The F-test for two variances lets you determine whether two populations have the same spread (variance). This matters because many statistical procedures, including pooled t-tests and ANOVA, assume equal variances across groups. If that assumption fails, your results from those procedures can be unreliable.

more resources to help you study

practice questions

F-Ratio for Variance Comparison

The F-ratio compares the variances of two independent samples drawn from normally distributed populations.

$F = \frac{s_1^2}{s_2^2}$

$s_1^2$ = sample variance of the first sample
$s_2^2$ = sample variance of the second sample

By convention, you place the larger sample variance in the numerator and the smaller in the denominator. This guarantees the F-ratio is always $\geq 1$ , which simplifies looking up critical values.

Each sample contributes its own degrees of freedom:

Numerator degrees of freedom: $df_1 = n_1 - 1$ , where $n_1$ is the sample size associated with the larger variance
Denominator degrees of freedom: $df_2 = n_2 - 1$ , where $n_2$ is the sample size associated with the smaller variance

Be careful here: $df_1$ always belongs to whichever sample you placed in the numerator, not necessarily "sample 1" from the problem statement. Mixing these up is a common mistake.

Interpretation of F-Ratio Results

The hypotheses for this test are:

Null hypothesis $(H_0)$ : $\sigma_1^2 = \sigma_2^2$ (the population variances are equal, sometimes called homogeneity of variances)
Alternative hypothesis $(H_a)$ : $\sigma_1^2 \neq \sigma_2^2$ (the population variances are not equal)

An F-ratio close to 1 means the two sample variances are similar, which supports $H_0$ . An F-ratio much larger than 1 suggests the variances differ meaningfully.

Decision rule:

Choose a significance level (typically $\alpha = 0.05$ ).
Find the critical F-value from an F-distribution table using $df_1$ and $df_2$ .
If your calculated F-ratio > critical F-value, reject $H_0$ and conclude the population variances are not equal.
If your calculated F-ratio $\leq$ critical F-value, fail to reject $H_0$ . You don't have enough evidence to say the variances differ.

Note that when you always place the larger variance in the numerator, you're effectively running a one-tailed test in the right tail. If the research question calls for a true two-tailed test (testing whether either variance could be larger), you need to halve $\alpha$ before looking up the critical value, or use the p-value approach and compare to $\alpha/2$ .

F-ratio for variance comparison, Introduction to Statistics Using Google Sheets

Appropriateness of the F-Test

The F-test for equality of two variances requires three conditions:

The two samples are independent of each other
Both populations are normally distributed
Sample sizes are relatively small (generally $n < 30$ )

The biggest limitation of this test is that it's very sensitive to non-normality. Even mild skewness or heavy tails can inflate the Type I error rate, especially with small samples. This is different from the t-test, which is fairly robust to non-normality. With the variance F-test, the normality assumption really matters.

If you suspect non-normality, consider alternatives like Levene's test or the Brown-Forsythe test, both of which are more resistant to departures from normality.

For larger samples ( $n > 30$ ), the sampling distribution of variances becomes more stable, making the F-test somewhat more robust. Still, you should check normality before applying the test using:

Graphical methods: histograms, Q-Q plots
Formal tests: Shapiro-Wilk test

Statistical Errors and Power

Two types of errors can occur with any hypothesis test, including this one:

Type I error: Rejecting $H_0$ when the population variances actually are equal. The probability of this equals your significance level $\alpha$ .
Type II error: Failing to reject $H_0$ when the population variances actually differ. The probability of this is denoted $\beta$ .

Statistical power $(1 - \beta)$ is the probability of correctly rejecting a false $H_0$ . Power increases when:

Sample sizes are larger
The true difference between the variances (effect size) is larger
You use a higher significance level $\alpha$ (though this also raises Type I error risk)

For the variance F-test specifically, power tends to be low unless the difference in variances is quite large or sample sizes are substantial. This is worth keeping in mind: a "fail to reject" result doesn't necessarily mean the variances are equal; you may simply lack the power to detect a real difference.