Using the F Table to Evaluate an ANOVA

Why Compare an ANOVA Statistic With a Table?

A one-way ANOVA produces an $F$ statistic that measures how large the variation between group means is relative to the variation within the groups. The statistic alone does not determine the test decision: it must be judged against a reference distribution. An F table provides a critical value for a chosen significance level and a specified pair of degrees of freedom. Comparing the calculated statistic with that value gives a direct reject-or-not decision.

This tutorial assumes that the ANOVA calculation has produced its mean squares and $F$ statistic. The focus here is how to locate the appropriate table entry and interpret the comparison, not how to reconstruct the ANOVA table.

Model and assumptions: For group $i$, the observation on individual $j$ is modeled as $Y_{ij}=\mu_i+\epsilon_{ij}$. The standard one-way ANOVA F test assumes independent observations, normally distributed errors within each group, and a common error variance $\sigma^2$ across groups. The table-based decision is appropriate when these assumptions are reasonable for the analysis.

The Hypotheses and the F-Table Entry

For $k$ groups, the null hypothesis is that all population means are equal, $H_0:\mu_1=\mu_2=\cdots=\mu_k$. The alternative is that the means are not all equal, $H_a:$ not all $\mu_i$ are equal. The alternative does not say that every mean differs from every other mean.

The ANOVA statistic is $F=MS_{\text{between}}/MS_{\text{within}}$. Its numerator degrees of freedom are $df_1=k-1$, and its denominator degrees of freedom are $df_2=N-k$, where $N$ is the total number of observations. These degrees of freedom determine which F distribution, and therefore which table entry, is used.

Definition: The upper-tail critical value $F_{\alpha;df_1,df_2}$ is the value for which a random variable $F$ with numerator degrees of freedom $df_1$ and denominator degrees of freedom $df_2$ has upper-tail probability $\alpha$: $$ P\left(F \geq F_{\alpha;df_1,df_2}\right)=\alpha. $$ For a one-way ANOVA, compare the calculated statistic with this upper-tail critical value. Reject $H_0$ if $F_{\text{calculated}}\geq F_{\alpha;df_1,df_2}$; otherwise, do not reject $H_0$.

F tables commonly organize entries by significance level and degrees of freedom, but the layout varies. Some have separate blocks for $\alpha$, some put numerator degrees of freedom across the top, and others reverse the row and column positions. Read the table headings rather than assuming an orientation. In the convention used here, the first degree of freedom is the numerator degree of freedom and the second is the denominator degree of freedom.

For reference, this small extract shows approximate upper-tail critical values. The entries are rounded, so a printed table may differ slightly in its final digit.

$df_1$	$df_2$	$\alpha=0.05$	$\alpha=0.01$
2	12	3.885	6.927
3	16	3.239	5.292

The significance level sets the probability of a Type I error for the test procedure under its assumptions. Choosing a smaller $\alpha$ makes the rejection threshold larger: stronger evidence against $H_0$ is required to reject. The F test is an upper-tail test because unusually large ratios of between-group to within-group variation count as evidence against equal means.

Why the Rejection Region Is in the Upper Tail

Under $H_0$, the group means are equal, so the between-group mean square and within-group mean square both estimate the same error variance. Their ratio is then expected to be around 1, though it varies from sample to sample. If the group means differ more than expected from within-group variation, $MS_{\text{between}}$ tends to be relatively large and the ratio $F$ rises. Values far out in the upper tail are therefore inconsistent with the equal-means model.

The critical value is chosen so that, when $H_0$ and the model assumptions hold, the probability of obtaining an F statistic at least that large is $\alpha$. Thus the table-based decision is:

Find the degrees of freedom.
Use $df_1=k-1$ for the between-groups numerator and $df_2=N-k$ for the within-groups denominator.

Choose the significance level.
Locate the table block or column for the specified $\alpha$, then find the entry for the correct pair of degrees of freedom.

Compare and decide.
If the calculated F statistic is at least the critical value, reject $H_0$. If it is below the critical value, do not reject $H_0$.

This rule follows directly from the definition of the critical value: the rejection region consists of values in the upper tail with probability $\alpha$. A statistic in that region has $F_{\text{calculated}}\geq F_{\alpha;df_1,df_2}$; a statistic outside it has $F_{\text{calculated}}<F_{\alpha;df_1,df_2}$. Equality is included in the rejection region.

Worked Table Comparisons

Worked Example: Rejecting at the 0.05 Level

Suppose a hypothetical comparison of three teaching formats uses 15 students, with five students in each group. An ANOVA summary gives $SS_{\text{between}}=34$ and $SS_{\text{within}}=48$. At $\alpha=0.05$, determine the table-based decision.

There are $k=3$ groups and $N=15$ observations, so $$ df_1=k-1=3-1=2,\qquad df_2=N-k=15-3=12. $$ The mean squares and statistic are $$ MS_{\text{between}}=\frac{34}{2}=17,\qquad MS_{\text{within}}=\frac{48}{12}=4,\qquad F_{\text{calculated}}=\frac{17}{4}=4.25. $$

In the $\alpha=0.05$ table block, use $df_1=2$ and $df_2=12$. The critical value is approximately $F_{0.05;2,12}=3.885$. Since $4.25>3.885$, reject $H_0$ at the 0.05 significance level. The observed ratio is in the rejection region for this test.

Worked Example: Not Rejecting at the 0.05 Level

A hypothetical ANOVA compares four fertilizer treatments using 20 plots. Its summary gives $SS_{\text{between}}=43.5$ and $SS_{\text{within}}=80$. Decide whether to reject equal population means at $\alpha=0.05$.

There are $k=4$ groups and $N=20$, giving $$ df_1=4-1=3,\qquad df_2=20-4=16. $$ The corresponding mean squares and statistic are $$ MS_{\text{between}}=\frac{43.5}{3}=14.5,\qquad MS_{\text{within}}=\frac{80}{16}=5,\qquad F_{\text{calculated}}=\frac{14.5}{5}=2.90. $$

The table entry for $\alpha=0.05$, $df_1=3$, and $df_2=16$ is approximately 3.239. Since $2.90<3.239$, do not reject $H_0$ at the 0.05 level. This decision means the statistic did not reach the chosen rejection threshold; it does not establish that the population means are equal.

Worked Example: Comparing Decisions at Two Significance Levels

A hypothetical study compares three storage methods. For 15 stored batches, the ANOVA summary gives $SS_{\text{between}}=20.4$ and $SS_{\text{within}}=24$. Determine the decisions at $\alpha=0.05$ and $\alpha=0.01$.

With $k=3$ groups and $N=15$, the degrees of freedom are $df_1=2$ and $df_2=12$. The mean squares and test statistic are $$ MS_{\text{between}}=\frac{20.4}{2}=10.2,\qquad MS_{\text{within}}=\frac{24}{12}=2,\qquad F_{\text{calculated}}=\frac{10.2}{2}=5.10. $$

For these degrees of freedom, the critical value is approximately 3.885 at $\alpha=0.05$, and 6.927 at $\alpha=0.01$. Since $5.10>3.885$, reject $H_0$ at the 0.05 level. Since $5.10<6.927$, do not reject $H_0$ at the 0.01 level. The statistic falls between the two critical values, so the table also indicates that the p-value is between 0.01 and 0.05.

Using a Table to Bound a P-Value

A critical-value table can often show a range for the p-value even when it cannot give the exact p-value. For a fixed pair of degrees of freedom, a larger upper-tail critical value corresponds to a smaller upper-tail probability. In the last example, the observed $F=5.10$ is larger than the 0.05 critical value but smaller than the 0.01 critical value. Therefore its upper-tail probability is less than 0.05 but greater than 0.01.

If the calculated statistic exceeds the largest critical value shown in the relevant table, report the corresponding bound, such as $p<0.01$, rather than claiming an exact p-value. If it is below a listed critical value, the comparison gives the appropriate lower bound. A table with only a few significance levels gives less detail than statistical software, but it still supports a valid decision at any listed level.

Some tables do not contain the exact denominator degrees of freedom needed. Do not silently use a value from a different row or column. Check the table’s instructions; if it specifies a rounding or conservative lookup convention, follow it and identify the approximation. When precision matters, use software to obtain the critical value or p-value for the exact degrees of freedom.

Common Mistakes

Using the wrong tail. The ANOVA F test uses the upper tail. A calculated statistic smaller than 1 is not automatically significant; compare it with the upper-tail critical value.
Reversing the degrees of freedom. The numerator uses $df_1=k-1$; the denominator uses $df_2=N-k$. Check which direction the table labels as numerator and denominator.
Using the wrong significance level. The critical value for $\alpha=0.01$ is larger than the one for $\alpha=0.05$ at the same degrees of freedom. Select the block matching the level specified for the test.
Treating “do not reject” as proof of equality. A statistic below the cutoff is not sufficient evidence to reject at that $\alpha$; it does not prove that the null hypothesis is true.
Rounding before comparing. Keep the calculated statistic at adequate precision until after the comparison. If it is very close to the critical value, use a more precise table entry or software rather than letting early rounding determine the decision.

Key takeaway: For a one-way ANOVA, select the upper-tail critical value using $df_1=k-1$, $df_2=N-k$, and the specified $\alpha$. Reject $H_0$ when the calculated $F$ is at least that value; otherwise, do not reject.

Check Your Understanding

Use the upper-tail F-table decision rule in each question.

A one-way ANOVA has five groups and 30 observations. What are the numerator and denominator degrees of freedom?
At fixed degrees of freedom, why is the critical value for $\alpha=0.01$ larger than the critical value for $\alpha=0.05$?
For $df_1=2$ and $df_2=12$, a calculated statistic is 3.70. Using the table extract, what is the decision at $\alpha=0.05$?
For $df_1=2$ and $df_2=12$, a calculated statistic is 7.10. What p-value bound can be read from the table extract?
Explain why a decision not to reject $H_0$ does not establish that all population means are equal.

\(df_1\)	\(df_2\)	\(\alpha=0.05\)	\(\alpha=0.01\)
2	12	3.885	6.927
3	16	3.239	5.292