with a cross or a separate point on the box plot. Whether you remove it depends on the wording. For a description of the distribution you keep it and comment on it; for a cleaned mean you delete it and state how many values went.

相关系数与最小二乘回归线的解释

回归线写作 y = a + bx，其中 b = Sxy / Sxx，a = ȳ - b x̄，公式册里两个求和项的定义要照抄，不要用计算器统计模式里的现成数字反推。斜率必须带单位解释：x 每增加一个单位，y 平均变化 b 个单位。相关系数 r 只衡量线性关系的强弱与方向，r 接近 0 说明没有线性关系，不等于两个变量无关。数据范围内预测叫内插，超出范围叫外推，外推的结论要写保留意见。举个具体的界：x 的观测范围是 3 到 11，取 7 预测属内插，取 15 预测属外推，后者的点估计还能算，因果解释就站不住了。

The least squares line is written y = a + bx with b = Sxy/Sxx and a = ȳ - b x̄. Interpret the gradient with units: each additional unit of x changes y by b on average. The correlation coefficient r measures only the strength and direction of a linear relationship, so a value near zero rules out linearity, not association. Prediction inside the observed range is interpolation and is defensible; prediction outside it is extrapolation and needs a caveat.

条件概率、独立与互斥不是一回事

P(A|B) = P(A∩B) / P(B)，独立的条件是 P(A∩B) = P(A)P(B)，两条式子要能在同一道题里互相推出。互斥和独立经常被混：两个概率都大于 0 的事件如果互斥，一个发生另一个就不可能发生，所以一定不独立。这是阅卷时错得最多的一处，学生常在维恩图上把两个圆分开放，转头就写 independent。树状图的乘法用在连续两支，加法用在同一层的不同分支；把 P(B|A) 与 P(A|B) 读反是另一处高频错误，题目给的条件顺序要照着用。

Conditional probability is P(A|B) = P(A ∩ B)/P(B), and independence means P(A ∩ B) = P(A)P(B). Mutually exclusive and independent are different properties: if two events with positive probability are mutually exclusive, one rules out the other, so they cannot be independent. Drawing two separate circles on a Venn diagram and labelling them independent is a common slip.

离散随机变量的期望与方差

E(X) 用 Σ x P(X = x)，Var(X) 用 E(X²) - [E(X)]²，后一个式子里 E(X²) 要先把每个值平方再加权，不是先求期望再平方。线性变换的法则是 E(aX + b) = aE(X) + b，Var(aX + b) = a² Var(X)，常数项在方差里直接消失。概率分布表里所有概率加起来必须等于 1，题目往往就在这里埋一个待定的常数 k。做题时先把 x² 与 x²P(X = x) 两列补在表下面，求和后直接套公式，比在草稿纸上记简化式稳当；算出来的方差如果是 0，说明这个随机变量其实没有随机性。

For a discrete random variable, E(X) = Σ x P(X = x) and Var(X) = E(X²) - [E(X)]². Square each value before weighting; squaring the mean afterwards gives a different and wrong number. Under a linear transformation E(aX + b) = aE(X) + b while Var(aX + b) = a² Var(X), since a constant shift changes no spread. The probabilities in a distribution table must sum to one, which is usually how an unknown constant k is found.

二项分布的四条前提与计算器写法

X ~ B(n, p) 成立要四条同时满足：试验次数 n 固定、每次只有成功与失败两种结果、每次成功的概率 p 相同、各次试验相互独立，少写一条就丢分。单个值用 binompdf，区间用 binomcdf，答案保留 3 位有效数字，并把原始表达式写出来，例如 P(X ≥ 8) = 1 - P(X ≤ 7)，批卷先看表达式再看数。很多学生会按计算器却不会写出 X ~ B(20, 0.3) 这一行，白丢一分。判断独立性最快的抓手是看抽样方式：十个球里连抽两个不放回，每次成功的概率 p 会变，这时候二项分布就不适用了。

Four conditions must all hold for X ~ B(n, p): a fixed number of trials, two outcomes per trial, a constant probability of success, and independence between trials. Use the binomial pdf for a single value and the cdf for an interval, quote three significant figures, and write the expression you keyed in, such as P(X ≥ 8) = 1 - P(X ≤ 7). The distribution statement X ~ B(20, 0.3) is worth a mark on its own.

正态分布的标准化与逆正态

X ~ N(μ, σ²) 里的第二个参数是方差而不是标准差，抄错这一步后面全错。标准化用 z = (x - μ) / σ，再查表或走计算器的 Normal CD；已知概率反求 x 时用 Inverse Normal，注意尾部的选择，Area 是从左尾开始的累积概率。对称性可以直接用：P(X > μ + a) = P(X < μ - a)，这类题用对称性比查表快，也少一层四舍五入误差。算完回头检查量级：要找的 x 在均值右侧，z 就应当是正的，符号反了说明公式里减数的顺序写颠倒了。

In X ~ N(μ, σ²) the second parameter is the variance, not the standard deviation. Standardise with z = (x - μ)/σ, and when the probability is given and x is wanted, use the inverse normal with the area taken from the left tail. Symmetry shortcuts such as P(X > μ + a) = P(X < μ - a) are quicker than table lookups and avoid rounding drift.

正态近似二项分布与连续性修正

n 较大、p 接近 0.5 时，B(n, p) 可以用 N(np, np(1-p)) 近似，均值与方差直接代入，再对整数边界做连续性修正：P(X ≤ 8) 变成 P(Y < 8.5)，P(X ≥ 8) 变成 P(Y > 7.5)。Edexcel 公式册不给适用阈值，题目一般自己提示 use a normal approximation，判断依据是 np 与 n(1-p) 都不算小；各本教材写的经验值并不一致，答题时跟着题干走，不要自己设一个硬门槛。修正的符号按包含边界就往外扩半格来记：P(X ≤ 8) 含 8，所以用 8.5 作上界；P(X > 8) 不含 8，所以 8.5 当下界。

When n is large and p is near 0.5, B(n, p) can be approximated by N(np, np(1-p)), with a continuity correction applied to integer boundaries: P(X ≤ 8) becomes P(Y < 8.5) and P(X ≥ 8) becomes P(Y > 7.5). The Edexcel formula booklet publishes no threshold, and questions normally tell you to use a normal approximation, so follow the wording rather than inventing a cutoff.

假设检验的五个步骤与结论写法

二项检验的骨架是固定的：写 H0 与 H1（单尾还是双尾看题干）、写出检验统计量服从的分布、在给定的显著性水平下算临界域或 p 值、比较、再用一句话落到情境里。双尾检验把 α 平分到两端，5% 就变成两端各 2.5%。结论不能写接受 H0，正确说法是在 5% 显著性水平下没有充分证据拒绝 H0，这两句话在阅卷标准里是两回事。有充分证据时，要说明 p 是小于还是大于原假设里的取值。临界域的做法是把累积概率逐项读到刚超过 α 的位置，p 值法则是直接算观测值及更极端情况的累积概率，两种写法都给分，同一道题里只用一种。

A binomial hypothesis test follows a fixed skeleton: state H0 and H1, name the distribution of the test statistic, find the critical region or the p-value at the given significance level, compare, then write the conclusion in context. Two-tailed tests split alpha between the tails, so 5% becomes 2.5% at each end. Never write that you accept H0; the correct wording is that there is insufficient evidence to reject it.

十二周复习节奏：按题型推进而不是按章节

前四周把统计知识点过一遍，每节配十道短题，重点放在判断该用哪个分布和整理数据；第五到第八周专做 Section A，限时 55 分钟，错题按错误类型归档而不是按题号归档；第九到第十一周做整卷，并把大样本数据集（large data set）的题单独过一遍，Edexcel 考纲要求学生对考试局指定的数据集做比较与解释；第十二周只看公式册位置、计算器流程与错题本，不再开新题。每周留一次二十分钟的口头复述，不看笔记讲清二项分布与正态分布的差别，讲不顺的地方就是下周要补的章节。

A twelve-week plan works better organised by question type than by chapter. Spend the first four weeks covering content with ten short questions per topic, focusing on distribution choice and data handling, then four weeks on timed Section A practice at 55 minutes with mistakes sorted by error type. The next three weeks go to full papers plus the large data set questions the Edexcel specification expects students to interpret and compare. The final week is formula booklet, calculator routines and the error log only.

计算器流程与考场动作

统计卷的分差常常出在操作上：进检验模式前先清空上一次的列表，二项与正态的输入位置不同，p 值不要手动四舍五入到四位以上再拿去比较。答卷上保留原始表达式与中间值，阅卷看的是过程。常见的错误是把单尾的 p 值直接乘 2 当成双尾结果，而双尾的临界域要两端各算一次再合并，遇到恰好落在边界上的情况还要说明取不取等号。开考前顺手把计算器恢复成常规显示模式，清掉上一次留下的列表与统计内存，这一步花十秒，能省掉后面重算一整题的麻烦。

Marks are often lost to keystrokes rather than mathematics: clear previous lists before entering a new test, check which menu holds the binomial and normal functions, and compare unrounded p-values. Keep the expression and intermediate values in your written answer, because the mark scheme reads the method. Doubling a one-tailed p-value is not the same as finding a genuine two-tailed critical region, and a value sitting exactly on the boundary needs a statement about whether the equality is included.

海外直订Higher Mathematics for AQA GCSE 高等数学AQA GCSE¥645.0📱 去京东购买京东精选

更多咨询请联系16621398022（同微信）

需要一对一辅导？Need One-on-One Tutoring?

A-Level / GCSE / IB / KS2-KS3 全科辅导

16621398022 同微信

Visit aleveler.com