Deriving The Student t Pivot For A Normal Sample

Why the Student t Pivot Arises

When a normal population has known standard deviation, the sample mean can be standardized using that known quantity. In practice, the population standard deviation is often unknown, so the sample standard deviation is used instead. That substitution makes the standardized quantity random in both its numerator and denominator. Under an independent normal sampling model, the resulting ratio has an exact Student t distribution.

The purpose here is to derive that distributional result. The central point is not merely that the sample size determines the degrees of freedom: it is that the sample mean and sample variance have particular distributions and are independent when the observations are independent and normal. This derivation is aligned with the SOA VEE Mathematical Statistics syllabus.

Definition: Let $X_1,\ldots,X_n$ be independent observations from $N(\mu,\sigma^2)$, where $\mu$ is any real number, $\sigma>0$, and $n\geq 2$. Define the sample mean and unbiased sample variance by $$ \bar X=\frac{1}{n}\sum_{i=1}^n X_i,\qquad S^2=\frac{1}{n-1}\sum_{i=1}^n (X_i-\bar X)^2. $$ The Student t pivot for $\mu$ is $$ T=\frac{\bar X-\mu}{S/\sqrt n} =\frac{\sqrt n(\bar X-\mu)}{S}. $$ Under the stated model, $T$ has a Student t distribution with $n-1$ degrees of freedom.

The pivot is a function of the data and the parameter $\mu$, but its distribution does not depend on either unknown population parameter $\mu$ or $\sigma$. That parameter-free distribution is what makes the statistic useful. The divisor $n-1$ in $S^2$ is essential to the result.

The Two Distributional Ingredients

First consider the standardized sample mean. Since independent normal variables have a normal sum, $\bar X$ is normal with mean $\mu$ and variance $\sigma^2/n$. Therefore $$ Z=\frac{\sqrt n(\bar X-\mu)}{\sigma}\sim N(0,1). $$ The other ingredient is the scaled sample variance: $$ Q=\frac{(n-1)S^2}{\sigma^2} =\frac{\sum_{i=1}^n(X_i-\bar X)^2}{\sigma^2}. $$ For an independent normal sample, $Q$ has a chi-square distribution with $n-1$ degrees of freedom. Crucially, $Z$ and $Q$ are independent.

The reason for the degrees of freedom and independence can be seen by expressing the standardized observations as coordinates in a rotated coordinate system. Write $Y_i=(X_i-\mu)/\sigma$. The vector $(Y_1,\ldots,Y_n)$ has independent standard normal coordinates. Its density is proportional to $\exp(-\sum_iY_i^2/2)$, so an orthogonal change of coordinates preserves the density and leaves the transformed coordinates independent standard normals.

Choose the sample-mean direction.
Take the first axis to point along $(1,\ldots,1)/\sqrt n$. The coordinate on this axis is $Z=\sqrt n(\bar X-\mu)/\sigma$, so it is standard normal.

Use the remaining directions for deviations.
The other $n-1$ orthogonal coordinates describe deviations from the sample mean. Their squared values sum to $\sum_i(X_i-\bar X)^2/\sigma^2=Q$.

Read off the distribution and independence.
The $n-1$ remaining coordinates are independent standard normals, independent of the first coordinate. Their squared sum is therefore chi-square with $n-1$ degrees of freedom, and $Q$ is independent of $Z$.

This also explains why the variance has $n-1$ degrees of freedom: the deviations sum to zero, so only $n-1$ orthogonal deviation coordinates remain after the sample-mean direction is accounted for.

Combining the Ingredients

A Student t random variable with $\nu$ degrees of freedom is defined as the ratio of a standard normal random variable to the square root of an independent chi-square random variable divided by its degrees of freedom. Thus, if $Z\sim N(0,1)$, $Q\sim\chi^2_\nu$, and $Z$ and $Q$ are independent, then $$ \frac{Z}{\sqrt{Q/\nu}}\sim t_\nu. $$ For the normal sample, use $\nu=n-1$. Since $Q=(n-1)S^2/\sigma^2$, it follows that $$ \frac{Z}{\sqrt{Q/(n-1)}} = \frac{\sqrt n(\bar X-\mu)/\sigma} {\sqrt{\left((n-1)S^2/\sigma^2\right)/(n-1)}} = \frac{\sqrt n(\bar X-\mu)/\sigma}{S/\sigma} = \frac{\sqrt n(\bar X-\mu)}{S}. $$ The final expression is exactly $T$, giving the required result.

Theorem: If $X_1,\ldots,X_n$ are independent and identically distributed as $N(\mu,\sigma^2)$, with $\sigma>0$ and $n\geq2$, then $$ \frac{\bar X-\mu}{S/\sqrt n}\sim t_{n-1}. $$ The exact result requires both normality and independence. If the observations are dependent or the population is not normal, this derivation does not establish an exact Student t distribution.

Worked Examples

Worked Example: Computing the Pivot From a Small Normal Sample

A hypothetical laboratory measures a chemical concentration in four independent specimens. Assume the measurements are a random sample from $N(\mu,\sigma^2)$, with hypothesized population mean $\mu=9$ units and population standard deviation $\sigma=2$ units. The observations are 8, 10, 12, and 14 units. Calculate the pivot and identify its reference degrees of freedom.

Method and assumptions. Use the one-sample Student t pivot. The stated model is independent normal sampling, $n=4$, $\mu=9$, and $\sigma=2$. The value of $\sigma$ is useful for checking the normal and chi-square components, although it cancels from the pivot.

The sample mean is $$ \bar X=\frac{8+10+12+14}{4}=\frac{44}{4}=11. $$ The deviations from 11 are $-3,-1,1,3$, and their squared sum is $$ (-3)^2+(-1)^2+1^2+3^2=9+1+1+9=20. $$ Therefore $$ S^2=\frac{20}{4-1}=\frac{20}{3},\qquad S=\sqrt{\frac{20}{3}}. $$ The pivot is $$ T=\frac{\sqrt4(11-9)}{\sqrt{20/3}} =\frac{4}{\sqrt{20/3}} =\sqrt{\frac{16}{20/3}} =\sqrt{\frac{12}{5}} \approx 1.5492. $$ Its degrees of freedom are $n-1=3$.

Check using the two components. Here $Z=\sqrt4(11-9)/2=2$, while $Q=3(20/3)/2^2=5$. Hence $Z/\sqrt{Q/3}=2/\sqrt{5/3}=\sqrt{12/5}$, agreeing with the direct calculation. The common mistake avoided is dividing the sum of squared deviations by $n$: $S^2$ uses $n-1$, not $n$.

Worked Example: Verifying the Normal-to-t Ratio

A hypothetical independent sample of five instrument readings is 3, 5, 5, 7, and 10 units. Assume $X_i\sim N(5,2^2)$, so $\mu=5$ and $\sigma=2$. Calculate the sample pivot by both its direct formula and its normal-over-chi-square representation.

Method and assumptions. Apply the result for an independent normal sample, with $n=5$. The observations have sum $3+5+5+7+10=30$, so $$ \bar X=\frac{30}{5}=6. $$ The deviations from 6 are $-3,-1,-1,1,4$, whose squared sum is $$ 9+1+1+1+16=28. $$ Thus $S^2=28/(5-1)=7$ and $S=\sqrt7$. The direct pivot is $$ T=\frac{\sqrt5(6-5)}{\sqrt7} =\sqrt{\frac{5}{7}} \approx 0.8452. $$ It has $5-1=4$ degrees of freedom.

For the component check, $$ Z=\frac{\sqrt5(6-5)}{2}=\frac{\sqrt5}{2}, \qquad Q=\frac{4(7)}{2^2}=7. $$ Therefore $$ \frac{Z}{\sqrt{Q/4}} =\frac{\sqrt5/2}{\sqrt{7/4}} =\frac{\sqrt5/2}{\sqrt7/2} =\sqrt{\frac57} \approx 0.8452. $$ This matches the direct pivot. The common mistake avoided is to leave a stray factor of $2$ in the ratio: the numerator and denominator both contain a factor of $1/2$, which cancels.

Worked Example: Detecting a Divisor-n Error

For a hypothetical independent normal sample of size $n=3$, suppose the sample summaries are $\bar X=12$ and $S^2=8$. Under the model $X_i\sim N(10,\sigma^2)$, with $\mu=10$ and $\sigma$ unknown, calculate the pivot. Then find the incorrect value produced by using the variance calculated with divisor $n$ to form the standard error.

Method and assumptions. Use the sample-variance definition with divisor $n-1$, and then compare it with the deliberately incorrect divisor-$n$ calculation. The supplied summaries are treated as the observed summaries of the stated independent normal sample.

The correct pivot is $$ T=\frac{\sqrt3(12-10)}{\sqrt8} =\frac{2\sqrt3}{2\sqrt2} =\sqrt{\frac32} \approx 1.2247. $$ The correct reference degrees of freedom are $3-1=2$.

Because $S^2=8$ uses divisor $n-1=2$, the sum of squared deviations is $2(8)=16$. The variance computed incorrectly using divisor $n=3$ would be $16/3$. If that is then used in the standard error, the incorrect standard error is $$ \sqrt{\frac{16/3}{3}}=\sqrt{\frac{16}{9}}=\frac43. $$ The resulting incorrect pivot is $$ \frac{12-10}{4/3}=\frac{2}{4/3}=\frac32=1.5000. $$ The correct value is $1.2247$, not $1.5000$. The common mistake is replacing $S^2$, defined with divisor $n-1$, by a variance with divisor $n$ while still treating it as the sample variance for the t pivot.

Assumptions and Exam Traps

The exact t result is a statement about the sampling distribution before observing a particular sample. It does not say that every observed ratio has a t-shaped distribution, nor does it require the observed value of $T$ to be near zero. It says that repeated samples under the stated model produce values of $T$ according to $t_{n-1}$.

Normality matters for exactness. The independence between the standardized sample mean and sample variance, and the chi-square result for the scaled variance, follow from normal sampling. Without normality, the exact derivation is not justified.
Independence matters. Correlated observations do not generally have the joint normal density and orthogonal-coordinate properties used in the derivation.
Use the correct variance convention. The sample variance in the pivot is $S^2=\sum(X_i-\bar X)^2/(n-1)$. The sample standard deviation is $S=\sqrt{S^2}$, not $S^2$.
Count degrees of freedom from the sample size. There are $n-1$ deviation coordinates after estimating the sample mean. The pivot therefore uses $t_{n-1}$, not $t_n$.
The case $n=1$ is excluded. With one observation there is no residual variation estimate using divisor $n-1$, so this pivot is undefined.

This derivation is the foundation for exact one-mean t procedures: once the pivot is known, probability statements about it can be rearranged to make statements about $\mu$. The key result here is the pivot and its distribution; using it to construct a particular interval is a separate step.

Key takeaway: For an independent normal sample, $Z=\sqrt n(\bar X-\mu)/\sigma$ is standard normal, $Q=(n-1)S^2/\sigma^2$ is chi-square with $n-1$ degrees of freedom, and $Z$ is independent of $Q$. Their ratio is the Student t pivot $\sqrt n(\bar X-\mu)/S$.

Check Your Understanding

For each problem, use the independent-normal result only when its assumptions are stated or satisfied.

An independent normal sample has $n=5$, $\bar X=12.4$, and $S^2=10$. Under $\mu=11.4$, calculate the pivot and its degrees of freedom. (Answer: $T=1/\sqrt2\approx0.7071$, with 4 degrees of freedom.)
For an independent normal sample of size $8$, the observed standardized mean component is $Z=-1.5$ and the scaled variance component is $Q=14$. Calculate the t ratio and its degrees of freedom. (Answer: $T=-1.5/\sqrt2\approx-1.0607$, with 7 degrees of freedom.)
An independent normal sample has $n=6$, known model standard deviation $\sigma=2$, and observed sample variance $S^2=4$. If $Z=0.8$, calculate $Q$, the t pivot, and its degrees of freedom. (Answer: $Q=5$, $T=0.8$, with 5 degrees of freedom.)
A sample of seven observations is independent but comes from an exponential population rather than a normal population. Does the derivation here establish that the usual sample-mean t ratio has an exact $t_6$ distribution? (Answer: No; the exact result here requires a normal population.)
An independent normal sample has $n=4$, residual sum of squares $\sum_i(X_i-\bar X)^2=18$, $\sigma^2=6$, and $\bar X-\mu=1.5$. Calculate $S^2$, $Q$, the pivot, and its degrees of freedom. (Answer: $S^2=6$, $Q=3$, $T=1.2247$, with 3 degrees of freedom.)

What You'll Learn