Calculate degrees of freedom for t-tests, ANOVA, chi-square, regression, and Welch's test with step-by-step formulas and a comprehensive reference table.
Degrees of freedom (df) represent the number of independent pieces of information available to estimate a parameter. They appear in virtually every statistical test — t-tests, ANOVA, chi-square, regression — and using the wrong df produces incorrect p-values and critical values. Despite their importance, df formulas vary by test and are easy to confuse.
This calculator computes degrees of freedom for ten common statistical tests: one-sample t, two-sample t (equal variance), Welch's t (unequal variance with Satterthwaite approximation), paired t, one-way ANOVA, two-way ANOVA, chi-square goodness of fit, chi-square independence, simple regression, and multiple regression. Each test displays all relevant df components (between, within, total, residual) with the exact formula used.
The presets cover the most common scenarios, and the reference table provides a single-page summary of df formulas for all tests. Whether you're a student working through homework or a researcher double-checking a statistical analysis, this tool ensures you get the df right every time.
Degrees of freedom formulas differ across statistical tests, and mixing them up is a common source of errors in homework, exams, and published research. This calculator covers all major tests in one place, eliminating the need to remember or look up individual formulas.
The step-by-step display shows exactly how each df component is computed from your inputs, making it an excellent learning tool. The comprehensive reference table serves as a permanent cheat sheet. For Welch's t-test, the Satterthwaite approximation is computed automatically — a calculation that's tedious by hand and easy to get wrong.
One-Sample t: df = n − 1 Two-Sample t: df = n₁ + n₂ − 2 Welch's t: df = (s₁²/n₁ + s₂²/n₂)² / [(s₁²/n₁)²/(n₁−1) + (s₂²/n₂)²/(n₂−1)] Paired t: df = n_pairs − 1 One-Way ANOVA: df_between = k − 1, df_within = N − k Two-Way ANOVA: df_A = a−1, df_B = b−1, df_AB = (a−1)(b−1) Chi-Square GoF: df = k − 1 Chi-Square Independence: df = (r−1)(c−1) Regression: df_regression = p, df_residual = n − p − 1
Result: df_between = 2, df_within = 57, df_total = 59
With 3 groups and 60 total observations, df_between = 3 − 1 = 2 (numerator for F-test), df_within = 60 − 3 = 57 (denominator), and df_total = 59. The F-test uses F(2, 57).
Imagine you have 5 numbers that must average to 10 (sum to 50). You can freely choose the first 4 numbers, but the 5th is forced by the constraint. You have 4 degrees of freedom. This extends to statistical estimation: every parameter you estimate from the data consumes one degree of freedom, leaving fewer for estimating variability.
In a one-sample t-test, you estimate the population mean from the data, consuming 1 df. With n observations, df = n − 1. In two-sample tests, you estimate two group means, but the pooled variance formula already accounts for this, giving df = n₁ + n₂ − 2.
When group variances are unequal, the standard t-test df formula doesn't work because the pooled variance is no longer appropriate. Welch proposed using separate variance estimates, but the resulting test statistic doesn't follow an exact t-distribution. Satterthwaite's approximation finds the effective df by matching the first two moments of the distribution. The result is typically a non-integer df between min(n₁, n₂) − 1 and n₁ + n₂ − 2.
In regression and machine learning, the trade-off between model complexity and residual df is central. Adding predictors improves fit but reduces df_residual. Adjusted R² penalizes for lost df, and information criteria (AIC, BIC) explicitly include df as a penalty term. The principle of parsimony says to use the fewest parameters that adequately explain the data — which means preserving as many degrees of freedom as possible.
df determine the shape of the sampling distribution used to compute p-values and critical values. Using wrong df gives wrong p-values, potentially leading to incorrect conclusions about statistical significance.
If you have n data points and know their mean, only n−1 values are free to vary — the last one is determined by the constraint that they must sum to n × mean. Each constraint (estimated parameter) removes one degree of freedom.
Use Welch's when you cannot assume equal variances between groups. The Welch-Satterthwaite approximation adjusts df downward, making the test more conservative. Many statisticians recommend always using Welch's as the default.
In one-way ANOVA, total df (N−1) splits into between-groups df (k−1) and within-groups df (N−k). The between df measures variation explained by group membership; within df measures residual variation. The F-statistic is the ratio of between to within mean squares.
Low df means heavier tails on the t-distribution, requiring larger test statistics for significance. This reduces statistical power — the ability to detect real effects. It's why larger samples generally provide more powerful tests.
Each predictor added to a regression model uses one degree of freedom from the residual. With n observations and p predictors, df_residual = n − p − 1 (the extra 1 accounts for estimating the intercept). Too many predictors relative to n leads to overfitting.