T-Distribution Calculator (Critical Values and P-Values)

What the t-distribution is

The t-distribution (also called Student’s t-distribution) is the workhorse probability distribution for hypothesis testing on small samples where the population standard deviation is unknown. William Sealy Gosset published it in 1908 under the pseudonym “Student” while working at Guinness Brewery, where he could not publish under his own name due to company policy. The name stuck.

The probability density function is:

f(t; ν) = Γ((ν+1)/2) / [√(νπ) · Γ(ν/2)] × (1 + t²/ν)^(−(ν+1)/2)

Where ν (the Greek letter nu) is the degrees of freedom and Γ is the gamma function. The distribution is symmetric around zero, bell-shaped like the normal distribution, but with heavier tails. For ν approaching infinity, the t-distribution converges to the standard normal.

Why heavier tails matter

With a small sample, you have less information about the underlying population, so extreme values of the test statistic are relatively more likely than the normal distribution would suggest. The t-distribution accounts for this by giving more probability mass to the tails. A 95 percent critical value for the normal distribution is 1.96; the equivalent t-critical value at ν = 5 is 2.57. As ν grows, the critical value shrinks back toward 1.96. By ν = 30, the difference is already small (2.04 vs 1.96).

This is why textbooks often say “use the t-distribution when n < 30 and the normal distribution when n ≥ 30.” The rule is approximate. Modern practice is to always use the t-distribution for hypothesis tests with sample standard deviation, regardless of sample size, because t-table values for large df are almost identical to normal-table values anyway.

The two main uses

Forward problem (find p-value from t): You computed a t-statistic from your data (e.g., t = 2.3 with df = 15). What is the probability of observing a value this extreme by chance under the null hypothesis? The answer is the p-value, found by evaluating the t-distribution’s tail integral. If p < 0.05, your result is typically considered statistically significant.

Inverse problem (find critical t from alpha): You want to construct a 95 percent confidence interval (alpha = 0.05, two-tailed). What is the t-value that has 2.5 percent of the distribution in each tail? Look up the critical t for that alpha and df, then build your CI as point estimate ± critical_t × standard_error.

This calculator handles both directions.

One-tailed vs two-tailed

A one-tailed test asks “is my parameter greater than (or less than) a specific value?” — only one tail of the distribution matters. The critical t at alpha = 0.05 with df = 10 is 1.812 (one-tailed) or 2.228 (two-tailed). The two-tailed critical value is larger because you split the alpha between both tails (2.5 percent each), requiring more extreme values to reject.

Use one-tailed only when the alternative hypothesis is genuinely directional (e.g., a new drug is BETTER than placebo, not just different). One-tailed gives more statistical power for the same alpha, but it requires committing to the direction before seeing the data. Most published research uses two-tailed by default because it is more conservative and avoids researcher-degrees-of-freedom issues.

Standard critical t-values (reference table)

For practical work, memorizing the asymptotic values (1.645 for one-tailed at 0.05, 1.96 for two-tailed at 0.05, 2.58 for two-tailed at 0.01) covers most cases when df is reasonably large. For small df, you need the actual t-table or this calculator.

How df is determined

Degrees of freedom depend on the test:

• One-sample t-test (testing one mean against a value): df = n − 1
• Independent two-sample t-test (equal variances): df = n₁ + n₂ − 2
• Welch’s t-test (unequal variances): df ≈ (s₁²/n₁ + s₂²/n₂)² / [(s₁²/n₁)²/(n₁−1) + (s₂²/n₂)²/(n₂−1)] (fractional)
• Paired t-test: df = n − 1 where n is number of pairs
• Linear regression slope test: df = n − k − 1 where k is number of predictors
• Chi-square test: different distribution, but follows similar tail-area logic with its own df

Always check what df formula applies to your specific test.

A worked example

You measure the heights of 12 trees in a forest and compute a sample mean of 18.4 m with a sample standard deviation of 3.2 m. The expected population mean (from other forests) is 17 m. Is your forest significantly taller?

t = (18.4 − 17) / (3.2 / √12) = 1.4 / 0.924 = 1.515

Degrees of freedom: df = n − 1 = 11

Looking up the t-distribution with df = 11:

• Two-tailed p-value at t = 1.515: about 0.158
• This is greater than 0.05, so you fail to reject the null hypothesis. The 1.4 m difference is not statistically significant at this sample size.

To get significance at α = 0.05 (two-tailed, df = 11), you would need |t| > 2.201, requiring a larger effect or a larger sample.

Why “Student”

Gosset worked at Guinness on quality control problems with small barley samples — large samples were expensive. The brewery wanted to keep his contributions secret to avoid revealing their methods. Allowing publication only under a pseudonym was the compromise. The pseudonym stuck because the methodology became foundational to statistics, and renaming after the original author proved impossible after the field adopted “Student’s t” universally.

Computational notes

The t-distribution PDF involves the gamma function and a power of (1 + t²/ν). For small ν, this has heavy tails; numerical integration over the tails requires care at very small ν (the variance is undefined for ν ≤ 2, and the mean is undefined for ν ≤ 1). This calculator uses standard incomplete-beta-function expansions that are accurate to 6 to 8 decimal places across a wide range of ν and t values.

For modern statistical software, t-distribution tables are obsolete: tools like R, Python (scipy.stats), Excel, and this calculator compute exact values from the formula on demand. The historical t-tables you see in textbooks were the only practical option before electronic computers; today they’re a reference convenience.

Relationship to other distributions

The t-distribution arises naturally when you divide a standard normal random variable by the square root of a chi-square random variable divided by its degrees of freedom. Symbolically: t = Z / √(X²/ν), where Z is N(0,1) and X² is chi-square with ν degrees of freedom, independent of Z.

This construction means t shows up wherever a normal-distributed statistic gets normalized by an estimated standard deviation (rather than the true standard deviation). The F-distribution, used in ANOVA, is a ratio of two chi-squares and shares some structural similarity. Both reduce to the normal in appropriate limits.