F (Variance Ratio) Test

 

Menu location: Analysis_Parametric_F (Variance Ratio).

 

This function tests the equality of the variances of two random samples from a normal distribution.

 

F is the ratio of variances (largest as numerator) from samples of size n1 and n2. Degrees of freedom are n1-1 and n2-1 corresponding to the numerator and denominator sample variances.

 

Only the upper tail probability need be considered because the larger variance is always used as the numerator in the variance ratio F (Altman, 1991; Armitage and Berry, 1994). For a two sided test the smaller of the two tail probabilities is doubled; with the larger variance on top that is usually the upper tail, but not always when the degrees of freedom differ. Analysis of variance can utilise a one sided probability because the numerator and denominator of the variance ratio are predetermined.

 

Example

Test workbook (Parametric worksheet: Slight/no symptoms, Marked symptoms).

 

Consider the following scores, on a scale of 0 to 100, from nine patients with slight or no symptoms and seven patients with marked symptoms (an illustration rather than a real study). The question is whether the scores vary more in one group than in the other.

Slight/no symptoms Marked symptoms
34 5
45 8
49 18
55 24
58 60
59 84
60 96
62  
86  

 

To analyse these data in StatsDirect open the test workbook using the file open function of the file menu. Then select F (Variance Ratio) from the Parametric section of the Analysis menu. Select the columns marked "Slight/no symptoms" and "Marked symptoms" when prompted for data.

 

For this example:

 

Variance ratio/F test

 

Variable Name DF (n-1) Variance
Marked symptoms 6 1404.809524
Slight/no symptoms 8 202.277778

 

F = 6.944952

 

Upper side P = 0.0077

Two sided P = 0.0153

 

The variance of the scores is nearly seven times larger in the patients with marked symptoms, and the two sided test is significant at the 5% level. The equal variances assumed by the ordinary unpaired t test would be doubtful for these data, so the unequal variances form of that test, or a nonparametric method, would be safer.

 

 

P values

confidence intervals