Written and maintained by CASRAI Editorial Board
Published
On this page
In a one-way analysis of variance with k groups and N observations in total, the degrees of freedom are k − 1 between groups, N − k within groups (error) and N − 1 in total. The F statistic is the ratio of two mean squares, so it always carries two degrees of freedom: the numerator df (k − 1) and the denominator df (N − k). You need both to look up a critical value or to report the result as F(df₁, df₂).
What degrees of freedom mean in ANOVA
ANOVA splits the total variation in a data set into a part explained by the grouping factor (treatments) and a part left over as error. Each sum of squares is divided by its degrees of freedom to give a mean square, which is a variance estimate. Degrees of freedom count how many independent pieces of information go into each sum of squares: the group means can vary freely except that they must be consistent with the grand mean, and each observation can vary around its own group mean except that the deviations within a group must sum to zero.
The model behind the one-way table writes each observation as Yij = μ + τi + εij, with errors assumed independent and normally distributed with constant variance (NIST/SEMATECH e-Handbook, section 7.4.3.2). The degrees of freedom follow from that structure.
Analysis of variance degrees of freedom formulas
The NIST/SEMATECH e-Handbook (section 7.4.3.3) lays out the one-way ANOVA table as follows. Every symbol is defined below the list.
- Treatments (between groups): df = k − 1; MST = SST / (k − 1)
- Error (within groups): df = N − k; MSE = SSE / (N − k)
- Total (corrected): df = N − 1
- Test statistic: F = MST / MSE
k is the number of groups (levels of the factor), N is the total number of observations across all groups (the sum of the group sizes ni), SST is the treatment sum of squares, SSE is the error sum of squares, MST and MSE are the corresponding mean squares. The identity (k − 1) + (N − k) = N − 1 always holds, which gives you a quick arithmetic check on any ANOVA table.
Why F has two degrees of freedom
F is a ratio of two independent variance estimates, and each estimate has its own degrees of freedom. The numerator (MST) uses k − 1 and the denominator (MSE) uses N − k. The F distribution is therefore a family indexed by the pair (df₁, df₂), and the critical value for a given significance level α is read from the table row and column for that pair. The test rejects the hypothesis of equal group means when the computed F exceeds that tabular value (NIST/SEMATECH, section 7.4.3.3).
Worked example: calculating each df step by step
The NIST handbook illustrates one-way ANOVA with resistor readings taken at three temperatures, five readings each. The data are: level 1 = 6.9, 5.4, 5.8, 4.6, 4.0 (mean 5.34); level 2 = 8.3, 6.8, 7.8, 9.2, 6.5 (mean 7.72); level 3 = 8.0, 10.5, 8.1, 6.9, 9.3 (mean 8.56).
- Count groups and observations. k = 3 groups, ni = 5 each, so N = 3 × 5 = 15.
- Between-groups df. k − 1 = 3 − 1 = 2.
- Within-groups df. N − k = 15 − 3 = 12.
- Total df. N − 1 = 15 − 1 = 14. Check: 2 + 12 = 14.
- Sums of squares. The grand total is 108.1, so the grand mean is 108.1 / 15 = 7.2067. SST = 27.897 and SSE = 17.452, giving a total of 45.349.
- Mean squares. MST = 27.897 / 2 = 13.949. MSE = 17.452 / 12 = 1.454.
- F statistic. F = 13.949 / 1.454 = 9.59.
- Critical value. For α = 0.05 with 2 and 12 degrees of freedom, the critical F is 3.89. Because 9.59 > 3.89, the equal-means hypothesis is rejected; the exact p-value is about 0.0032.
| Source | SS | df | MS | F |
|---|---|---|---|---|
| Treatments | 27.897 | 2 | 13.949 | 9.59 |
| Error | 17.452 | 12 | 1.454 | |
| Total (corrected) | 45.349 | 14 |
Sums of squares, mean squares and F reproduce the NIST table; the p-value and the critical value 3.89 were recomputed from the F distribution with 2 and 12 degrees of freedom (critical value 3.885 before rounding).
Interpreting the degrees of freedom and the F result
| What you see | How to read it |
|---|---|
| Numerator df = k − 1 | Tells you how many groups were compared: df₁ = 2 means three groups. |
| Denominator df = N − k | Reflects how much information estimates the error variance; here 12 for 15 observations in 3 groups. More error df means a tighter critical value. |
| F larger than the critical value for (df₁, df₂) | Reject equal means at that α. Here 9.59 exceeds 3.89. |
| df₁ + df₂ ≠ total df | The identity (k − 1) + (N − k) = N − 1 is broken, so a count is wrong. |
| Significant F | At least one group mean differs; it does not say which. Follow-up comparisons are needed. |
For the example, the proportion of total variation explained by temperature is SST / SStotal = 27.897 / 45.349 ≈ 0.62 (an eta-squared computed from the table above).
Two-way ANOVA degrees of freedom
With two factors A (a levels) and B (b levels) and r replicates per cell, the balanced design has N = a × b × r observations. NIST section 7.4.3.7 gives: factor A df = a − 1, factor B df = b − 1, interaction AB df = (a − 1)(b − 1), error df = N − ab, total df = N − 1. For its example with a = 3, b = 4 and r = 2 (N = 24) these are 2, 3, 6, 12 and a total of 23, and 2 + 3 + 6 + 12 = 23. The error df is N − ab rather than N − k because the model now estimates one mean per cell (ab cells), not one per group. For a side-by-side view of the two designs see one-way vs two-way ANOVA; the data table below collects both sets of formulas with the numeric examples.
Assumptions and common mistakes
- Using N instead of k. The numerator df is k − 1 (groups), not N − 1. Confusing the two is the most frequent counting slip.
- Miscounting groups. Count the levels of the factor, not the number of participants per level.
- Reporting only one df. F needs both: F(2, 12), not F(12).
- Using the one-way error df in a two-way design. Use N − ab (balanced case) and remember the interaction term has its own df.
- Ignoring assumptions. The F test relies on independent observations, normally distributed errors and equal error variances across groups (NIST section 7.4.3.2). Degrees of freedom do not repair a violated assumption.
- Reading a significant F as a pairwise finding. It only says the group means are not all equal.
How to report ANOVA degrees of freedom (APA style)
Report the F statistic with both degrees of freedom in parentheses, then the exact p-value. For the example: F(2, 12) = 9.59, p = .003. Adding an effect size is good practice, for instance η² = .62. Italicise the statistical symbols. See also the ANOVA dictionary entry and the guide to types of test statistics.
How to calculate ANOVA degrees of freedom in Excel, R or Python
Excel
Degrees of freedom are not computed for you by a formula; derive them from the counts, then use the F functions. With k = 3 and N = 15 in cells, df₁ = k − 1 and df₂ = N − k:
=F.DIST.RT(9.59, 2, 12) returns the p-value (about 0.0032)
=F.INV.RT(0.05, 2, 12) returns the critical F (about 3.89)
The Data Analysis ToolPak option “Anova: Single Factor” prints the df column directly. See statistical analysis in Excel.
R
y <- c(6.9,5.4,5.8,4.6,4.0, 8.3,6.8,7.8,9.2,6.5, 8.0,10.5,8.1,6.9,9.3)
g <- factor(rep(c("L1","L2","L3"), each = 5))
summary(aov(y ~ g)) # Df column: 2 and 12
qf(0.95, 2, 12) # critical value
pf(9.59, 2, 12, lower.tail = FALSE) # p-value
Python
from scipy import stats
l1=[6.9,5.4,5.8,4.6,4.0]; l2=[8.3,6.8,7.8,9.2,6.5]; l3=[8.0,10.5,8.1,6.9,9.3]
F, p = stats.f_oneway(l1, l2, l3) # F about 9.59
k, N = 3, 15
df1, df2 = k - 1, N - k # 2, 12
crit = stats.f.ppf(0.95, df1, df2) # about 3.89
FAQ
What are the degrees of freedom for a one-way ANOVA?
Between groups k − 1, within groups N − k, total N − 1, where k is the number of groups and N the total sample size.
How do I find degrees of freedom for an F test in ANOVA?
Numerator df is the between-groups df (k − 1) and denominator df is the within-groups df (N − k). Use that pair to read the critical F value.
Why is the within-groups df N − k?
Each of the N observations contributes information, but k group means were estimated from the data, which uses up k degrees of freedom.
Can the degrees of freedom be non-integers?
In the standard equal-variance one-way ANOVA the df are whole numbers. Corrected procedures for unequal variances use adjusted, fractional df; that is outside the standard table above.
How many degrees of freedom does a two-way ANOVA have?
(a − 1) for factor A, (b − 1) for factor B, (a − 1)(b − 1) for the interaction, N − ab for error and N − 1 in total in a balanced design.
Sources
- NIST/SEMATECH e-Handbook of Statistical Methods, section 7.4.3.3, One-way ANOVA: itl.nist.gov/div898/handbook/prc/section4/prc433.htm (read 2026-11-12).
- NIST/SEMATECH e-Handbook, section 7.4.3.2, The one-way model: prc432.htm (read 2026-11-12).
- NIST/SEMATECH e-Handbook, section 7.4.3.7, Two-way ANOVA: prc437.htm (read 2026-11-12).
Data: ANOVA degrees of freedom formulas and worked numeric examples (one-way and two-way)
Degrees-of-freedom formulas for each source of variation in balanced one-way and two-way ANOVA with the numeric values for the NIST/SEMATECH worked examples, compiled from the NIST e-Handbook ANOVA sections.
| Design and source of variation | Formula | Example df |
|---|---|---|
| One-way: treatments (between) | k - 1 | 2 |
| One-way: error (within) | N - k | 12 |
| One-way: total | N - 1 | 14 |
| Two-way: factor A | a - 1 | 2 |
| Two-way: factor B | b - 1 | 3 |
| Two-way: interaction AB | (a - 1)(b - 1) | 6 |
| Two-way: error | N - ab | 12 |
| Two-way: total | N - 1 | 23 |
Sources
- NIST/SEMATECH e-Handbook 7.4.3.3 One-way ANOVA (accessed 2026-11-12)
- NIST/SEMATECH e-Handbook 7.4.3.7 Two-way ANOVA (accessed 2026-11-12)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







