A-Level数学 假设检验 显著性水平 临界值
假设检验是A-Level数学统计模块的高频考点，也是很多同学觉得抽象难懂的部分。它看似绕口，其实核心逻辑只有一句话：先假定旧说法成立，再判断手上的样本数据是否"离谱"到无法用巧合解释。本文用中英双语拆解假设检验的完整流程，从零假设写到结论措辞，附一道完整的二项分布例题，帮你一次掌握。
Hypothesis testing is one of the most frequently examined topics in A-Level Maths statistics papers, yet many students find it abstract and confusing. The underlying logic is only one sentence: assume the old claim is true, then judge whether the sample data you collected is too surprising to be explained by chance. This article breaks down the full procedure in Chinese and English, from writing hypotheses to wording the final conclusion, with a complete binomial worked example, so you can master it in one pass.
一、什么是假设检验 What is a hypothesis test
假设检验是一种用样本数据对总体参数作判断的正式统计方法，思想来源是反证法。先假定零假设成立，算出在零假设为真的前提下观察到当前样本、甚至更极端样本的概率；如果这个概率非常小，说明样本与零假设严重不符，我们就有理由拒绝零假设。要特别记住：检验永远不能"证明"某个说法，它只能告诉我们证据够不够强。比如怀疑一枚硬币不均匀，可以抛50次并记录正面次数，再检验"正面概率等于0.5"这个说法是否还站得住。
A hypothesis test is a formal statistical procedure that uses sample data to make a judgement about a population parameter, and its logic comes from proof by contradiction. We first assume the null hypothesis is true, then calculate the probability of observing the sample result, or something even more extreme, under that assumption. If this probability is very small, the sample is badly inconsistent with the null hypothesis, so we have reason to reject it. Always remember that a test never proves a claim; it only tells us whether the evidence is strong enough. If you suspect a coin is biased, for example, you can toss it 50 times and test whether the claim that the probability of heads equals 0.5 still stands.
二、零假设与备择假设 Null and alternative hypotheses
零假设用H0表示，是被检验的、当前被接受的陈述，永远写成等号形式，例如H0: p等于0.45，或H0: μ等于30。备择假设用H1表示，是你怀疑或想要证明的方向，只能取三种形式：小于、大于、不等于。答题时H0和H1必须用总体参数符号书写，并且H1要带明确方向。另外，检验的结论只能说"拒绝零假设"或"不能拒绝零假设"，不能说"接受零假设"，因为证据不足不代表零假设一定正确。
The null hypothesis, written H0, is the currently accepted statement that is put on trial, and it is always written with an equals sign, such as H0: p equals 0.45, or H0: μ equals 30. The alternative hypothesis, written H1, is the direction you suspect or want to demonstrate, and it can only take three forms: less than, greater than, or not equal to. In an exam answer both hypotheses must be written with the population parameter symbol, and H1 must show a clear direction. A test can only reject the null hypothesis or fail to reject it; you should never say the null is accepted, because weak evidence does not mean the null is actually correct.
三、检验统计量与显著性水平 Test statistic and significance level
检验统计量是由样本算出的一个数：在二项检验里通常是成功次数X，在正态检验里通常是样本均值。判断X是否"极端"，需要把对应概率与显著性水平比较。显著性水平用希腊字母α表示，常见取值是0.05、0.01、0.025，它的含义是：当零假设实际为真时，你愿意承担多大的风险去错误拒绝它。考试还常考p值：p值是在零假设为真时得到当前样本或更极端结果的概率。判据很简洁，p值小于α就拒绝零假设，大于等于α就不能拒绝。
The test statistic is a number calculated from the sample: usually the number of successes X in a binomial test, or the sample mean in a normal test. To judge whether X is extreme, you compare its probability with the significance level. The significance level is written with the Greek letter α and commonly takes values 0.05, 0.01 or 0.025; it is the risk you accept of wrongly rejecting the null hypothesis when the null is in fact true. Exams also test the p-value, the probability, assuming the null is true, of obtaining the observed result or something more extreme. The rule is simple: if the p-value is smaller than α, reject the null; if it is greater than or equal to α, fail to reject.
四、单尾检验与双尾检验 One-tailed and two-tailed tests
当备择假设带方向，比如H1写成p大于0.45或p小于0.45，检验就是单尾的，极端概率只在对应的一侧累计。当备择假设写成p不等于0.45，检验是双尾的，显著性水平要分成两半，左右两侧各用α的一半，即α/2。选择哪一种取决于题目问法：问"是否增加了"或"是否减少了"用单尾检验，问"是否改变了"或"是否与原来不同"用双尾检验。许多同学在第一步就把H1方向写反，导致后面整道题全错，务必先圈出题干中的关键词再下笔。
When the alternative hypothesis has a direction, such as H1: p greater than 0.45 or H1: p less than 0.45, the test is one-tailed and the extreme probability is collected on that single side. When the alternative is written as p not equal to 0.45, the test is two-tailed, and the significance level must be split in half, with α over 2 on each side. Which one you use depends on the wording of the question: increased or decreased points to a one-tailed test, while changed or different from points to a two-tailed test. Many students write the direction of H1 incorrectly in the very first step and then lose the entire question, so circle the key words in the stem before you start.
五、临界值与拒绝域 Critical value and rejection region
拒绝域又称临界域，是检验统计量的取值集合，样本结果落入其中就要拒绝零假设；临界值是拒绝域的边界。求上尾临界值的方法是：找出最小的整数c，使X大于等于c的概率不超过α。例如下面的例题中临界值为14，意思是只要样本里有14人或更多人偏好新口味，就能在5%水平拒绝零假设；13人则不够。下尾检验对称处理：找最大的整数c，使X小于等于c的概率不超过α。二项分布是离散分布，所以临界值法要检查边界两边的概率，不能凭感觉取整。
The rejection region, also called the critical region, is the set of values of the test statistic that would lead you to reject the null hypothesis, and the critical value is its boundary. For an upper-tail test, find the smallest integer c such that the probability of X being greater than or equal to c is no more than α. In the worked example below, the critical value is 14, meaning that if 14 or more people in the sample prefer the new flavour, you reject the null at the 5% level, while 13 is not enough. The lower-tail test is symmetric: find the largest integer c such that the probability of X being less than or equal to c is no more than α. Because the binomial distribution is discrete, you must check the probabilities on both sides of the boundary instead of rounding by intuition.
六、完整例题：二项分布假设检验 Worked example with the binomial distribution
题目：某饮料公司声称45%的消费者更喜欢新配方。随机调查20人，其中13人表示更喜欢新配方。在5%显著性水平下检验，喜欢新配方的消费者比例是否增加了。
Question: a drinks company claims that 45% of consumers prefer the new recipe. In a random sample of 20 people, 13 say they prefer the new recipe. Test at the 5% significance level whether the proportion of consumers who prefer the new recipe has increased.
解答：设X为样本中更喜欢新配方的人数，则X服从二项分布B(20, p)。第一步写假设，H0: p等于0.45，H1: p大于0.45，这是上尾单尾检验。第二步计算概率，在零假设成立时，P(X大于等于13)等于1减去P(X小于等于12)，查表或用计算器得1减0.94197约等于0.05803。第三步与显著性水平比较，0.05803大于0.05，说明样本结果不在拒绝域内。第四步下结论：没有充分证据拒绝零假设，不能认为喜欢新配方的比例增加了。如果用临界值法复核，P(X大于等于14)约等于0.0214小于0.05，因此临界值为14，拒绝域是X大于等于14，观察值13不在其中，两种方法结论一致。再想深一层：若显著性水平改为10%，因为0.0580小于0.10，结论会翻转为拒绝零假设，可见显著性水平的选择会直接影响结论。
Solution: let X be the number of people in the sample who prefer the new recipe, so X follows the binomial distribution B(20, p). Step one, write the hypotheses: H0: p equals 0.45 and H1: p greater than 0.45, an upper-tail one-tailed test. Step two, calculate the probability: under the null, P(X greater than or equal to 13) equals 1 minus P(X less than or equal to 12), which is 1 minus 0.94197, about 0.05803, from tables or a calculator. Step three, compare with the significance level: 0.05803 is greater than 0.05, so the sample result does not lie in the rejection region. Step four, conclude: there is insufficient evidence to reject the null hypothesis, so we cannot claim that the proportion of consumers who prefer the new recipe has increased. Cross-check with the critical value method: P(X greater than or equal to 14) is about 0.0214, which is below 0.05, so the critical value is 14 and the rejection region is X greater than or equal to 14; the observed value 13 is not inside it, and both methods agree. One more insight: if the significance level were 10%, then 0.0580 would be smaller than 0.10 and the conclusion would flip to rejecting the null, which shows how the choice of significance level directly changes the outcome.
七、第一类错误与第二类错误 Type I and Type II errors
第一类错误是零假设实际为真却被我们拒绝，它的发生概率恰好等于显著性水平α，因为α的定义就是"零假设为真却仍被拒绝"时愿意承担的风险。第二类错误是零假设实际为假却没有被拒绝，概率通常记为β，A-Level多数考卷只要求默写定义和理解含义，不要求计算β。调低α能减小第一类错误，但会让第二类错误变大，两者互相牵制；最有效的缓解办法是增大样本量，让两类错误同时下降。结论文末还要写清"拒绝"与"不能拒绝"的区别，一字之差就会丢采分点。
A Type I error is rejecting the null hypothesis when it is actually true, and its probability is exactly the significance level α, because α is defined as the risk you accept of rejecting a null that is in fact true. A Type II error is failing to reject the null hypothesis when it is actually false, and its probability is usually written β; most A-Level papers only ask for the definition and the concept rather than a calculation of β. Lowering α reduces the chance of a Type I error but increases the chance of a Type II error, since the two pull against each other, and the most effective remedy is a larger sample, which reduces both at the same time. In your final sentence, also keep the distinction between reject and fail to reject clear, because confusing the two wording loses marking points.
八、正态分布的假设检验 Hypothesis testing with the normal distribution
当总体服从正态分布且方差已知时，可以用样本均值构造检验统计量。设H0: μ等于μ0，若H1为μ大于μ0，则计算z等于(x拔减μ0)除以(σ除以根号n)，再把z与标准正态分布的上尾临界值比较；常用的双侧5%临界值是1.96，单侧5%临界值是1.645。若总体方差未知但样本量足够大，可用样本标准差估计σ，本质思路不变。正态检验的易错点是方向：H1用大于号时看右尾，用小于号时看左尾，用不等于号时双侧都要看，且临界值取1.96而不是1.645。
When the population follows a normal distribution and the variance is known, you can use the sample mean to build the test statistic. Set H0: μ equals μ0; if H1 is μ greater than μ0, calculate z equal to (x bar minus μ0) divided by (σ divided by root n), then compare z with the upper-tail critical value of the standard normal distribution. The common two-tailed 5% critical value is 1.96, while the one-tailed 5% critical value is 1.645. When the population variance is unknown but the sample is large, estimate σ with the sample standard deviation and the idea is unchanged. The usual trap is the direction: H1 with a greater-than sign looks at the right tail, a less-than sign looks at the left tail, and a not-equal sign needs both tails with 1.96 instead of 1.645.
九、考试技巧与常见错误 Exam technique and common pitfalls
按固定套路答题最稳：第一行写明X的分布；第二行写出H0与H1，参数必须带具体数值和方向；第三行列出P(X大于等于观察值)的计算并保留足够精度；第四行把概率与α明确比较；第五行用完整句子回到题目情境下结论。常见错误包括：上尾检验写成下尾、H1漏写方向、只算观察值本身的概率而漏掉更极端的结果、没有把概率与显著性水平作显式比较、结论只说"拒绝"而不解释证据强度。计算器上使用二项累积分布功能，一次即可得到P(X小于等于k)。另外考试常把临界值表印在公式册里，先确认自己查的是上尾表还是下尾表再抄数。
Answer in a fixed routine to stay safe: line one, state the distribution of X; line two, write H0 and H1 with the parameter value and direction included; line three, show the calculation of P(X greater than or equal to the observed value) with enough precision; line four, compare the probability with α explicitly; line five, conclude in a full sentence that returns to the context of the question. Common mistakes include testing the wrong tail, forgetting the direction in H1, calculating only the probability of the observed value while missing the more extreme outcomes, failing to compare the probability with the significance level, and ending with reject only without describing the strength of the evidence. Use the binomial cumulative distribution function on your calculator to obtain P(X less than or equal to k) in one step. Critical value tables are usually printed in the formula booklet, so check whether you are reading an upper-tail or a lower-tail table before copying the number.
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需要一对一辅导？Need One-on-One Tutoring?
A-Level / GCSE / IB / KS 全科辅导，中英双语授课
或截图后用「扫一扫 → 相册」选择本图