Overview
Degrees of freedom, df, are the number of values that are free to vary once some quantities are fixed. Many statistical tests need them to find the right critical values.
Enter the sample size, estimated parameters and number of groups, then a test code: 1 general (n − parameters), 2 chi-square goodness of fit (groups − 1), 3 one-sample t (n − 1) or 4 ANOVA within-groups (n − groups).
Worked examples
| Inputs | Result |
|---|---|
| Sample Size (n)120Estimated Parameters4Number of Groups3TestGeneral | Degrees of freedom (general): 116Inputs n=120, parameters=4, groups=3 |
Show 3 more examples
| Sample Size (n)30Estimated Parameters1Number of Groups1Testt-test | Degrees of freedom (one-sample t): 29Inputs n=30, parameters=1, groups=1 |
| Sample Size (n)60Estimated Parameters1Number of Groups4TestANOVA | Degrees of freedom (ANOVA within-groups): 56Inputs n=60, parameters=1, groups=4 |
| Sample Size (n)200Estimated Parameters1Number of Groups6TestChi-square | Degrees of freedom (chi-square goodness-of-fit): 5Inputs n=200, parameters=1, groups=6 |
Degrees of freedom (general) by sample size
With the calculator's starting values, the degrees of freedom (general) is 116. The table shows how it changes with the sample size.
| Sample Size (n) | Degrees of freedom (general) |
|---|---|
| 60 | 56 |
| 80 | 76 |
| 100 | 96 |
| 120starting value | 116 |
| 140 | 136 |
| 160 | 156 |
| 180 | 176 |
Other inputs: Estimated Parameters: 4 · Number of Groups: 3. Change them in the calculator above for your own numbers.
How it works
Each estimated quantity uses up one degree of freedom. A one-sample t-test estimates the mean, so df = n − 1. A goodness-of-fit test with k categories has k − 1. ANOVA's within-groups df is n − k. Regression residual df is n − the number of estimated coefficients.
Practical tips
- 120 observations with 4 estimated parameters gives 116 general df.
- A chi-square test of independence for an r × c table uses (r − 1)(c − 1).
- A two-sample t-test with equal variances uses n₁ + n₂ − 2.
Frequently asked questions
Why does df matter?
It sets the shape of the t, chi-square and F distributions, and so the critical values.
What's df for a contingency table?
(rows − 1) × (columns − 1).
What if df is less than 1?
The test can't be run. You need more data or fewer parameters.