Using t Tables for Small-Sample Normal Calculations

Why t Tables Need Careful Reading

For small samples from a normal population, calculations involving an estimated standard deviation often lead to a Student t variable. A t table lets you find a critical value or narrow down a tail probability when a calculator or software is unavailable. The arithmetic is usually manageable; the main risks are using the wrong degrees of freedom, confusing a one-sided tail with a two-sided probability, or reading a table heading as the opposite tail from the one it represents.

This tutorial focuses on reading and interpolating t tables for small-sample normal calculations. The examples use the upper-tail convention, but the same reasoning works with other table formats once their headings are interpreted correctly. The techniques are aligned with the SOA VEE Mathematical Statistics syllabus.

Definition: Let \(T\sim t_\nu\), a Student t variable with \(\nu\) degrees of freedom. Define \(t_{\nu,\alpha}\) as the positive number satisfying \(P(T>t_{\nu,\alpha})=\alpha\), where \(0<\alpha<0.5\). By symmetry, \(P(T<-t_{\nu,\alpha})=\alpha\), and \(P(|T|>t_{\nu,\alpha})=2\alpha\).

Read the Table Heading Before the Number

In an upper-tail table, a row gives the degrees of freedom \(\nu\), and a column gives an upper-tail probability \(\alpha\). The entry is \(t_{\nu,\alpha}\), the value with probability \(\alpha\) to its right. For example, in the row \(\nu=5\), the entry in the \(\alpha=0.05\) column is 2.015. Thus \(P(T>2.015)=0.05\) for \(T\sim t_5\); it is not a two-sided probability of 0.05.

Here are selected upper-tail values. Tables may round entries differently in the last digit, and some tables use cumulative probabilities instead of upper-tail probabilities. Always check the heading and follow the convention printed in the table you are using.

Degrees of freedom \(\nu\)	\(\alpha=0.10\)	\(\alpha=0.05\)	\(\alpha=0.025\)	\(\alpha=0.01\)
4	1.533	2.132	2.776	3.747
5	1.476	2.015	2.571	3.365
8	1.397	1.860	2.306	2.896
9	1.383	1.833	2.262	2.821
12	1.356	1.782	2.179	2.681

A two-sided probability splits equally between the two tails because the Student t distribution is symmetric about zero. Consequently, the positive cutoff that leaves total probability \(\alpha\) outside has upper-tail area \(\alpha/2\). This is why a two-sided 5% cutoff uses the \(\alpha=0.025\) column, not the \(\alpha=0.05\) column.

Worked Example: Finding a Two-Sided Critical Value

A hypothetical laboratory measures the response time of \(n=10\) independent test units. Assume the observations are from a normal population and the population variance is unknown. Find the positive critical value for a two-sided 10% calculation based on the usual studentized sample mean.

Method. The studentized mean has a \(t\) distribution with \(n-1\) degrees of freedom under these assumptions. Here \(\nu=10-1=9\). A two-sided 10% calculation places 5% in each tail, so the needed upper-tail probability is \(\alpha=0.05\).

t_{9,0.05}=1.833

The positive critical value is 1.833, and by symmetry the two cutoffs are \(-1.833\) and \(1.833\). The probability between them is \(1-0.10=0.90\). Indeed, each outer tail has probability 0.05, so the combined probability outside the cutoffs is \(0.05+0.05=0.10\).

Check and common mistake avoided. The degrees of freedom are 9, not 10, and the two-sided 10% level calls for the one-sided 5% column. Using the 10% column would give 1.383 and would leave 10% in each tail, or 20% outside in total.

Using Symmetry for Tail Probabilities

A table of positive upper-tail critical values can answer several kinds of probability questions. If \(c>0\), symmetry gives \(P(T<-c)=P(T>c)\). If the event is \(|T|>c\), add the equal left and right tails. If the event is \(|T|<c\), subtract those tails from 1.

P(T<-c)=P(T>c), \qquad P(|T|>c)=2P(T>c), \qquad P(|T|<c)=1-2P(T>c)

When the observed value falls between two table entries, the table generally gives a range for the tail probability. Linear interpolation can provide an estimate between those endpoints. It is an approximation, not an exact identity.

Interpolation rule: Suppose \(t_1<t<t_2\), and the table gives upper-tail probabilities \(p_1=P(T>t_1)\) and \(p_2=P(T>t_2)\), with \(p_1>p_2\). Linear interpolation estimates \(P(T>t)\) by \(p_1+\dfrac{t-t_1}{t_2-t_1}(p_2-p_1)\). Use the two table entries that bracket \(t\); do not extrapolate beyond the table interval.

Why Linear Interpolation Gives an Estimate

The upper-tail probability \(P(T>t)\) decreases as \(t\) increases. Between two known points, interpolation estimates its change at a constant rate. This is the straight-line approximation between the table values. If \(t\) is a fraction \(r\) of the way from \(t_1\) to \(t_2\), then the estimate is the same fraction of the way from \(p_1\) to \(p_2\):

r=\frac{t-t_1}{t_2-t_1}, \qquad \widehat{P(T>t)}=p_1+r(p_2-p_1)

The t-distribution tail is curved rather than exactly a straight line, so the interpolated result is approximate. The closer the table entries are to the target, the more local the approximation. If the requested value is outside the available interval, use a bound from the table or find a table with a suitable interval rather than extending the line past its endpoints.

Identify the distribution and degrees of freedom.
Write down the specified \(t_\nu\) distribution. For a studentized mean from \(n\) independent normal observations, \(\nu=n-1\).

Translate the event into a tail area.
Decide whether the question asks for one tail, both tails, or a central probability. Use symmetry for negative cutoffs.

Bracket the value in the correct row.
Find table entries on either side of the given positive value, and record their matching upper-tail probabilities.

Interpolate only when needed.
Compute the fraction of the distance between the critical values, apply that fraction to the probability difference, and then convert to a two-sided probability if required.

Worked Example: Interpolating a Two-Sided Tail Probability

A statistic has distribution \(T\sim t_5\) and is observed to be \(-2.10\). Estimate \(P(|T|\geq 2.10)\) using the table.

Method. The absolute value is 2.10, so first estimate the right-tail probability \(P(T>2.10)\). In the \(\nu=5\) row, the adjacent entries are \(t_{5,0.05}=2.015\) and \(t_{5,0.025}=2.571\). The value 2.10 lies between them, so these entries bracket it.

The fraction of the critical-value interval from 2.015 to 2.10 is

r=\frac{2.10-2.015}{2.571-2.015} =\frac{0.085}{0.556} \approx 0.1529

The upper-tail probability decreases from 0.05 to 0.025 across this interval. Interpolating gives

P(T>2.10)\approx 0.05+0.1529(0.025-0.05) =0.05-0.003823 =0.04618

By symmetry, the two-sided probability is approximately

P(|T|\geq 2.10)\approx 2(0.04618)=0.09236\approx 0.0924

Check and common mistake avoided. Since 2.10 is greater than 2.015, its right-tail probability must be less than 0.05; the estimate 0.04618 has the correct direction. The two-sided answer is about 0.0924. Do not use a table entry below 2.10 and extrapolate from it: interpolation must use entries on either side of the target.

From a Sample Summary to a Table-Based Probability

For a sample of \(n\) independent observations from a normal population with mean \(\mu\) and unknown standard deviation \(\sigma\), let \(\bar X\) be the sample mean and \(S\) the sample standard deviation. The studentized statistic \((\bar X-\mu)/(S/\sqrt n)\) has a \(t_{n-1}\) distribution. This result applies under the stated normal-sample assumptions; the table calculation does not make the result exact for a nonnormal population or dependent observations.

Once a statistic has been calculated, table lookup proceeds just as it does for a specified \(T\). Calculate the statistic with the given sample summary, use the correct degrees of freedom, bracket its magnitude, and interpret the resulting tail area in the context of the question.

Worked Example: Estimating a Two-Sided Tail Area from a Sample

A hypothetical quality-control study takes an independent sample of \(n=8\) measurements from a normal population. The sample mean is \(\bar x=52.4\), the sample standard deviation is \(s=3.2\), and the reference mean is \(\mu_0=50\). Estimate the two-sided tail probability associated with the studentized statistic.

Method. Under the normal-sample model, the statistic under the reference mean has a \(t\) distribution with \(n-1=7\) degrees of freedom. Its observed value is

t=\frac{\bar x-\mu_0}{s/\sqrt n} =\frac{52.4-50}{3.2/\sqrt 8} =\frac{2.4}{3.2/2.8284} \approx\frac{2.4}{1.1314} \approx 2.121

For \(\nu=7\), the upper-tail entries bracketing 2.121 are \(t_{7,0.05}=1.895\) and \(t_{7,0.025}=2.365\). The interpolation fraction is

r=\frac{2.121-1.895}{2.365-1.895} =\frac{0.226}{0.470} \approx 0.4809

Thus the estimated upper-tail probability is

P(T_7>2.121)\approx 0.05+0.4809(0.025-0.05) \approx 0.03798

The two-sided tail probability is approximately \(2(0.03798)=0.07596\), or 0.0760 to four decimal places. As a check, 2.121 is between the two critical values, so the one-sided tail must be between 0.025 and 0.05, and the two-sided probability must be between 0.05 and 0.10.

Common mistake avoided. The degrees of freedom are \(8-1=7\), not 8. Also, the table interpolation estimates a tail probability; it does not turn the estimate into an exact probability.

Limits of Table-Based Answers

A table may contain only a few tail probabilities, so it cannot always determine a precise answer. If a value lies between two critical values, report a probability interval when interpolation is not requested. For instance, if \(t_{8,0.05}<t<t_{8,0.025}\), then \(0.025<P(T_8>t)<0.05\). For a two-sided event, double both endpoints, giving \(0.05<P(|T_8|>t)<0.10\).

A table may also provide only a rounded critical value. Treat additional digits obtained from arithmetic with appropriate caution: the precision of the table limits the precision of the result. Give an interpolated probability as an estimate, not as an exact value.

Common Table-Reading Traps

Confusing one tail with two. The upper-tail column \(\alpha=0.05\) leaves 5% on one side, or 10% across both sides at the symmetric cutoffs.
Choosing the wrong column for a two-sided level. A two-sided level of 0.05 requires upper-tail area 0.025 for its positive critical value.
Using the sample size as the degrees of freedom. For the studentized mean of \(n\) independent normal observations, the degrees of freedom are \(n-1\).
Interpolating in the wrong direction. As a positive t value increases, its upper-tail probability decreases.
Extrapolating instead of bracketing. If the target exceeds the largest critical value being used, those entries do not bracket it. Use another entry or report the table-supported bound.

Key takeaway: State the degrees of freedom, read the table's probability convention, and translate one-sided or two-sided events before selecting an entry. For an intermediate value, interpolate between the two entries that bracket it; use symmetry to obtain left-tail or two-sided probabilities.

Check Your Understanding

For each problem, use the upper-tail convention and treat any requested interpolation as a linear estimate.

A studentized mean is based on \(n=7\) independent observations from a normal population with unknown variance. What is the degrees of freedom, and what positive critical value leaves 5% in the upper tail? (Answer: \(\nu=6,\ t_{6,0.05}=1.943\))
For \(T\sim t_{10}\), estimate \(P(T>1.90)\) by interpolation. Use \(t_{10,0.05}=1.812\) and \(t_{10,0.025}=2.228\). (Answer: approximately 0.0447)
For \(T\sim t_4\), estimate \(P(|T|\geq 2.50)\), using \(t_{4,0.05}=2.132\) and \(t_{4,0.025}=2.776\). (Answer: approximately 0.0714)
A normal-sample calculation has \(n=9\), \(\bar x=14.6\), \(s=4\), and reference mean \(\mu_0=12\). Calculate the observed studentized statistic and estimate its two-sided tail probability using \(t_{8,0.05}=1.860\) and \(t_{8,0.025}=2.306\). (Answer: \(t=1.950\), two-sided probability approximately 0.0899)
For \(T\sim t_{12}\), estimate \(P(T>1.50)\) using \(t_{12,0.10}=1.356\) and \(t_{12,0.05}=1.782\). (Answer: approximately 0.0831)

What You'll Learn