What Is the t-Distribution? (Definition)
Every t-distribution belongs to a family of curves, one for each value of degrees of freedom (df). With very few degrees of freedom — small samples — the curve is short and wide, putting more probability in the tails than the normal curve does. As df grows, the t-distribution gets taller and narrower, and once df reaches roughly 30, it is nearly indistinguishable from the standard normal distribution.
The heavier tails exist for a practical reason: when you estimate the population standard deviation (σ) using the sample standard deviation (s), that estimate carries its own uncertainty, especially with few data points. The t-distribution builds in a margin for that extra uncertainty, so confidence intervals and hypothesis tests based on small samples are not falsely narrow.
As degrees of freedom increase, the t-distribution curve narrows and its tails thin, approaching the standard normal distribution.
- Shape: Symmetric, bell-shaped, centered at 0, with heavier tails than the normal distribution
- Degrees of freedom (df): Controls the exact shape — typically df = n − 1 for one sample
- Convergence: As df → ∞, the t-distribution approaches the standard normal (Z) distribution
- Used for: t-tests, confidence intervals for the mean, and regression coefficient tests when σ is unknown
- Origin: Developed in 1908 by William Sealy Gosset, published under the pen name "Student"
- Critical value: Found in a t-table using df and α (significance level)
t-Distribution Formula and Degrees of Freedom
x̄ = sample mean
μ = population mean under H₀
s = sample standard deviation
n = sample size
s/√n = standard error
The numerator measures the gap between your observed sample mean and the value claimed under the null hypothesis. The denominator — the standard error — rescales that gap into units of "how many standard errors away" the sample mean falls. A larger |t| means the sample mean sits further from the hypothesized value relative to the natural variability of the data.
Degrees of Freedom Explained
Degrees of freedom represent the number of independent values that are free to vary once certain constraints — like the sample mean — are fixed. For a one-sample t-test, df = n − 1, because once the sample mean is known, only n − 1 of the observations can vary freely; the last one is determined.
| Test | Degrees of Freedom Formula | Example (n₁=12, n₂=15) |
|---|---|---|
| One-sample t-test | df = n − 1 | df = 11 |
| Paired t-test | df = n − 1 (n = number of pairs) | df = 11 (for 12 pairs) |
| Two-sample t-test (equal variances) | df = n₁ + n₂ − 2 | df = 25 |
| Two-sample t-test (Welch's, unequal variances) | Welch–Satterthwaite formula (non-integer, often rounded down) | df ≈ 22–24 |
| Simple linear regression (slope test) | df = n − 2 | df = n − 2 |
Lower df produces a flatter, wider curve with more area in the tails — so the critical value needed to reject H₀ is larger. As df climbs past about 30, the critical values stop changing much and converge toward the fixed z-values from the normal distribution.
t-Distribution vs. Normal Distribution vs. Z-Distribution
The standard normal (Z) distribution and the t-distribution are both symmetric, bell-shaped, and centered at zero — but they answer different questions about uncertainty. The table below summarizes the structural differences.
| Feature | t-Distribution | Standard Normal (Z) Distribution |
|---|---|---|
| Population SD (σ) | Unknown — estimated by sample s | Known |
| Tail thickness | Heavier (more probability in tails) | Lighter |
| Shape | A family of curves, one per df | A single fixed curve |
| Best for sample size | Small samples (n < 30 typically) | Large samples (n ≥ 30) or known σ |
| Critical value at α=0.05, two-tailed | Varies — e.g. ±2.262 at df=9 | Fixed at ±1.96 |
| As n increases | Converges to the normal distribution | Stays the same |
When to Use the t-Distribution vs. the Normal Distribution
📊 t-Distribution vs. Z-Distribution Decision Guide
The t-distribution and the normal distribution are both symmetric and bell-shaped, but the t-distribution has heavier tails to account for the extra uncertainty of estimating the population standard deviation from a small sample. As sample size increases, the t-distribution becomes nearly identical to the normal distribution.
t-Distribution Table (Critical Values)
A t-table lists critical t-values for combinations of degrees of freedom (rows) and significance level / tail type (columns). Find your row by df, then your column by α and whether the test is one-tailed or two-tailed; the intersection is your critical value t*.
| df | One-tail α=0.10 Two-tail α=0.20 |
One-tail α=0.05 Two-tail α=0.10 |
One-tail α=0.025 Two-tail α=0.05 |
One-tail α=0.01 Two-tail α=0.02 |
One-tail α=0.005 Two-tail α=0.01 |
|---|---|---|---|---|---|
| 1 | 1.376 | 1.886 | 12.706 | 31.821 | 63.657 |
| 2 | 1.061 | 1.886 | 4.303 | 6.965 | 9.925 |
| 5 | 1.476 | 2.015 | 2.571 | 3.365 | 4.032 |
| 9 | 1.383 | 1.833 | 2.262 | 2.821 | 3.250 |
| 10 | 1.372 | 1.812 | 2.228 | 2.764 | 3.169 |
| 15 | 1.341 | 1.753 | 2.131 | 2.602 | 2.947 |
| 19 | 1.328 | 1.729 | 2.093 | 2.539 | 2.861 |
| 20 | 1.325 | 1.725 | 2.086 | 2.528 | 2.845 |
| 25 | 1.316 | 1.708 | 2.060 | 2.485 | 2.787 |
| 30 | 1.310 | 1.697 | 2.042 | 2.457 | 2.750 |
| 50 | 1.299 | 1.676 | 2.009 | 2.403 | 2.678 |
| 100 | 1.290 | 1.660 | 1.984 | 2.364 | 2.626 |
| ∞ (Z-table) | 1.282 | 1.645 | 1.960 | 2.326 | 2.576 |
The highlighted column (α = 0.05, two-tailed) is the most commonly used in introductory statistics courses. Notice that the bottom row, labeled df = ∞, matches the critical values from the standard normal z-table — this is the convergence described in the comparison section above. For the complete table covering every df from 1 to 100, see the dedicated t-distribution table page, or download the printable PDF reference.
t-Distribution Examples — 3 Fully Solved
The three examples below show the t-distribution in three of its most common roles: building a confidence interval, running a one-sample t-test, and comparing two independent groups. Formulas follow standard notation from the NIST Engineering Statistics Handbook.
Example 1 — Confidence Interval Using the t-Distribution
Problem: A quality engineer measures the fill weight of 12 cereal boxes and finds x̄ = 498 g with s = 6 g. Construct a 95% confidence interval for the true mean fill weight.
t* = critical value at df = n−1
s/√n = standard error
Degrees of freedom: df = n − 1 = 12 − 1 = 11
Critical value: For a 95% confidence interval (two-tailed, α = 0.05), the t-table gives t* with df = 11 as approximately 2.201
Standard error: SE = s/√n = 6/√12 = 6/3.464 = 1.732
Margin of error: ME = t* × SE = 2.201 × 1.732 = 3.81
Interval: 498 ± 3.81 → (494.19 g, 501.81 g)
✅ Conclusion: We are 95% confident the true mean fill weight of all cereal boxes is between 494.19 g and 501.81 g. Because σ was unknown and n was small, the t-distribution (not the normal distribution) was the correct choice.
Example 2 — One-Sample t-Test
Problem: A coffee shop chain claims its average customer wait time is 4 minutes. A sample of 16 customers shows x̄ = 4.6 minutes with s = 1.2 minutes. At α = 0.05, is the claim supported?
μ₀ = claimed mean = 4
df = n − 1 = 15
Hypotheses: H₀: μ = 4 | H₁: μ ≠ 4 (two-tailed)
α = 0.05. With df = 15, the critical value from the t-table is t* = ±2.131
Test: σ is unknown and n = 16 is small, so the one-sample t-test applies. See the full one-sample t-test guide.
Test statistic:
SE = 1.2/√16 = 1.2/4 = 0.3
t = (4.6 − 4) / 0.3 = 0.6 / 0.3 = 2.0
p-value: For t = 2.0 with df = 15 (two-tailed): p ≈ 0.063
Decision: p = 0.063 > 0.05 → Fail to Reject H₀. Also: |t| = 2.0 < t* = 2.131.
❌ Conclusion: There is not enough evidence at the 5% level to conclude the average wait time differs from 4 minutes, although the result (p = 0.063) is close to the threshold and a larger sample might detect a difference.
Example 3 — Two-Sample t-Test (Independent Groups)
Problem: A teacher compares two teaching methods. Group A (n₁ = 14) scores x̄₁ = 78 with s₁ = 8. Group B (n₂ = 14) scores x̄₂ = 82 with s₂ = 7. Assuming equal variances, test at α = 0.05 whether the methods produce different mean scores.
s_p = pooled standard deviation
df = n₁ + n₂ − 2 = 26
Hypotheses: H₀: μ₁ = μ₂ | H₁: μ₁ ≠ μ₂ (two-tailed). See the full two-sample t-test guide.
α = 0.05. With df = 26, the critical value from the t-table is t* ≈ ±2.056
Pooled standard deviation:
s_p² = [(13)(8²) + (13)(7²)] / 26 = [13(64) + 13(49)] / 26 = (832 + 637) / 26 = 56.5
s_p = √56.5 = 7.52
Test statistic:
SE = s_p √(1/14 + 1/14) = 7.52 × √(0.1429) = 7.52 × 0.378 = 2.843
t = (78 − 82) / 2.843 = −4 / 2.843 = −1.407
p-value: For |t| = 1.407 with df = 26 (two-tailed): p ≈ 0.171
Decision: p = 0.171 > 0.05 → Fail to Reject H₀. |t| = 1.407 < t* = 2.056.
❌ Conclusion: The data does not provide statistically significant evidence (p = 0.171) that the two teaching methods produce different average scores at the 5% level, despite Group B's higher sample mean.
Real-World Applications of the t-Distribution
The t-distribution underlies almost every small-sample statistical procedure in practice. Wherever a sample standard deviation stands in for an unknown population value, the t-distribution governs the resulting critical values and p-values.
Clinical Trials
Small pilot studies and early-phase trials use t-tests built on the t-distribution to compare treatment and control groups when only a handful of patients are enrolled.
Quality Control
Manufacturers sampling a small batch of parts use the t-distribution to build confidence intervals for mean dimensions, weights, or tolerances when σ is not known from historical data.
Regression Analysis
The t-distribution tests whether individual regression coefficients differ significantly from zero — a core part of simple linear regression output.
Psychology & Education Research
Studies with limited participant pools rely on the t-distribution for paired and independent-samples comparisons of test scores, reaction times, and survey responses.
Early-Stage A/B Testing
When an experiment hasn't yet collected enough sessions to justify a normal approximation, t-distribution-based confidence intervals give more honest, wider estimates of the true effect.
Agricultural & Field Research
Field trials comparing crop yields across a small number of plots use t-tests, the original setting for which Gosset developed the distribution at Guinness's experimental farms.
t-Distribution Cheat Sheet
Formula Summary
| Use Case | Formula | df |
|---|---|---|
| t-statistic (one sample) | t = (x̄ − μ)/(s/√n) | n − 1 |
| Confidence interval | x̄ ± t*(s/√n) | n − 1 |
| Two-sample t (pooled) | (x̄₁−x̄₂)/(s_p√(1/n₁+1/n₂)) | n₁+n₂−2 |
| Paired t-test | t = d̄/(s_d/√n) | n − 1 |
| Regression slope test | t = b/SE(b) | n − 2 |
Symbols Glossary
| Symbol | Name | Meaning |
|---|---|---|
| t | t-statistic | Standardized distance of sample mean from μ₀, in SE units |
| t* | Critical value | Threshold from the t-table for a given df and α |
| df | Degrees of freedom | Determines the exact shape of the t-curve |
| x̄ | Sample mean | Average of the observed sample |
| s | Sample standard deviation | Estimate of spread from the sample |
| n | Sample size | Number of observations |
| s/√n | Standard error | Estimated standard deviation of the sample mean |
| α | Significance level | Pre-set probability threshold for rejecting H₀ |
Common Misconceptions About the t-Distribution
| What People Say | Why It's Wrong | What's Correct |
|---|---|---|
| "The t-distribution is only for tiny samples (n < 10)" | The t-distribution is technically correct for any n when σ is unknown | Use it whenever σ is unknown, regardless of sample size; for large n it simply matches the normal distribution closely |
| "Degrees of freedom is just the sample size" | df accounts for constraints like the estimated mean | df = n − 1 for one sample because one parameter (the mean) is estimated from the data |
| "A higher t-value always means a more important result" | Statistical significance ≠ practical significance | A large |t| with a tiny effect size can still be practically meaningless — check effect size too |
| "The t-distribution and normal distribution are the same thing" | They converge only as df → ∞ | At small df, the t-distribution has noticeably heavier tails and larger critical values |
Interactive t-Distribution Calculator
Enter your sample statistics below to compute the t-statistic and p-value. This is the same one-sample calculation engine used for the one-sample t-test; select "T-Test" and enter your sample standard deviation.
🔬 t-Distribution Calculator (One-Sample t-Test)
Frequently Asked Questions
The t-distribution is a continuous probability distribution that resembles a bell curve but has thicker, heavier tails than the normal distribution. It is used to analyze sample data when the population standard deviation is unknown and sample sizes are small.
The t-statistic is calculated as t = (x̄ − μ) / (s / √n), where x̄ is the sample mean, μ is the population mean under the null hypothesis, s is the sample standard deviation, and n is the sample size. Degrees of freedom for a one-sample test equal n − 1.
The primary difference lies in the tails: the t-distribution has heavier, thicker tails than the standard normal distribution. This accounts for the extra uncertainty of estimating the population standard deviation from a small sample. Additionally, the t-distribution is a family of curves defined by degrees of freedom, whereas the standard normal distribution is a single fixed curve.
Use the t-distribution when the population standard deviation σ is unknown and the sample size is small (commonly n < 30). If the sample size is large (n ≥ 30) and σ is still unknown, the t-distribution remains technically correct, though the normal distribution provides a close approximation. See the full hypothesis testing examples guide for worked comparisons.
For a one-sample t-test, degrees of freedom equal the sample size minus one (df = n − 1). For a two-sample t-test with equal variances, df = n₁ + n₂ − 2. For Welch's two-sample t-test with unequal variances, df is calculated using the Welch–Satterthwaite formula and is often a non-integer.
William Sealy Gosset, a chemist at the Guinness Brewery in Dublin, developed the t-distribution in 1908 to handle small-sample quality control problems. Guinness prohibited employees from publishing under their own names, so Gosset published his work under the pseudonym "Student" — hence "Student's t-distribution."
As sample size increases, degrees of freedom rise, causing the tails of the t-distribution to thin and the peak to grow taller. As n approaches infinity, the t-distribution becomes identical to the standard normal (Z) distribution — which is why the bottom row of a t-table (df = ∞) matches the z-table values.
The t-distribution sets the margin of error for confidence intervals on small samples using CI = x̄ ± (t* · s/√n), where t* is the critical value from the t-table corresponding to the chosen confidence level and degrees of freedom. See confidence interval for the mean for a deeper walkthrough.
Sources and References
This guide cross-references the following primary and secondary sources for formulas, critical values, and historical background.
- NIST Engineering Statistics Handbook — Percent Point Function of the t-Distribution. National Institute of Standards and Technology. itl.nist.gov
- Penn State STAT 415 — Introduction to Mathematical Statistics. Penn State Eberly College of Science. online.stat.psu.edu
- OpenStax Introductory Statistics — Ch. 8: Confidence Intervals; Ch. 9–10: Hypothesis Testing. Rice University. openstax.org
- "Student" [Gosset, W.S.] (1908) — "The Probable Error of a Mean." Biometrika, 6(1), 1–25.
- Fisher, R.A. (1925) — Statistical Methods for Research Workers. Edinburgh: Oliver and Boyd.