What is the F-distribution percent point?
The F-distribution percent point (also called the inverse F, the quantile, or the critical value) is the value x such that the cumulative F-distribution with degrees of freedom d1 (numerator) and d2 (denominator) equals a chosen probability. It is the inverse of the F-distribution CDF and is exactly the critical value you look up in F-tables for ANOVA, regression overall significance tests, and variance-ratio (equality-of-variances) tests.
How to use the calculator
Pick a cumulative mode. Choose Lower cumulative P if your probability is \(P = \Pr(F \le x)\) — for example P = 0.95 returns the value below which 95% of the distribution lies. Choose Upper cumulative Q if you have a tail probability \(Q = \Pr(F > x)\) — for example Q = 0.05 returns the usual upper 5% critical value. Enter the probability (strictly between 0 and 1), the numerator degrees of freedom d1, and the denominator degrees of freedom d2, then submit.
The formula explained
The F-distribution CDF is written using the regularized incomplete beta function I: with \(a = d_1/2\), \(b = d_2/2\) and the substitution \(w = d_1 \cdot x / (d_1 \cdot x + d_2)\), we have $$\text{F-CDF}(x) = I_w(a, b).$$ To invert it, the calculator solves \(I_w(a, b) = \text{targetP}\) for \(w\) by bisection on \((0, 1)\), evaluating \(I_w\) with a Lanczos log-gamma and the Numerical Recipes continued fraction. It then back-substitutes $$x = \frac{d_2 \cdot w}{d_1 \cdot (1 - w)}.$$ In Upper mode the target probability becomes \(\text{targetP} = 1 - Q\).
Worked example
For P = 0.95, d1 = 5, d2 = 10 we seek the upper 5% F critical value. With \(a = 2.5\) and \(b = 5\), inverting \(I_w(2.5, 5) = 0.95\) gives \(w \approx 0.6245\), and $$x = \frac{10 \times 0.6245}{5 \times 0.3755} \approx 3.3258,$$ matching the tabulated \(F_{0.05}(5, 10) = 3.3258\).
F Critical Values Table (Upper 5% Level)
The table lists upper-tail critical values \(F_{0.05}(d_1, d_2)\), the point with 5% of the F-distribution to its right. These correspond to an upper-tail probability \(Q = 0.05\), so the lower-tail probability is \(P = 0.95\). To reproduce a value, enter P = 0.95 (or Q = 0.05) with the numerator degrees of freedom \(d_1\) and denominator degrees of freedom \(d_2\). Columns are \(d_1\) (numerator) and rows are \(d_2\) (denominator). Values are rounded to 4 decimals.
| d2 \ d1 | 1 | 2 | 3 | 4 | 5 | 6 | 8 | 10 | 12 |
|---|---|---|---|---|---|---|---|---|---|
| 5 | 6.6079 | 5.7861 | 5.4095 | 5.1922 | 5.0503 | 4.9503 | 4.8183 | 4.7351 | 4.6777 |
| 10 | 4.9646 | 4.1028 | 3.7083 | 3.4780 | 3.3258 | 3.2172 | 3.0717 | 2.9782 | 2.9130 |
| 15 | 4.5431 | 3.6823 | 3.2874 | 3.0556 | 2.9013 | 2.7905 | 2.6408 | 2.5437 | 2.4753 |
| 20 | 4.3513 | 3.4928 | 3.0984 | 2.8661 | 2.7109 | 2.5990 | 2.4471 | 2.3479 | 2.2776 |
| 30 | 4.1709 | 3.3158 | 2.9223 | 2.6896 | 2.5336 | 2.4205 | 2.2662 | 2.1646 | 2.0921 |
| 60 | 4.0012 | 3.1504 | 2.7581 | 2.5252 | 2.3683 | 2.2541 | 2.0970 | 1.9926 | 1.9174 |
| 120 | 3.9201 | 3.0718 | 2.6802 | 2.4472 | 2.2899 | 2.1750 | 2.0164 | 1.9105 | 1.8337 |
| ∞ | 3.8415 | 2.9957 | 2.6049 | 2.3719 | 2.2141 | 2.0986 | 1.9384 | 1.8307 | 1.7522 |
Example readings: \(F_{0.05}(5,10) = 3.3258\), \(F_{0.05}(1,10) = 4.9646\) and \(F_{0.05}(2,10) = 4.1028\). The ∞ row equals \(\chi^2_{0.05}(d_1)/d_1\); for \(d_1 = 1\) this is \(1.96^2 = 3.8415\).
Interpreting the Critical Value
Right-tailed test. In ANOVA and regression F-tests, the null hypothesis \(H_0\) is rejected at significance level \(\alpha\) when the observed statistic exceeds the upper-tail critical value:
$$F_{\text{obs}} > F_{\alpha}(d_1, d_2) \;\Rightarrow\; \text{reject } H_0.$$
Here \(d_1\) is the numerator degrees of freedom (for example, the number of groups minus 1 in one-way ANOVA) and \(d_2\) is the denominator degrees of freedom (for example, the residual degrees of freedom).
P and Q. The calculator accepts a lower-tail probability \(P = \Pr(F \le x)\) or an upper-tail probability \(Q = \Pr(F > x)\), with \(P = 1 - Q\). The significance level maps directly to the upper tail, \(\alpha = Q\), so a 5% test uses \(Q = 0.05\), which is \(P = 0.95\).
Two-sided variance-ratio tests. To test whether two variances are equal against a two-sided alternative, split the significance level and use two critical values, \(F_{\alpha/2}(d_1, d_2)\) (upper) and \(F_{1-\alpha/2}(d_1, d_2)\) (lower). Reject \(H_0\) if the ratio falls above the upper value or below the lower value. The lower value follows from the reciprocal identity:
$$F_{1-\alpha}(d_1, d_2) = \frac{1}{F_{\alpha}(d_2, d_1)}.$$
Note that the degrees of freedom swap. For example, from the table \(F_{0.05}(10, 5) = 4.7351\), so \(F_{0.95}(5, 10) = 1/4.7351 \approx 0.2112\).
Behavior as \(d_2\) grows. Each column of the table decreases as \(d_2\) increases, because the denominator estimate becomes more precise. As \(d_2 \to \infty\), the critical value approaches \(\chi^2_{\alpha}(d_1)/d_1\). The table's ∞ row shows this limit.
Shape. The F-distribution is defined only for \(x > 0\) and is right-skewed, especially for small degrees of freedom. This is why the lower and upper critical values are not symmetric around 1 and why the reciprocal identity is needed for the lower tail.
FAQ
What is the difference between P and Q? P is the lower-tail (cumulative) probability up to x; Q is the upper-tail probability beyond x. They are related by \(P = 1 - Q\).
Why must the probability be strictly between 0 and 1? A probability of 0 corresponds to the quantile \(x = 0\) and a probability of 1 corresponds to \(x \to \infty\), neither of which is a finite, meaningful critical value.
Can degrees of freedom be non-integer? Yes — the math works for any positive real d1 and d2, though ANOVA applications normally use integers.