Why Degrees of Freedom Matter
When a population standard deviation is unknown, inference about a population mean uses the sample standard deviation to estimate it. That extra estimation makes the standardized sample mean more variable than a standard normal variable. The Student t distribution accounts for this additional uncertainty, and its critical value depends on the sample size through the degrees of freedom.
The practical task is precise: identify the correct degrees of freedom, determine how much probability lies in each tail, and select the matching t quantile. Choosing a critical value from the wrong row or tail changes the interval width and its coverage. The calculations below use hypothetical measurement studies; no interest rate or mortality assumption is involved.
For a one-sample mean procedure based on \(n\) independent observations from a normal population with mean \(\mu\) and unknown standard deviation \(\sigma\), the degrees of freedom are \(\nu=n-1\). The sample standard deviation \(s\) is calculated using deviations from the sample mean, which uses one degree of freedom. In this setting, the standardized statistic \((\bar X-\mu)/(s/\sqrt n)\) has a \(t_{n-1}\) distribution. The exact t procedures discussed here rely on the stated normal-sample model.
Choose the Tail Before the Table Entry
A confidence level and a tail probability are related, but they are not interchangeable. For a two-sided confidence interval with confidence level \(1-\alpha\), the total probability outside the interval is \(\alpha\), split equally between its two tails. The positive critical value is therefore \(t_{1-\alpha/2,\nu}\). For a one-sided bound with confidence level \(1-\alpha\), all of the excluded probability is in one tail, so the positive critical value is \(t_{1-\alpha,\nu}\).
For example, a 95% two-sided interval uses the 0.975 quantile, not the 0.95 quantile. A 95% one-sided bound uses the 0.95 quantile. Both procedures use the same degrees of freedom when they are based on the same sample; their critical values differ because the tail allocations differ.
| Procedure | Total excluded probability | Positive critical value |
|---|---|---|
| 95% two-sided interval | 0.05, with 0.025 in each tail | \(t_{0.975,\nu}\) |
| 95% one-sided bound | 0.05 in one tail | \(t_{0.95,\nu}\) |
| 90% two-sided interval | 0.10, with 0.05 in each tail | \(t_{0.95,\nu}\) |
A t table may label columns by upper-tail area rather than cumulative probability. Under an upper-tail convention, the column headed 0.025 gives \(t_{0.975,\nu}\), since \(P(T>t_{0.975,\nu})=0.025\). Always check the table’s convention before reading a value. Use the row for the exact degrees of freedom; do not substitute the sample size itself.
Worked Example: Selecting a Two-Sided Critical Value
A hypothetical lab measures a chemical concentration in a random sample of 10 independent units. Assume the measurements follow a normal distribution with unknown mean \(\mu\) and unknown standard deviation \(\sigma\). The sample mean is 42.0 units and the sample standard deviation is 6.0 units. Find the 95% t interval for \(\mu\). No interest rate or mortality assumption applies.
Method and degrees of freedom. This is a one-sample, two-sided Student t interval. The sample size is \(n=10\), so \(\nu=n-1=9\). With \(\alpha=0.05\), each tail contains \(\alpha/2=0.025\), and the required table entry is \(t_{0.975,9}=2.262\).
Calculate the standard error. The standard error is the sample standard deviation divided by the square root of the sample size:
Calculate the margin and endpoints. The margin of error is \(2.262(1.8974)=4.2919\). Thus the interval is \(42.0\pm4.2919\), with lower endpoint \(42.0-4.2919=37.7081\) and upper endpoint \(42.0+4.2919=46.2919\). Rounded only at the final step, the 95% interval is 37.71 to 46.29 units.
Check and trap avoided. The interval is centered at 42.0 and has total width \(46.29-37.71=8.58\) units, twice the unrounded margin to the displayed precision. Using \(t_{0.95,9}\) would incorrectly allocate all 5% to one tail; the two-sided interval requires \(t_{0.975,9}\).
Why the t Critical Value Is Larger
For a fixed central confidence level, a t distribution with finite degrees of freedom has heavier tails than the standard normal distribution. Its positive critical value is consequently larger than the corresponding normal critical value. As the degrees of freedom increase, the t distribution approaches the standard normal distribution, and its critical values approach the normal quantiles.
The reason is visible in the standard representation of a Student t variable. If \(Z\sim N(0,1)\) and \(U\) has a chi-square distribution with \(\nu\) degrees of freedom, independently, then \(T=Z/\sqrt{U/\nu}\) has a \(t_\nu\) distribution. The denominator is random, rather than fixed at 1 as it is for \(Z\). When \(U/\nu\) happens to be small, division by its square root makes \(|T|\) large. This creates more tail probability and requires a more distant critical value to capture the same central probability.
As \(\nu\) grows, \(U/\nu\) concentrates near 1, so the random denominator is increasingly close to 1. The t distribution then becomes increasingly similar to the standard normal distribution. This explains both the wider intervals at small sample sizes and why the t and normal critical values become close for large samples.
Worked Example: A One-Sided Upper Bound
A hypothetical quality-control study samples 16 independent items. Assume a normal population with unknown mean \(\mu\) and standard deviation \(\sigma\). The measured characteristic has sample mean 120 units and sample standard deviation 8 units. Find a 90% one-sided upper confidence bound for \(\mu\). No interest rate or mortality assumption applies.
Method and degrees of freedom. For an upper bound with 90% confidence, \(\alpha=0.10\) is in the upper tail. The degrees of freedom are \(\nu=16-1=15\), and the required quantile is \(t_{0.90,15}=1.341\). This is not the two-sided 90% critical value, which would use \(t_{0.95,15}\).
Calculate the standard error and bound. Since \(\sqrt{16}=4\), the standard error is \(8/4=2\) units. The upper-bound margin is \(1.341(2)=2.682\) units. The upper confidence bound is \(120+2.682=122.682\), or 122.68 units rounded to two decimal places.
Check and trap avoided. An upper bound is formed by adding the positive critical value to the sample mean; subtracting it would form a lower bound. The result has the appropriate one-sided tail allocation: \(P(T>1.341)\) is approximately 0.10 for 15 degrees of freedom.
How Degrees of Freedom Affect Interval Width
For a fixed sample size, the degrees of freedom determine which t critical value is used. At a fixed confidence level, lower degrees of freedom produce a larger positive critical value. Because the margin of error multiplies the standard error by that critical value, fewer degrees of freedom make the interval wider. The effect is especially noticeable for small samples; it becomes less pronounced as the degrees of freedom increase.
A comparison with a normal critical value isolates the effect of the multiplier. It does not mean the normal procedure is appropriate when the population standard deviation is unknown. In that setting, replacing the t critical value with a normal one would fail to account for the uncertainty from estimating \(\sigma\).
Worked Example: Comparing t and Normal Critical Values
A hypothetical study observes a sample of five independent measurements from a normal population with unknown mean \(\mu\) and unknown standard deviation \(\sigma\). The sample mean is 10.0 units and the sample standard deviation is 2.5 units. Compare the 95% t interval’s width with the width produced by using the 95% standard normal critical value. No interest rate or mortality assumption applies.
Find the t critical value. Here \(n=5\), so \(\nu=5-1=4\). A 95% two-sided interval uses \(t_{0.975,4}=2.776\). For comparison, the 95% standard normal critical value is \(z_{0.975}=1.960\).
Find the common standard error. The standard error is \(2.5/\sqrt5=2.5/2.2361=1.1180\) units.
Calculate both margins and widths. The t margin is \(2.776(1.1180)=3.1036\) units, giving a t interval from \(10.0-3.1036=6.8964\) to \(10.0+3.1036=13.1036\). Its width is \(13.1036-6.8964=6.2072\), or 6.21 units to two decimal places.
The normal-based comparison has margin \(1.960(1.1180)=2.1913\) units and width \(2(2.1913)=4.3826\), or 4.38 units to two decimal places. The t interval is wider because \(2.776>1.960\). For inference with unknown \(\sigma\) under the stated model, the t interval is the appropriate one.
Check and trap avoided. Both comparisons use the same standard error, so the width difference comes from the critical value. Using \(n=5\) rather than \(\nu=4\) to select a t-table row would be an incorrect degrees-of-freedom choice.
A Reliable Selection Process
Before calculating an interval or bound, write down the sample size, degrees of freedom, confidence level, and tail allocation. That short check prevents most critical-value errors.
Decide whether the required result is two-sided, a lower bound, or an upper bound.
For a one-sample t procedure, use \(\nu=n-1\), not \(n\).
For two-sided confidence level \(1-\alpha\), use \(\alpha/2\) in each tail. For a one-sided bound, use \(\alpha\) in its single tail.
For example, a cumulative probability of 0.975 may appear as an upper-tail area of 0.025.
A two-sided interval is centered at \(\bar X\); an upper bound adds the margin, and a lower bound subtracts it.
Common Exam Traps
- Using \(n\) degrees of freedom. For a one-sample t procedure, the correct value is \(n-1\).
- Confusing confidence level with cumulative probability. A 95% two-sided interval uses the 0.975 quantile, not the 0.95 quantile.
- Ignoring a table’s tail convention. A 0.025 upper-tail column corresponds to cumulative probability 0.975.
- Using a normal critical value when \(\sigma\) is unknown. A t critical value accounts for estimating the population standard deviation with \(s\).
- Assuming every t critical value is the same. The value depends on both the tail probability and the degrees of freedom.
Selecting and interpreting t critical values is aligned with the SOA VEE Mathematical Statistics syllabus. The central habit is to settle the degrees of freedom and tail allocation before looking up a number; the interval arithmetic follows only after those choices are correct.
Check Your Understanding
For each problem, identify the degrees of freedom and the appropriate critical value or interval. Round interval endpoints to two decimal places where needed.
- A two-sided 90% interval is based on 12 independent observations from a normal population with unknown standard deviation. Which t quantile and degrees of freedom are required? (Answer: \(\nu=11\), \(t_{0.95,11}=1.796\).)
- A researcher wants a 95% one-sided lower bound from a sample of 8 independent normal observations. Which positive critical value is required? (Answer: \(\nu=7\), \(t_{0.95,7}=1.895\).)
- A two-sided 99% t interval uses a sample of 21 observations. Give the degrees of freedom and critical value. (Answer: \(\nu=20\), \(t_{0.995,20}=2.845\).)
- A hypothetical normal-population study has \(n=25\) and sample standard deviation \(s=5\). Compare the 90% two-sided t and normal interval widths. (Answer: \(\nu=24\), \(t^*=1.711\); the standard error is 1, so the widths are 3.42 for t and 3.29 for normal.)
- A hypothetical sample of 6 independent measurements from a normal population has \(\bar X=18.4\) and \(s=3.0\). Find the 95% two-sided t interval. (Answer: \(\nu=5\), \(t^*=2.571\), standard error \(3/\sqrt6=1.2247\), interval \((15.25,21.55)\).)