|
Activity Number:
|
283
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Education
|
| Abstract - #303467 |
|
Title:
|
Preliminary Variance Tests for Deciding to Use Pooled vs. Nonpooled t-Tests
|
|
Author(s):
|
Jared L. Martin* and Lewis H. Shoemaker*+
|
|
Companies:
|
Millersville University of Pennsylvania
|
|
Address:
|
Department of Mathematics , Millersville, PA, 17551,
|
|
Keywords:
|
two sample t-test ; preliminary F-test ; unequal variances
|
|
Abstract:
|
Detecting differences between two population means is commonly achieved using a pooled t-test. This test is based on the assumption that the populations are normally distributed with equal variances. When the populations have unequal variances, the alternative is the non-pooled t-test. The problem is to decide which method to use when given data with unknown variances especially when sample sizes are not equal. Doing a preliminary F-test for equal variances is a possible solution. Unfortunately, the F test is unreliable when deviating from the normal assumption. A possible alternative is to use a robust test for equal variances before proceeding with the appropriate t-test. This paper investigates four possible approaches to the problem of which t-test to use. Level and power are investigated when sampling from various distributions and for equal and unequal variances.
|
