Date and Time:  Wednesday, February 11, 1998, 4:10 p.m.

Place:  321 Snedecor Hall

Speaker:     Thomas Loughin,
                       Kansas State University

Title:  Testing Independence with Correlated Binary Data

This talk will present two seemingly unrelated research problems which have arisen through consulting.

The first deals with the problem of performing tests for independence in contingency tables with multiple column responses. This situation arises often, for example, in surveys in which respondents are allowed to mark more than one response to a question with several possible response categories (``What kind of car(s) do you own?'' --- many people own more than one car.) Each unit may therefore provide more than one response to a crosstabulation of these response frequencies against some categorical grouping, such as gender.

The second problem is one of controlling the error rate for a large number of correlated tests. ``Quantitative Trait Loci'' (QTL) are areas on a genome which are related to a given trait observed in the specimen. For example, in alfalfa plants the trait may be resistance or susceptibility to a disease. A large number, perhaps hundreds, of binary genetic markers, indicating presence or absence of proteins at different locations on the gene, can be observed on each plant. It is of interest to determine which of these markers may be associated with the binary trait.

It turns out that both problems can be formulated as tests of independence between elements of a vector of correlated binary variates and a fixed group identification. Solutions to both problems are presented which use the same basic set of tools: Pearson's chi-square test for contingency tables and a resampling procedure. It is shown that these two problems are direct categorical analogs to the ANOVA F-test and Tukey's method of multiple comparisons.
 

Coffee:  3:45 p.m., 104 Snedecor

Seminar schedules and abstracts are available via WWW:
   http://www.stat.iastate.edu