Mapping Sample Size to Student t Degrees of Freedom

Why Sample Size Determines the t Row

When a population standard deviation is unknown, a one-sample t statistic uses the sample standard deviation in its denominator. The t distribution used for probabilities and critical values depends on how many independent observations were used to calculate that standard deviation. For a one-sample statistic, that number is summarized by \(n-1\) degrees of freedom, not \(n\).

The practical task is usually short but consequential: identify the usable sample size, subtract one, and use that degrees-of-freedom row in the t table. A wrong row can change a tail-area bound or critical value, even when the test statistic itself has been calculated correctly.

Definition: For \(n\geq 2\) independent observations \(X_1,\ldots,X_n\) from a normal population with mean \(\mu\) and variance \(\sigma^2\), define the sample mean and 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 one-sample Student t statistic for testing or standardizing a hypothesized population mean \(\mu_0\) is

$$ T=\frac{\bar X-\mu_0}{S/\sqrt{n}}. $$ Under the null hypothesis \(H_0:\mu=\mu_0\), this statistic has a Student t distribution with \(\nu=n-1\) degrees of freedom. The exact t result requires independent observations from a normal population and an unknown population variance. For \(n=1\), \(S^2\) and this statistic are not defined.

Why One Degree of Freedom Is Used Up

The sample variance is calculated from deviations around the sample mean, rather than around a known population mean. Those deviations cannot vary freely in all \(n\) directions: their sum must be zero. Once \(n-1\) deviations are specified, the remaining deviation is fixed by that constraint.

The identity below makes the same loss of one degree of freedom visible in the normal-sample result. It separates the total squared deviation around the known \(\mu\) into variation around the sample mean and variation of the sample mean itself:

\sum_{i=1}^{n}(X_i-\mu)^2 = \sum_{i=1}^{n}(X_i-\bar X)^2+n(\bar X-\mu)^2.

For an independent normal sample, the first term, divided by \(\sigma^2\), has a chi-square distribution with \(n-1\) degrees of freedom. In particular, \((n-1)S^2/\sigma^2\sim\chi^2_{n-1}\), and it is independent of \(\bar X\). Meanwhile, \((\bar X-\mu)/(\sigma/\sqrt n)\) is standard normal. Dividing that standard normal variable by the square root of the independent chi-square variable divided by its degrees of freedom gives a Student t variable. Thus the denominator of that ratio has \(n-1\) degrees of freedom, which is why the t statistic uses that same number.

This exact distributional result is specific to the normal-sample setting. If observations are not independent, or the population is not normal, the statistic does not in general have an exact \(t_{n-1}\) distribution. A t approximation may be considered under other conditions, but it should not be described as this exact result.

Mapping the Observations to the Table Row

Here, \(n\) means the number of observations actually included in the one-sample calculation. If a study records 20 people but two measurements are unusable and excluded, then \(n=18\) for the calculation, and the degrees of freedom are \(18-1=17\). The number of observations originally approached or enrolled is not automatically the sample size used in the statistic.

Once \(\nu=n-1\) is found, use the row labelled \(\nu\) in the t table. Check the table heading before reading a value: some tables label upper-tail probabilities, while others label two-tail probabilities. The t distribution is symmetric about zero. For a positive critical value \(c\), a two-sided tail probability of \(\alpha\) means \(P(|T|>c)=\alpha\), so the probability in each tail is \(\alpha/2\).

Table convention: In this tutorial, \(t_{\nu,\alpha}\) denotes the positive value with \(P(T_\nu>t_{\nu,\alpha})=\alpha\). For a two-sided test or confidence interval with total tail probability \(\alpha\), use \(t_{\nu,\alpha/2}\). Always match the table’s probability heading to the tail probability you need.

For example, a 95% two-sided confidence interval uses a total of 0.05 outside its endpoints, or 0.025 in each tail. With \(n=10\), the degrees of freedom are 9, so the critical value is \(t_{9,0.025}=2.262\). It is not the value from the row for 10 degrees of freedom.

Worked Example: Selecting a Critical Value for a Confidence Interval

A hypothetical laboratory measures the concentration of a compound in 10 independent specimens. Assume the measurements are a random sample from a normal population with unknown mean \(\mu\) and unknown variance \(\sigma^2\). The sample mean is 72.0 units and the sample standard deviation is 6.0 units. Construct a 95% t confidence interval for \(\mu\).

Method and degrees of freedom. This is a one-sample interval with unknown population variance, so use the t distribution. There are \(n=10\) usable observations, giving \( \nu=n-1=10-1=9\) degrees of freedom. For a 95% two-sided interval, each tail has probability \(0.05/2=0.025\). The correct table entry is \(t_{9,0.025}=2.262\).

Calculate the standard error and margin. The standard error is \(S/\sqrt n\):

\frac{6.0}{\sqrt{10}} = \frac{6.0}{3.1623} \approx 1.8974, \qquad 2.262(1.8974)\approx 4.2918.

Form the interval. Subtract and add the margin to the sample mean:

72.0-4.2918=67.7082, \qquad 72.0+4.2918=76.2918.

The 95% t confidence interval is \((67.71,\ 76.29)\) units, rounded to two decimal places. As a check, it is centered at 72.0 and has equal margins of about 4.29 units. Common mistake avoided: using the row for 10 degrees of freedom because there are 10 observations. Estimating the mean leaves 9 degrees of freedom, so row 9 is required.

Worked Example: Using the Row for a One-Sided Test

A hypothetical calibration process produces 16 independent readings. Assume a normal population with unknown mean \(\mu\) and variance \(\sigma^2\). The observed sample mean is 103.2 units and the sample standard deviation is 8.0 units. Test \(H_0:\mu=100\) against \(H_A:\mu>100\), using the t table to assess the result at significance levels 0.10 and 0.05.

Method and degrees of freedom. Since the population variance is unknown and the observations are an independent normal sample, use the one-sample t statistic. Here \(n=16\), so \(\nu=16-1=15\). Under \(H_0\), the statistic has a \(t_{15}\) distribution.

Calculate the statistic. The standard error is \(8.0/\sqrt{16}=8.0/4=2.0\). Therefore,

t_{\mathrm{obs}} = \frac{103.2-100}{2.0} = \frac{3.2}{2.0} = 1.6.

Compare with upper-tail critical values. For 15 degrees of freedom, the table gives \(t_{15,0.10}=1.341\) and \(t_{15,0.05}=1.753\). Since \(1.341<1.6<1.753\), the observed statistic exceeds the 0.10 critical value but not the 0.05 critical value. Thus the one-sided p-value is between 0.05 and 0.10: reject \(H_0\) at the 0.10 level, but do not reject at the 0.05 level. The table does not justify reporting an exact p-value. Common mistake avoided: using a two-sided 0.05 column for a one-sided alternative. This test places its rejection area in the upper tail only.

Worked Example: Counting Usable Observations and Bounding a Two-Sided Probability

A hypothetical study planned to analyze nine independent device lifetimes. Two recorded lifetimes are excluded because the measuring equipment failed, leaving seven usable values. Assume those values are a random sample from a normal population with unknown mean \(\mu\) and unknown variance \(\sigma^2\). The usable values have \(\bar X=14.0\) hours and \(S=\sqrt{7}\) hours. For \(H_0:\mu=12\) versus \(H_A:\mu\ne12\), find the t statistic and use the table to bound the two-sided p-value.

Method and degrees of freedom. Count the observations used to calculate the reported mean and standard deviation: \(n=7\), not the nine originally planned. Thus \(\nu=7-1=6\), and under the null the statistic has a \(t_6\) distribution.

Calculate the statistic. Since \(\sqrt n=\sqrt7\), the standard error is \(\sqrt7/\sqrt7=1\) hour. Hence

t_{\mathrm{obs}} = \frac{14.0-12}{1} = 2.0.

Bound the probability. In the \(t_6\) row, \(t_{6,0.05}=1.943\) and \(t_{6,0.025}=2.447\). The observed value lies between them: \(1.943<2.0<2.447\). Therefore the one-tail probability above 2.0 is between 0.025 and 0.05. By symmetry, the two-sided p-value is between \(2(0.025)=0.05\) and \(2(0.05)=0.10\). The table supports \(0.05<p<0.10\). Check: the statistic is positive, but the two-sided calculation includes equally extreme values in both tails. Common mistake avoided: counting the two excluded readings, or doubling the degrees of freedom, when selecting the row.

Practical Checks Before Reading a t Table

Before using a table, write down the observations that entered the calculation and the statistic being used. This simple check catches most row-selection errors. It also keeps the sample size separate from other numbers in the question, such as the number invited, the hypothesized mean, or the number of degrees of freedom itself.

Identify the one-sample statistic.
Confirm that the standard error uses the sample standard deviation \(S\), with the factor \(1/\sqrt n\), rather than a known population standard deviation.

Count the usable observations.
Use the number of independent observations included in \(\bar X\) and \(S\). Excluded or unusable measurements do not contribute to \(n\).

Subtract one.
For the one-sample t statistic, set \(\nu=n-1\). Do not subtract the null mean or use \(n\) as the row label.

Match the tail convention.
Choose the row for \(\nu\), then use an upper-tail probability for a one-sided calculation or half the total tail probability for each side of a two-sided calculation.

The t distribution has heavier tails than the standard normal distribution at finite degrees of freedom. As degrees of freedom increase, its critical values approach the corresponding standard normal critical values. This does not change the row rule: use \(n-1\) for a one-sample t statistic, and select the probability column based on the stated alternative or confidence level.

Key takeaway: For a one-sample Student t statistic based on \(n\) independent normal observations with unknown variance, the degrees of freedom are \(n-1\). Count the usable observations, use the matching t-table row, and distinguish one-tail probabilities from total two-tail probabilities.

Check Your Understanding

For each problem, identify the degrees of freedom before interpreting the stated table entry or critical value.

A one-sample t interval uses 18 independent normal observations and has 90% two-sided confidence. What row and upper-tail probability should be used? (Answer: \(\nu=17\); use \(t_{17,0.05}\).)
A study enrolled 25 people, but one measurement was unusable and excluded before calculating the sample mean and standard deviation. What are \(n\) and the t degrees of freedom? (Answer: \(n=24\), \(\nu=23\).)
For a one-sided upper-tail test, \(n=11\) and the observed statistic is 1.8. In the \(t_{10}\) row, \(t_{10,0.10}=1.372\) and \(t_{10,0.05}=1.812\). What probability bound follows? (Answer: \(0.05<p<0.10\), using \(\nu=10\).)
A two-sided 95% interval is based on nine independent normal observations with unknown variance. What t critical value is needed? (Answer: \(\nu=8\), use \(t_{8,0.025}=2.306\).)
A two-sided test has six usable independent normal observations and an observed statistic of 2.5. In the \(t_5\) row, \(t_{5,0.05}=2.015\) and \(t_{5,0.025}=2.571\). Bound the two-sided p-value. (Answer: \(0.05<p<0.10\), using \(\nu=5\).)

What You'll Learn