You are on page 1of 18

Chapter 9

The t- and F-distributions


Small sample size and possible unknown
variance
• Assume the random variable is normally
distributed (in the population)
• For small sample size, sample variance s can
vary a lot. (unknown variance).
• s is now also a random variable
• Sample size n < 30 (otherwise Central Limit
Theorem)
• (n - 1 degrees of freedom)
Known variance
Procedure
• Variance known?
– Yes – proceed with normal distribution
– No
• One sample?
– Yes – proceed with t-statistic
– No
» Test sample variances for equality (F-test)
• They are equal – Pooled variance formula
• Unequal – adjusted degrees of freedom formula
TESTING VARIANCES FOR EQUALITY (P211)

• When doing a t-test for comparing the mean


of two samples, we assume the variances to
be equal.
• Actually, we need to test to make sure.
• There is no easy sampling distribution for .
• We do have a sampling distribution for the
test statistic
• Called the F-distribution.
F-DISTRIBUTION
• Also called sometimes the Fisher distribution, after Ronald
Fisher, but care should be taken as there are a number of
distributions that has the name Fisher associated.
• Formula for the F-distribution is quite complex.
• F has two parameters, related to two degrees of freedom.
F-distribution
GENERAL COMMENTS
• Variances are positive, hence F-statistic is
always positive
• If , then we expect , and F ≈ 1
• Hence, we accept H0 when F is close to 1
• Reject H0 when F is close to 0 or much larger
than 1.
• Consider , based on sample sizes n1 and n2.
• A degree of freedom each for the numerator
and the denominator.
• Symbol : Fa,b
• Note the reciprocal property:
EXAMPLE 14A (page 212)

• Test statistic: (put the larger in the numerator,


above)
• Degrees of freedom (n2 - 1, n2 - 1)
• n1 = 6, n2 = 5, degrees of freedom (5,4).
• Sampling distribution is F5,4
EXAMPLE 14A (page 212)
• Two-sided test at 5% level.
• Need to find upper and lower 2.5% points of
F5,4.
• Use the reciprocal property
• Only need (from F-table, p330)
Find
EXAMPLE 14A (page 212-213)
EXAMPLE 14A (page 212-213)
• Observed F-value:

• Do not reject null hypothesis. Assumption of


equal variances holds.
• Always swops the variances such that F > 1 (by
taking the reciprocal).
Let’s recap….
Two sample test with equal variances
• We’ve done the F-test and accepted H0:
• Assume both populations has the same variance
• Improve the variance estimate with the pooled variance:

• S2 has (n1 – 1) + (n2 – 1) = n1+ n2 – 2 degrees of freedom


• Two-sample t-statistic:

• Example 11A, p208


Two sample test with unequal variances
• We’ve done the F-test and accepted H0:
• Assume both populations has dissimilar variance
• Test statistic has approximately the t-distribution, with
degree of freedom given by

• Two-sample t-statistic:

• Example 17B, p215

You might also like