The F-Distribution and Comparing Two Variances

Lesson 42 of STA301 explores Snedecor's F-Distribution, named in honor of Sir Ronald A. Fisher, and its primary application in comparing the variances of two independent normal populations. The F-distribution is defined as the ratio of two independent Chi-Square random variables, each divided by its respective degrees of freedom: F = (chi1^2 / nu1) / (chi2^2 / nu2), where nu1 represents the numerator degrees of freedom and nu2 represents the denominator degrees of freedom. The lecture outlines the foundational mathematical properties of the F-distribution: it is continuous and defined exclusively for positive values (F > 0); its curve is non-symmetrical and skewed to the right; its exact shape depends entirely on the pair of degrees of freedom (nu1, nu2); its mean is nu2 / (nu2 - 2) for nu2 > 2; and it satisfies the Reciprocal Property: F_(1 - alpha, nu1, nu2) = 1 / F_(alpha, nu2, nu1). This reciprocal property allows researchers to determine lower critical values from standard tables that tabulate only upper-tail percentages. The primary inferential application is testing whether two independent normal populations possess equal variances: H0: sigma1^2 = sigma2^2 (or sigma1^2 / sigma2^2 = 1) against two-tailed H1: sigma1^2 != sigma2^2 or directional alternatives. Testing variance equality is essential not only for comparing product consistency in quality control, but also for validating the homoscedasticity assumption required before executing independent two-sample pooled t-tests. The test statistic is the variance ratio: F = s1^2 / s2^2, where s1^2 and s2^2 are the unbiased sample variances. Conventionally, the larger sample variance is placed in the numerator so the calculated F-ratio exceeds unity (F >= 1), comparing against upper-tail critical values with df1 = n1 - 1 and df2 = n2 - 1. The lecture also derives the 100*(1 - alpha)% Confidence Interval for the variance ratio sigma1^2 / sigma2^2. In Virtual University final term examinations, F-test calculations and reciprocal critical value determinations are high-yield questions. Revising sta301 handouts and practicing solved past papers guarantees success.