📚 Investigation 2: Confidence Levels and Intervals | 探究二：置信水平与置信区间
In statistical inference, we rarely know the true value of a population parameter like the mean weight of all IB diploma candidates or the proportion of defective components from a production line. Instead, we take a sample and use it to estimate the parameter. A point estimate such as the sample mean is useful, but it gives no indication of its precision. Confidence intervals address this by providing a range of plausible values for the parameter, together with a confidence level that expresses how certain we are that the range captures the true value. This investigation explores the relationship between confidence levels and intervals, how they are constructed, and what factors influence their width.
在统计推断中，我们很少知道总体参数的真实值，例如所有 IB 文凭考生的平均体重或生产线上缺陷部件的比例。我们通常抽取一个样本，用它来估计参数。像样本均值这样的点估计很有用，但并未反映其精确度。置信区间通过给出参数的一个合理取值范围，并附以置信水平来表达我们有多大的把握认为该范围包含了真实值。本次探究将探讨置信水平与置信区间之间的关系、如何构建它们以及哪些因素会影响其宽度。
1. What Is a Confidence Interval? | 什么是置信区间？
data. It takes the form: point estimate ± margin of error. For a population mean μ, the 95% confidence interval using a known σ is x̄ ± z* × (σ/√n). The probability that the interval contains μ is the confidence level, say 0.95, in the long run over many repeated samples.
置信区间是由样本数据计算得出的总体参数的区间估计，其形式为：点估计 ± 误差范围。对于在已知总体标准差 σ 的情况下对总体均值 μ 进行的估计，95% 置信区间为 x̄ ± z* × (σ/√n)。从长期多次重复抽样来看，该区间包含 μ 的概率就是置信水平，例如 0.95。
The endpoints of the interval are random variables because they depend on the sample. The population parameter is fixed. So a 95% confidence level does not mean there is a 95% chance that μ lies in a single computed interval; rather, if we repeated the sampling procedure many times, about 95% of the resulting intervals would capture μ.
区间的端点依赖于样本，因而是随机变量，总体参数则是固定的。因此，95% 的置信水平并不意味着对于计算出的单个区间，μ 有 95% 的概率落在其中；而是说，如果我们多次重复抽样，则大约有 95% 的区间会包含 μ。
2. Understanding Confidence Levels | 理解置信水平
The confidence level, denoted as (1 - α)×100%, is chosen by the researcher. Common choices are 90%, 95%, and 99% in each tail. The critical value z* is the z-score that cuts off the upper tail area α/2.
置信水平用 (1 - α)×100% 表示，由研究者选定。常见的选择有 90%、95% 和 99%。其补数 α 是显著性水平，在双侧区间中会平分到左右两侧尾部，每侧尾部面积为 α/2。临界值 z* 就是对应上侧尾部面积为 α/2 的 z 分数。
For a 95% α/2 = 0.025. The critical value from the standard normal distribution is z* = 1.96. For 90% it is 1.645, and for 99% ad to larger critical values, which increases the margin of error and widens the interval.
对于 95% 置信区间，α = 0.05，α/2 = 0.025，标准正态分布的临界值为 z* = 1.96。90% 的临界值为 1.645，99% 的临界值为 2.576。更高的置信水平需要更大的临界值，这会增加误差范围并使区间变宽。
3. The Central Limit Theorem and Sampling Distributions | 中心极限定理与抽样分布
Confidence intervals for the mean rely on the sampling distribution of the sample mean x̄ for a sufficiently large sample size (n ≥ 30 is a common guideline), the distribution of x̄ is approximately normal with mean μ and standard error σ/√n, regardless of the shape of the population distribution.
对均值的置信区间依赖于样本均值 x̄ 的抽样分布。根据中心极限定理，对于足够大的样本量（常用指南为 n ≥ 30），无论总体分布形状如何，x̄ 的分布都近似服从正态分布，其均值为 μ，标准误差为 σ/√n。
If the population is itself normal, then x̄ is exactly normal even for small samples. This theoretical foundation allows us to use the normal model to construct intervals. The standard error σ/√n measures the variability of x̄ from sample to sample; a smaller standard error indicates more precise estimates.
如果总体本身就是正态分布，那么即使样本量很小，x̄ 也精确服从正态分布。这一理论基础使得我们能够使用正态模型来构建区间。标准误差 σ/√n 衡量的是不同样本间 x̄ 的变异性；标准误差越小，估计越精确。
4. Constructing a CI for μ When σ Is Known | 已知 σ 时构造均值置信区间
eviation σ is known, the formula for a confidence interval for the population mean μ is: x̄ ± z* × (σ/√n). To compute the interval, identify x̄, σ, n, and the desired confidence level, then find the associated z* from tables or software.
当总体标准差 σ 已知时，总体均值 μ 的置信区间公式为：x̄ ± z* × (σ/√n)。计算时，需要明确 x̄、σ、n 和所需的置信水平，然后从表格或软件中找出相应的 z*。
sample mean of 68 points. The population standard deviation is known to be 12 points. A 95% confidence interval for μ is: 68 ± 1.96 × (12/√36) = 68 ± 1.96 × 2 = 68 ± 3.92, giving an interval from 64.08 to 71.92.
示例：一位 IB 数学教师随机记录了 36 名学生的成绩，得到样本均值为 68 分。已知总体标准差为 12 分。μ 的 95% 置信区间为：68 ± 1.96 × (12/√36) = 68 ± 1.96 × 2 = 68 ± 3.92，即区间 (64.08, 71.92)。
We interpret this as: we are 95% between 64.08 and 71.92. The margin of error is 3.92. Changing the confidence level to 99% would expand the interval to 68 ± 2.576 × 2 = (62.85, 73.15).
对此我们可以解释为：我们有 95% 的把握认为所有学生的真实平均分介于 64.08 和 71.92 之间。误差范围为 3.92。若将置信水平改为 99%，则区间扩大为 68 ± 2.576 × 2 = (62.85, 73.15)。
5. Margin of Error and Its Calculation | 误差范围及其计算
nterval estimate. ME = critical value × standard error. In the known-σ case, ME = z* × (σ/√n). The margin of error is influenced by three factors: the confidence level (which determines z*), the population standard deviation σ, and the sample size n.
误差范围 (ME) 量化了区间估计的精确度。ME = 临界值 × 标准误差。在 σ 已知的情况下，ME = z* × (σ/√n)。误差范围受三个因素的影响：置信水平（决定 z*）、总体标准差 σ 和样本量 n。
Larger confidence levels increase ME because z* grows; larger σ increases ME because the data are more spread; larger n decreases ME because the denominator √n grows, making the standard error smaller. In practice, researchers can control n and the confidence level, but σ is a fixed population characteristic.
较高的置信水平会增加 ME，因为 z* 变大；较大的 σ 会增加 ME，因为数据更分散；较大的 n 会降低 ME，因为分母 √n 增大使得标准误差变小。在实际研究中，研究者可以控制 n 和置信水平，但 σ 是固定的总体特征。
The expression for ME also shows that to halve the margin of error, the sample size must be quadrupled, assuming σ and z* are held constant. This square-root relationship is important in planning sample sizes.
ME 的表达式还表明，若要保持 σ 和 z* 不变，要将误差范围减半，样本量必须增加为原来的四倍。这种平方根关系在设计样本量时十分重要。
6. Impact of Sample Size on Interval Width | 样本量对区间宽度的影响
Sample size n appears in the denominator of the standard error. As n increases, the standard error decreases, and the interval becomes narrower. This reflects the fact that larger samples provide more information about the population, yielding more precise estimates.
样本量 n 出现在标准误差的分母中。随着 n 增加，标准误差减小，区间变窄。这反映出更大的样本能提供更多关于总体的信息，从而得到更精确的估计。
linear; moving from n = 36 to n = 100 is a 2.78-fold increase in sample size but only a 40% reduction in ME.
仍以上例来说明，σ = 12，z* = 1.96。若 n = 36，ME = 3.92。若将样本量增加到 100，则 ME = 1.96 × (12/√100) = 1.96 × 1.2 = 2.35。区间宽度从 7.84 降至 4.70。这种关系并非线性：样本量从 36 增加到 100 是 2.78 倍的增长，而 ME 只减少了约 40%。
)². This ensures the final confidence interval has the desired precision for a given confidence level.
在设计探究时，可以通过试点研究估计 σ，然后由研究者确定期望的 ME，从而求出所需的样本量：n = (z*σ/ME)²。这能保证最终的置信区间在给定的置信水平下达到期望的精度。
7. The Role of Confidence Level: Trade-offs | 置信水平的作用：权衡
Choosing a higher confidence level increases the probability that the interval captures the parameter, but it also makes the interval wider. There is a trade-off between confidence and precision. A 99% CI is more reliable in terms of coverage but may be too wide to be useful for decision-making. A 90% CI is narrower but captures the parameter less frequently.
选择更高的置信水平会增加区间包含参数的概率，但同时也会使区间变宽。置信度与精确度之间存在权衡。99% 置信区间在覆盖率方面更可靠，但可能因过宽而难以用于决策。90% 置信区间较窄，但能包含参数的频率更低。
In an IB investigation, you might simulate repeated sampling from a known population to observe this trade-off. For example, generate 100 samples of size 30 from a normal distribution, construct both 90% and 99% CIs for each, and record how many intervals contain the true μ. Typically, around 90 of the 90% intervals and 99 of the 99% intervals will cover μ, but the average width of the 99% intervals will be considerably larger.
在 IB 探究中，你可以从已知总体中进行模拟重复抽样来观察这种权衡。例如，从正态分布中生成 100 个 n=30 的样本，分别构建 90% 和 99% 置信区间，并记录有多少个区间包含真实的 μ。通常大约 90% 的 90% 区间和 99% 的 99% 区间会覆盖 μ，但 99% 区间的平均宽度要大得多。
This visual demonstration reinforces the correct interpretation: the confidence level is a property of the method, not of a single interval. Researchers must balance the need for high certainty with the desire for a narrow, informative interval.
这一可视化演示强化了正确的解读：置信水平是方法本身的属性，而不是单个区间的属性。研究者必须在对高确定性的需求和对窄而富有信息量的区间的期望之间取得平衡。
8. Unknown σ: Introducing the t-Distribution | σ 未知：引入 t 分布
In most realistic situations, the population standard deviation σ is unknown. When σ is replaced by the sample standard deviation s, the statistic (x̄ - μ)/(s/√n) no longer follows a z-distribution; instead it follows a t-distribution with n - 1 degrees of freedom. The t-distribution is similar to the normal but has heavier tails to account for the extra uncertainty in estimating σ.
在大多数真实情境中，总体标准差 σ 是未知的。当用样本标准差 s 替代 σ 时，统计量 (x̄ - μ)/(s/√n) 不再服从 z 分布，而是服从自由度为 n - 1 的 t 分布。t 分布与正态分布相似，但尾部更厚，以体现估计 σ 时引入的额外不确定性。
The confidence interval formula becomes: x̄ ± t* × (s/√n ribution approaches the standard normal distribution, so for large samples the z-interval is often acceptable as an approximation.
此时置信区间的公式变为：x̄ ± t* × (s/√n)，其中 t* 是对应所需置信水平和自由度 df = n - 1 的 t 分布临界值。随着 n 增大，t 分布趋近于标准正态分布，因此对于大样本，z 区间通常可作为可接受的近似。
aths students yields a mean study time of 15.2 hours per week with s = 3.5 hours. For a 95% CI, df = 9, t* from table is 2.262. Interval: 15.2 ± 2.262 × (3.5/√10 e z-interval would be (using z* = 1.96 gives ME = 2.17), reflecting the extra uncertainty.
示例：从 IB 数学学生中随机抽取 10 人，平均每周学习时间为 15.2 小时，s = 3.5 小时。对于 95% 置信区间，df = 9，查表得 t* = 2.262。区间为 15.2 ± 2.262 × (3.5/√10) = 15.2 ± 2.262 × 1.107 ≈ 15.2 ± 2.50，即 (12.70, 17.70) 小时。t 区间比直接用 z 区间（z* = 1.96，ME = 2.17）要宽，体现了额外的不确定性。
9. Confidence Intervals for Proportions | 比例的置信区间
When estimating a population proportion p, the point estimate is the sample proportion p̂ = x/n, where x is the number of successes. Provided the sample is large enough (np̂ ≥ 10 and n(1 - p̂) ≥ 10 as a common check), the sampling distribution of p̂ is approximately normal with mean p and standard error √[p̂(1 - p̂)/n].
当估计总体比例 p 时，点估计为样本比例 p̂ = x/n，其中 x 是成功次数。只要样本足够大（通常检验条件为 np̂ ≥ 10 且 n(1 - p̂) ≥ 10），p̂ 的抽样分布近似正态，均值为 p，标准误差为 √[p̂(1 - p̂)/n]。
The confidence interval for p is: p̂ ± z* × √[p̂(1 - p̂ r example, in a survey of 200 IB students, 132 say they use a graphing calculator daily. p̂ = 132/200 = 0.66. A 95% CI: 0.66 ± 1.96 × √0.66 × 0.34 / 200 = 0.66 ± 1.96 × √0.2244/200 = 0.66 ± 1.96 × 0.0335 ≈ 0.66 ± 0.0657, i.e. (0.594, 0.726). We are 95% confident that between 59.4% and 72.6% of all IB students use a graphing calculator daily.
p 的置信区间为：p̂ ± z* × √[p̂(1 - p̂)/n]。例如，一项对 200 名 IB 学生的调查显示，有 132 人每天使用图形计算器，p̂ = 132/200 = 0.66。其 95% 置信区间为：0.66 ± 1.96 × √0.66 × 0.34 / 200 = 0.66 ± 1.96 × 0.0335 ≈ 0.66 ± 0.0657，即 (0.594, 0.726)。我们有 95% 的把握认为所有 IB 学生中使用图形计算器的比例介于 59.4% 到 72.6% 之间。
For smaller samples or extreme proportions, the normal approximation may be poor. In those cases, the Wilson score interval or exact binomial methods are more appropriate. IB investigations may compare these methods to see how coverage probabilities vary with n and p.
对于小样本或极端比例，正态近似可能效果不佳。此时更适用 Wilson 得分区间或精确二项方法。IB 探究活动可以比较这些方法，观察覆盖率如何随 n 和 p 变化。
10. Common Misinterpretations of Confidence Intervals | 置信区间的常见误解
Misconception 1: "A 95% confidence interval means there is a 95% probability that the true mean is inside the interval." Actually, once the interval is computed, the true mean either is or is not in it; the probability refers to the method's long-run performance.
误解一：“95% 置信区间意味着真实均值有 95% 的概率落在该区间内。” 事实上，一旦区间被计算出来，真实均值要么在、要么不在其中；概率指的是该方法在长期重复中的表现。
Misconception 2: "The confidence interval captures 95% of the sample data." This is false. The confidence interval is about the population parameter, not about individual observations. A prediction interval would be needed for future observations.
误解二：“置信区间包含了 95% 的样本数据。” 这是错误的。置信区间针对的是总体参数，而非个体观测值。要预测未来观测值，需要使用预测区间。
Misconception 3: "A wider interval is always better because it is more confident." While a wider interval offers higher confidence, it sacrifices precision. An extremely wide interval (e.g., "the mean height is between 100 cm and 250 cm") provides almost no useful information. The goal is to find an interval that is both reliable and narrow enough to guide decisions.
误解三：“区间越宽越好，因为置信度更高。” 虽然更宽的区间提供更高的置信度，但它牺牲了精确性。一个极宽的区间（如“平均身高在 100 厘米到 250 厘米之间”）几乎不提供任何有用信息。目标应是找到一个既可靠又足够窄的区间来指导决策。
In IB investigations, you can design a questionnaire or task to test these misconceptions among peers, then demonstrate the correct interpretations using simulations. This connects statistical theory with common reasoning errors.
在 IB 探究中，你可以设计一份问卷或任务来测试同伴对这些概念的理解误区，然后通过模拟演示正确的解读。这能将统计理论与常见推理错误联系起来。
11. Practical Investigation: Simulating Confidence Intervals | 实践探究：模拟置信区间
A powerful way to understand confidence intervals is to run a simulation. Using technology such as GeoGebra, Desmos, or a spreadsheet, you can generate many random samples from a known population, compute a confidence interval for each sample, and observe the coverage rate. For a 95% confidence level, about 95 out of 100 intervals should include the true parameter value.
理解置信区间的一种有效方法是运行模拟。使用 GeoGebra、Desmos 或电子表格等工具，你可以从一个已知总体中生成许多随机样本，为每个样本计算置信区间，并观察覆盖率。对于 95% 的置信水平，每 100 个区间中大约应有 95 个包含真实的参数值。
ration, you could set μ = 50 and σ = 10, draw 100 samples of size n = 25, and construct 95% re those that miss the parameter; count them. This graphical approach makes the concept of confidence level tangible.
在 IB 探究论文中，你可以设定 μ = 50 和 σ = 10，抽取 100 个 n = 25 的样本，并构建 95% z 区间。将这些区间画成水平线段，并在 μ = 50 处画一条竖直线。没有穿过竖线的区间就是漏掉了参数；统计其数量。这种图形化方法使置信水平的概念变得具体可见。
Extend the simulation by varying the confidence level (90%, 95%, 99% cord the observed coverage proportion and average interval width. This investigation allows you to verify the theoretical properties and discuss the trade-off between confidence and precision with real data.
可以通过改变置信水平（90%、95%、99%）或样本量（n = 10、30、100）来扩展模拟，并记录观察到的覆盖比例和平均区间宽度。这项探究活动能让你验证理论性质，并用真实数据讨论置信度与精确度之间的权衡。
| Confidence Level 置信水平 | z* | Expected Coverage 预期覆盖率 | Average Width (σ=10, n=30) 平均宽度 |
|---|---|---|---|
| 90% | 1.645 | 90% | 6.01 |
| 95% | 1.960 | 95% | 7.16 |
| 99% | 2.576 | 99% | 9.41 |
This table summarises how the critical value and interval width increase with confidence level. A simulation-based investigation will produce experimental values close to these theoretical figures, deepening understanding of the long-run frequency interpretation.
该表格总结了临界值和区间宽度如何随置信水平增加。基于模拟的探究将得出接近这些理论数值的实验结果，从而加深对长期频率解释的理解。
12. Conclusion: From Confidence Levels to Informed Decisions | 结论：从置信水平到明智决策
Confidence intervals provide a mathematically rigorous way to express the uncertainty in sample-based estimates. By choosing a confidence level and computing the corresponding interval, statisticians and data scientists communicate not just a single guess but a plausible range, along with the reliability of the method. In IB mathematics, investigations into confidence levels and intervals bridge theoretical probability, sampling distributions, and real-world data analysis.
置信区间提供了一种数学上严谨的方法来表达基于样本估计的不确定性。通过选定置信水平并计算相应的区间，统计学家和数据科学家不仅传达了一个单一的猜测值，还传达了一个合理的取值范围以及该方法的可靠性。在 IB 数学中，对置信水平与置信区间的探究架起了理论概率、抽样分布与现实数据分析之间的桥梁。
Key takeaways: the confidence level is not the probability that a specific interval contains the parameter; the width of an interval is controlled by the critical value, variability, and sample size; and the t-distribution must be used when σ is unknown. Explorations that integrate simulation allow students to experience these ideas dynamically, moving beyond mere algebraic manipulation.
核心要点：置信水平并非某个具体区间包含参数的概率；区间的宽度受临界值、变异性和样本量的控制；σ 未知时必须使用 t 分布。融合模拟的探究活动能让学生动态体验这些思想，超越单纯的代数运算。
Whether you are estimating the mean IB score, the proportion of left-handed students, or the effect of a new teaching strategy, mastering confidence intervals equips you with an essential tool for statistical reasoning. As you design your IB Internal Assessment, consider how you can apply these concepts to your own data, checking assumptions and interpreting results with appropriate caution and insight.
无论你是在估计 IB 平均分、左撇子学生的比例，还是一种新教学策略的效果，掌握置信区间都能为你提供统计推理的必备工具。在设计你的 IB 内部评估时，请思考如何将这些概念应用到自己的数据中，检查假设条件，并以适度的谨慎和洞察力解释结果。
Published by TutorHao | IB Maths Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图