What Makes a Quantity Pivotal?
A sampling distribution describes how a quantity calculated from data varies across repeated samples. A pivotal quantity is useful because its distribution is the same for every allowed value of the unknown parameter. Its formula can contain the unknown parameter; what must not depend on that parameter is the resulting distribution. This distinction is central: the pivot is generally not directly observable as a number, but its known distribution lets us reason about it.
For example, if \(X_1,\ldots,X_n\) are independent observations from a normal distribution with unknown mean \(\mu\) and known standard deviation \(\sigma\), then \((\bar X-\mu)/(\sigma/\sqrt n)\) is a pivot. It contains \(\mu\), but for every possible \(\mu\) its distribution is standard normal. By contrast, \(\bar X\) is not a pivot in this model: its distribution is \(N(\mu,\sigma^2/n)\), which depends on the unknown \(\mu\).
A pivot is different from a statistic in the strict sense often used in estimation. A statistic is a function of the data alone and cannot contain an unknown parameter. A pivot may contain one. Once data have been observed, the pivot's expression still includes the unknown parameter, so its observed numerical value is not generally known. The parameter-free sampling distribution, however, is known. Pivots are therefore a bridge between sampling theory and interval estimation; here the focus is on their definition and distributions, not on converting them into intervals.
Standard Normal-Sample Pivots
Assume \(X_1,\ldots,X_n\) are independent and identically distributed as \(N(\mu,\sigma^2)\). The sample mean is \(\bar X=(1/n)\sum_{i=1}^n X_i\), and the sample variance is \(S^2=(1/(n-1))\sum_{i=1}^n(X_i-\bar X)^2\). These definitions use \(n-1\) in the sample variance.
| Sampling model and assumptions | Pivot | Distribution |
|---|---|---|
| Normal sample; \(\sigma\) known | \((\bar X-\mu)/(\sigma/\sqrt n)\) | \(N(0,1)\) |
| Normal sample; \(\mu,\sigma\) unknown | \((\bar X-\mu)/(S/\sqrt n)\) | \(t_{n-1}\) |
| Normal sample; \(\mu,\sigma\) unknown | \((n-1)S^2/\sigma^2\) | \(\chi^2_{n-1}\) |
| Independent exponential sample with rate \(\lambda\) | \(2\lambda\sum_{i=1}^n X_i\) | \(\chi^2_{2n}\) |
The notation \(t_{n-1}\) denotes a Student \(t\) distribution with \(n-1\) degrees of freedom, and \(\chi^2_k\) denotes a chi-square distribution with \(k\) degrees of freedom. The normal pivots require a normal population, not merely a large sample. The exponential pivot below requires independent observations with the same rate \(\lambda\), where the exponential density is \(f_X(x)=\lambda e^{-\lambda x}\) for \(x>0\). Changing these assumptions can change the sampling distributions, so the formulas should not be applied automatically to nonnormal or dependent samples.
Why the Normal Pivots Have These Distributions
The known-standard-deviation pivot follows by standardizing the sample mean. Under the normal-sample model, \(\bar X\) has distribution \(N(\mu,\sigma^2/n)\). Subtracting its mean and dividing by its standard deviation gives a standard normal variable:
The distribution on the right contains neither \(\mu\) nor \(\sigma\). If \(\sigma\) is known, this is a pivot even though \(\mu\) is unknown.
When \(\sigma\) is unknown, the sample standard deviation replaces it, but the resulting quantity has a \(t\) distribution rather than a standard normal distribution. The reason is the exact normal-sample result that \(\bar X\) and \(S^2\) are independent, with \(\sqrt n(\bar X-\mu)/\sigma\sim N(0,1)\) and \((n-1)S^2/\sigma^2\sim\chi^2_{n-1}\). By definition, the ratio of an independent standard normal variable to the square root of an independent chi-square variable divided by its degrees of freedom has a Student \(t\) distribution. Substituting these two normal-sample quantities gives:
The unknown scale \(\sigma\) cancels in the ratio. That cancellation, together with the independence result, explains why the \(t\) distribution does not depend on either unknown normal parameter. Normality is essential to this exact argument.
The variance pivot follows directly from the same normal-sample result: \((n-1)S^2/\sigma^2\sim\chi^2_{n-1}\). Its distribution is parameter-free even though its formula contains the unknown variance \(\sigma^2\). This pivot concerns the sample variance; it does not require knowing the population mean.
Worked Examples: Identifying and Checking Pivots
Worked Example: A Normal Mean With Known Standard Deviation
A hypothetical calibration process produces independent measurements from \(N(\mu,12^2)\). The population mean \(\mu\) is unknown, while the standard deviation of 12 units is known. A sample of \(n=16\) readings has mean \(\bar x=104.5\). Identify the pivot and find its value if a proposed value of the mean is \(\mu=100\).
Method. For a normal sample with known \(\sigma\), standardize the sample mean using \((\bar X-\mu)/(\sigma/\sqrt n)\). This is a pivot with distribution \(N(0,1)\).
Compute the standard error. The known standard deviation is \(\sigma=12\), so \(\sigma/\sqrt n=12/\sqrt{16}=12/4=3\).
Write the pivot for these data. Before substituting a proposed mean, its observed-data expression is \((104.5-\mu)/3\). Since the true \(\mu\) is unknown, this expression is not yet a known numerical value.
Evaluate at the proposed value. If \(\mu=100\), the pivot value is \((104.5-100)/3=4.5/3=1.5\). Under the model, the pivot has a standard normal distribution for every \(\mu\), so \(P(Q\leq1.5)=0.9332\) to four decimal places.
Check and common mistake. The standard error is 3, not 12: the standard deviation of a sample mean is \(\sigma/\sqrt n\). Dividing 4.5 by 12 would give 0.375, the wrong standardized value. The calculation also confirms that the pivot's distribution is \(N(0,1)\), while its realized value requires specifying a value of \(\mu\).
Worked Example: A Normal Mean With Unknown Standard Deviation
A hypothetical laboratory takes \(n=5\) independent observations from a normal population with both \(\mu\) and \(\sigma\) unknown. The observed measurements are 2, 4, 6, 8, and 10. Identify the appropriate pivot, calculate its expression from the data, and evaluate it at a proposed mean \(\mu=4\).
Method. With both normal parameters unknown, use the Student \(t\) pivot \((\bar X-\mu)/(S/\sqrt n)\), which has \(t_{n-1}\) distribution.
Calculate the sample mean. The sum is \(2+4+6+8+10=30\), so \(\bar x=30/5=6\).
Calculate the sample variance. The deviations from 6 are \(-4,-2,0,2,4\). Their squared sum is \(16+4+0+4+16=40\). Therefore, \(s^2=40/(5-1)=10\), and \(s=\sqrt{10}\).
Calculate the standard error. \(s/\sqrt n=\sqrt{10}/\sqrt5=\sqrt2\).
Write and evaluate the pivot. The observed-data expression is \((6-\mu)/\sqrt2\), with a \(t_4\) distribution. At the proposed value \(\mu=4\), it equals \(2/\sqrt2=\sqrt2\), approximately 1.4142.
Check and common mistake. The denominator uses the sample standard deviation divided by \(\sqrt5\), not \(s\) alone. The sample variance also divides by \(n-1=4\), not by 5. Using either incorrect denominator would not give the stated \(t_4\) pivot.
Worked Example: An Exponential-Rate Pivot
A hypothetical component's operating time \(X\), in years, is modeled by an exponential distribution with unknown rate \(\lambda\) per year. Four independent component lifetimes are \(0.2, 0.4, 0.6,\) and \(0.8\) years. Identify the pivot, calculate its observed-data expression, and evaluate it if a proposed rate is \(\lambda=1\) per year.
Method. For independent exponential observations with common rate \(\lambda\), their sum has a gamma distribution with shape \(n\) and rate \(\lambda\). Equivalently, \(2\lambda\sum X_i\) has a chi-square distribution with \(2n\) degrees of freedom.
Calculate the sample sum. \(0.2+0.4+0.6+0.8=2.0\) years.
Form the pivot. Here \(n=4\), so \(Q=2\lambda\sum X_i=2\lambda(2.0)=4\lambda\), where \(\lambda\) is measured per year. Its distribution is \(\chi^2_8\), independent of the value of \(\lambda\).
Evaluate at the proposed rate. At \(\lambda=1\) per year, \(Q=4(1)=4\). The units cancel because the rate is per year and the sample sum is in years.
Check and common mistake. The chi-square degrees of freedom are \(2n=8\), not \(n=4\). Omitting the factor of 2 would produce \(\lambda\sum X_i\), which has a gamma distribution in a different parameterization, not the stated chi-square pivot.
Verifying a Candidate and Avoiding Traps
To decide whether a proposed expression is pivotal, write down its sampling distribution under the stated model and check whether any unknown parameter remains in that distribution. Do not decide from the formula alone. An unknown parameter may appear in the expression, as in \((\bar X-\mu)/(\sigma/\sqrt n)\), yet disappear from its distribution. Conversely, a data-only statistic may have a distribution that depends on an unknown parameter, so being a statistic does not make it a pivot.
For example, if \(X\sim\operatorname{Binomial}(n,p)\) with unknown \(p\), then \(X\) is a statistic, but its probability mass function \(\binom{n}{x}p^x(1-p)^{n-x}\) depends on \(p\); thus \(X\) is not pivotal. Standardizing it as \((X-np)/\sqrt{np(1-p)}\) does not generally make an exact pivot: the distribution remains discrete and depends on \(p\). Approximate parameter-free behavior in a large sample is a separate issue from an exact pivot.
Record the distributional assumptions, which parameters are unknown, and which quantities are known.
Make clear which parts depend on the data and which include unknown parameters.
Use the model's sampling results and check whether the final distribution still involves any unknown parameter.
Confirm that the formula matches the model and parameterization before labeling it a pivot.
A frequent exam error is to treat every standardized statistic as pivotal. Standardization alone is not sufficient: the resulting distribution must be parameter-free under the exact stated assumptions. Another error is to confuse the distribution of a pivot with the numerical value obtained after observing data. For a pivot containing \(\mu\) or \(\sigma\), that value still changes with the unknown parameter, even though the pivot's sampling distribution is fixed.
Check Your Understanding
For each setting, identify the proposed pivot or explain why the proposed quantity is not pivotal. Answers include the key distribution or calculation.
- Independent observations come from \(N(\mu,5^2)\), with known standard deviation 5 and unknown \(\mu\). For \(n=25\), identify the standardized sample-mean pivot and its distribution. (Answer: \((\bar X-\mu)/(5/5)=\bar X-\mu\sim N(0,1)\).)
- An independent sample of size 7 is normal, with both parameters unknown. State the pivot for the mean and its degrees of freedom. (Answer: \((\bar X-\mu)/(S/\sqrt7)\sim t_6\).)
- Four independent lifetimes follow an exponential distribution with unknown rate \(\lambda\). Their observed sum is 5 years. Form the rate pivot, give its distribution, and evaluate it at \(\lambda=0.5\) per year. (Answer: \(2\lambda\sum X_i\sim\chi^2_8\); at the proposed rate its value is \(2(0.5)(5)=5\).)
- A count \(X\) has a binomial distribution with \(n=12\) and unknown success probability \(p\). Is \(X\) a pivot? State why. (Answer: No; its binomial distribution depends on \(p\).)
- A normal sample has \(n=16\), unknown mean and variance, and observed sample variance \(S^2=20\). Identify the variance pivot, its distribution, and its expression using the observed variance. (Answer: \(15S^2/\sigma^2\sim\chi^2_{15}\); with the observed variance its expression is \(300/\sigma^2\), which still has the parameter-free \(\chi^2_{15}\) sampling distribution.)