📚 Unbiased Estimators and Confidence Intervals | 无偏估计量与置信区间
In statistical inference, we often need to estimate unknown population parameters using sample data. An unbiased estimator is a statistic whose expected value equals the true parameter, ensuring that on average we are accurate. Confidence intervals then give a range of plausible values for the parameter, reflecting the uncertainty in our estimate. This article explores the theory behind unbiased estimators and the construction of confidence intervals, which are fundamental in A-level and IB Mathematics.
在统计推断中，我们常常需要用样本数据去估计未知的总体参数。无偏估计量是其期望值等于真实参数的统计量，能保证长期来看估计是准确的。置信区间则提供了一个参数可能取值的范围，反映了估计的不确定性。本文探讨无偏估计量的原理以及置信区间的构造方法，这些都是 A-level 和 IB 数学中的基础内容。
1. Populations, Samples and Statistics | 总体、样本和统计量
A population is the entire set of items or individuals of interest, characterised by parameters such as the population mean μ and population variance σ². A sample is a subset drawn from the population, used to calculate statistics like the sample mean x̄ and sample variance s², which serve as estimators of the corresponding parameters.
总体是我们感兴趣的全部个体或项目的集合，其特征由参数描述，比如总体均值 μ 和总体方差 σ²。样本是从总体中抽取的子集，用于计算统计量（例如样本均值 x̄ 和样本方差 s²），这些统计量作为对应参数的估计量。
The sampling distribution of a statistic describes how the statistic varies from sample to sample. Understanding its properties — particularly its mean and variance — is essential for assessing whether an estimator is unbiased and for constructing confidence intervals.
统计量的抽样分布描述了该统计量在不同样本之间的变异情况。理解其性质——特别是均值和方差——对于判断估计量是否无偏以及构造置信区间至关重要。
2. Point Estimators | 点估计量
A point estimator is a single value calculated from sample data that provides a “best guess” of an unknown population parameter. For example, the sample mean x̄ is a natural point estimator for the population mean μ. However, a good estimator should possess certain desirable properties: unbiasedness, efficiency, and consistency.
点估计量是从样本数据计算出的一个单一数值，用于提供一个关于未知总体参数的“最佳猜测”。例如，样本均值 x̄ 是总体均值 μ 的自然点估计量。然而，一个好的估计量应具备一些理想性质：无偏性、有效性和一致性。
An estimator is called unbiased if its expected value equals the true parameter value. It is efficient if it has a relatively small variance compared to other unbiased estimators. It is consistent if it converges in probability to the parameter as the sample size increases.
如果估计量的期望值等于真实的参数值，则称其为无偏估计量。如果它相对于其他无偏估计量具有较小的方差，则称其有效。如果当样本量增大时它依概率收敛于参数，则称其一致。
3. Defining Unbiasedness | 无偏性的定义
Formally, an estimator θ̂ of a parameter θ is unbiased if E(θ̂ n is taken over all possible samples of a given size. If the estimator is biased, then E(θ̂) ≠ θ, and the bias is defined as Bias(θ̂) = E(θ̂) − θ.
正式地说，参数 θ 的估计量 θ̂ 是无偏的，如果 E(θ̂) = θ。这里的期望是针对于所有给定容量的可能样本。如果估计量有偏，则 E(θ̂) ≠ θ，偏差定义为 Bias(θ̂) = E(θ̂) − θ。
Unbiasedness means that the estimator does not systematically overestimate or underestimate the parameter. It is a desirable property because it ensures the centre of the sampling distribution is at the true value.
无偏性意味着估计量不会系统性地高估或低估参数。这是一个理想的性质，因为它保证了抽样分布的中心就是真实值。
4. Unbiased Estimators of μ and σ² | 总体均值与方差的无偏估计量
The sample mean x̄ = (1/n)∑xi is an unbiased estimator of the population mean μ, because E(x̄ the population distribution, provided the sample is random and the population mean exists.
样本均值 x̄ = (1/n)∑xi 是总体均值 μ 的无偏估计量，因为 E(x̄) = μ。无论总体服从什么分布，只要样本是随机的且总体均值存在，这一点都成立。
For the population variance σ², the sample variance s² = (1/(n−1))∑(xi − x̄)² is an unbiased estimator. The denominator n−1 (known as Bessel’s correction) compensates for the fact that the sample mean is used, which introduces a slight negative bias into the sum of squared deviations. In contrast, using n in the denominator gives a biased estimator with expected value ((n−1)/n)σ².
对于总体方差 σ²，样本方差 s² = (1/(n−1))∑(xi − x̄)² 是无偏估计量。分母使用 n−1（称为贝塞尔校正）弥补了因使用样本均值而造成的偏差，后者使得离差平方和略微偏小。相反，分母用 n 会得到一个有偏估计量，其期望值为 ((n−1)/n)σ²。
5. Sampling Distribution of the Sample Mean | 样本均值的抽样分布
If a random sample of size n is drawn from a population with mean μ and variance σ², the sampling distribution of x̄ has mean μ and variance σ²/n. Therefore, the standard error of the mean is σ/√n.
若从一个均值为 μ、方差为 σ² 的总体中抽取容量为 n 的随机样本，则 x̄ 的抽样分布的均值为 μ，方差为 σ²/n。因此，均值的标准误差为 σ/√n。
When the population is normally distributed, x̄ is also normally distributed: x̄ ~ tral Limit Theorem (CLT) states that for large n (typically n ≥ 30), the distribution of x̄ is approximately normal. This is the foundation of confidence intervals when σ is known.
当总体服从正态分布时，x̄ 也服从正态分布：x̄ ~ N(μ, σ²/n)。即使总体不服从正态分布，根据中心极限定理（CLT）当 n 较大（通常 n ≥ 30）时，x̄ 的分布也近似正态。这是在 σ 已知时构造置信区间的基础。
6. Confidence Interval for μ with Known σ | σ已知时总体均值的置信区间
A 95% confidence interval for μ is an interval calculated from sample data that, in repeated sampling, would contain the true μ in 95% of such samples. It is not a probability statement about μ itself (which is fixed), but about the method of interval construction.
总体均值 μ 的 95% 置信区间是从样本数据计算出的一个区间，在重复抽样下，这一区间会包含真实的 μ 的概率为 95%。这并不是关于 μ 本身的概率陈述（μ 是固定的），而是关于区间构造方法可靠性的说明。
When σ is known and the population is normal (or n large), the interval takes the form:
x̄ ± z* × (σ/√n)
where z* is the critical value from the standard normal distribution. For a 95% confidence level, z* = 1.96.
当 σ 已知且总体服从正态分布（或 n 较大）时，置信区间的形式为：
x̄ ± z* × (σ/√n)
其中 z* 是标准正态分布的临界值。对于 95% 置信水平，z* = 1.96。
The margin of error is z* × (σ/√n). Increasing the sample size n reduces the margin of error, making the interval narrower.
误差幅度为 z* × (σ/√n)。增加样本容量 n 可以减小误差幅度，使区间变窄。
7. Confidence Interval for μ with Unknown σ | σ未知时总体均值的置信区间
In practice, σ is usually unknown and must be estimated by the sample standard deviation s. This introduces extra uncertainty, so we use the t-distribution instead of the normal. The confidence interval is:
x̄ ± t* × (s/√n)
where t* is the critical value from the t-distribution with (n−1) degrees of freedom.
在实践中，σ 通常是未知的，必须用样本标准差 s 来估计。这引入了额外的不确定性，因此我们使用 t 分布来代替正态分布。置信区间为：
x̄ ± t* × (s/√n)
其中 t* 是自由度为 (n−1) 的 t 分布的临界值。
The t-distribution has heavier tails than the normal distribution, resulting in wider confidence intervals for small n. As n increases, the t-distribution approaches the standard normal, and the z* and t* values converge.
t 分布比正态分布具有更厚的尾部，因此在样本容量较小时会给出更宽的置信区间。随着 n 增大，t 分布趋近于标准正态分布，z* 和 t* 的数值也会趋同。
8. Confidence Interval for a Population Proportion | 总体比例的置信区间
When estimating a population proportion p, the point estimator is the sample proportion p̂ the number of successes in n independent Bernoulli trials. The sampling distribution of p̂ has mean p and variance p(1−p)/n. For sufficiently large n (such that np̂ ≥ 10 and n(1−p̂) ≥ 10), p̂ is approximately normal.
在估计总体比例 p 时，点估计量是样本比例 p̂ = x/n，其中 x 是 n 次独立伯努利试验中的成功次数。p̂ 的抽样分布的均值为 p，方差为 p(1−p)/n。对于足够大的 n（满足 np̂ ≥ 10 且 n(1−p̂) ≥ 10），p̂ 近似服从正态分布。
The approximate confidence interval for p is:
p̂ ± z* × √(p̂(1−p̂)/n)
where z* is the appropriate normal critical value.
p 的近似置信区间为：
p̂ ± z* × √(p̂(1−p̂)/n)
其中 z* 为相应的正态临界值。
This interval is built on the assumption that p̂ is roughly normal and that the standard error is estimated by √(p̂(1−p̂)/n). For smaller samples, an exact binomial method (Clopper–Pearson interval) or Wilson score interval may be preferred.
这一区间建立在 p̂ 近似正态且标准误差由 √(p̂(1−p̂)/n) 估计的基础上。对于较小的样本，更推荐使用精确二项方法（Clopper–Pearson 区间）或 Wilson 得分区间。
9. Interpreting Confidence Intervals | 置信区间的解释
A common misinterpretation is that a 95% confidence interval has a 95% probability of containing the true parameter. The correct interpretation is: “If we were to take many random samples and compute a 95% confidence interval from each, about 95% of those intervals would contain the true parameter.” For a single computed interval, it either contains the parameter or it does not – no probability attaches to that single outcome.
一个常见的误解是认为 95% 置信区间有 95% 的概率包含真实参数。正确的解释是：“如果我们多次随机抽样并计算 95% 置信区间，那么大约 95% 的区间会包含真实参数。”对于某个具体计算出的区间，它要么包含参数，要么不包含——这一特定结果并不附带概率。
Confidence level refers to the long‑run success rate of the method, not to a probabilistic statement about the particular interval. This distinction is crucial in hypothesis testing and statistical literacy.
置信水平指的是该方法的长期成功率，而不是对某个特定区间的概率陈述。这一区别在假设检验和统计素养中至关重要。
10. Margin of Error and Sample Size Determination | 误差幅度与样本容量的确定
The margin of error (ME) for a mean is z* × σ/√n (or t* × s/√n). To achieve a desired ME, we can solve for n:
n = (z* × σ / ME)²
σ is known or estimated from a pilot study. For proportions, the formula is n = (z*)² × p̂(1−p̂) / ME². If no prior p̂ is available, using p̂ = 0.5 gives the most conservative (largest) sample size.
均值的误差幅度（ME）为 z* × σ/√n（或 t* × s/√n）。为达到期望的误差幅度，我们可以求解 n：
n = (z* × σ / ME)²
当 σ 已知或由先导研究估计得出。对于比例，公式为 n = (z*)² × p̂(1−p̂) / ME²。如果没有先验的 p̂，使用 p̂ = 0.5 可以得到最保守（最大）的样本量。
11. Relationship Between Unbiased Estimators and Confidence Intervals | 无偏估计量与置信区间的关系
Unbiased estimators provide the central point of confidence intervals. For the population mean, the unbiased estimator x̄ is exactly at the centre of the symmetric interval. The width of the interval reflects the precision of the estimator: the smaller the variance of the estimator, the narrower the interval for a given confidence level.
无偏估计量提供了置信区间的中心点。对于总体均值，无偏估计量 x̄ 恰好位于对称区间的中心。区间的宽度反映了估计量的精确度：估计量的方差越小，在给定置信水平下区间就越窄。
Using a biased estimator would shift the centre away from the true parameter, leading to systematically misleading intervals. Thus, unbiasedness is a property that supports the validity of confidence intervals, even though the interval width also depends on other factors.
使用有偏估计量会使中心点偏离真实参数，导致系统性的误导区间。因此，无偏性是支持置信区间有效性的一个性质，尽管区间宽度还取决于其他因素。
12. Summary and Common Pitfalls | 总结与常见易错点
- Unbiased estimator: E(θ̂) = θ. Sample mean is unbiased for μ; sample variance with (n−1) is unbiased for σ².
- 无偏估计量： E(θ̂) = θ。样本均值是对 μ 的无偏估计量；分母为 (n−1) 的样本方差是对 σ² 的无偏估计量。
- Confidence interval: an interval estimate that depends on the sample and is constructed so that it captures the parameter with a specified long‑run frequency.
- 置信区间： 一种依赖于样本的区间估计，并以指定的长期频率捕获参数。
- Common mistake: treating the confidence level as the probability that the parameter lies in the specific interval. Remember: the parameter is fixed, and the interval is random before data collection.
- 常见错误： 将置信水平视为参数落在特定区间内的概率。请记住：参数是固定的，区间在数据收集前是随机的。
- Sample size effect: rrower confidence interval. Bias does not decrease with n if the estimator is inherently biased.
- 样本量的影响： n 增大 → 标准误差减小 → 置信区间变窄。如果估计量本身有偏，偏差不会随 n 增大而消失。
Mastering the concepts of unbiasedness and confidence intervals equips students with essential tools for statistical inference, enabling them to critically evaluate data-based claims in science, economics, and everyday life.
掌握无偏性和置信区间的概念，为学生提供了统计推断的基本工具，使他们能够批判性地评估科学、经济学和日常生活中基于数据的论断。
Published by TutorHao | Mathematics 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