Degrees of Freedom Calculator

Did You Know?

Formulas

What Does df Mean Conceptually?

Degrees of freedom represent how many values are free to vary once certain constraints are applied. For example, if you know the mean of n numbers, only n−1 of them can vary freely — the last one is determined by the constraint that the mean is fixed. In regression, each parameter you estimate (slope, intercept) uses up one degree of freedom.

Think of it this way: when you compute a sample variance, you divide by n−1 (not n) because the sample mean is already fixed. That constraint "uses up" one degree of freedom. The same logic applies to every estimated parameter: each one reduces the number of independent pieces of information available for inference.

Worked Example: One-Sample t-test

Suppose you measure the heights of 25 students to test whether their mean height differs from a national average. You have n = 25 observations. The degrees of freedom for the one-sample t-test is df = n − 1 = 25 − 1 = 24. You use this df to look up the critical t-value in a t-table or to compute a confidence interval. With df = 24 and α = 0.05 (two-tailed), the critical t is approximately ±2.064.

For a 3×4 chi-square contingency table (e.g., 3 age groups × 4 income levels), df = (3−1)(4−1) = 2×3 = 6. You need 6 df to interpret the chi-square statistic and determine whether the two categorical variables are associated.

Chart Interpretation

The t-distribution curve shows the sampling distribution of the t-statistic for your computed df. Lower df produces heavier tails (more spread), meaning you need a larger t-value to reject the null hypothesis. As df increases, the curve approaches the normal distribution.

The df comparison bar chart illustrates typical df values across different test types. Use it to get a sense of how df varies: paired designs (fewer pairs) have less df than two-sample designs with larger total n.

When df is small (e.g., df < 10), the t-distribution has noticeably heavier tails than the normal. This is why small-sample t-tests are more conservative — you need stronger evidence to reject H₀.

General Rule: df = n − Parameters

A unifying principle: df = n − number of estimated parameters. For the sample mean, we estimate 1 parameter (μ), so df = n−1. For simple regression, we estimate 2 (intercept and slope), so df = n−2. For multiple regression with p predictors plus intercept, we estimate p+1 parameters, so df_residual = n − p − 1.

This rule helps you derive df for new situations: count how many parameters you estimate from the data, then subtract from the sample size (or relevant total).

Reporting df in Research

When reporting statistical results, always include degrees of freedom. For example: "t(24) = 2.15, p = 0.042" or "F(3, 36) = 4.52, p = 0.008". This allows readers to verify your analysis and assess the strength of your design. Meta-analyses often use df to weight effect sizes.

In APA style, report df in parentheses after the test statistic. For repeated-measures designs, report both between-subjects and within-subjects df where applicable.

Effect of df on Statistical Power

Higher df generally means more statistical power — you can detect smaller effects. This is why larger samples are preferred. However, adding parameters (e.g., more predictors in regression) reduces df and can reduce power if the added parameters do not explain meaningful variance.

For the t-distribution, critical values decrease as df increases. At df = ∞, the t-distribution equals the normal. At df = 1 (Cauchy-like), tails are very heavy and inference is challenging.

Summary Table: df by Test

Frequently Asked Questions

Official Data Sources

Limitations

• • t-tests assume normality (or large n). For non-normal data, consider non-parametric tests.
• • Chi-square tests require expected counts ≥ 5. Use Fisher exact for small tables.
• • ANOVA assumes equal variances and normality. Use Welch ANOVA or Kruskal-Wallis if violated.
• • Regression df assumes linearity and homoscedasticity. Check residuals.

Applications