Problems with a Type of Matched Subjects Design and Analysis

• Activity Number: 255
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Steve Verrill* and David Kretschmann
• Companies: USDA Forest Products Laboratory and USDA Forest Products Laboratory
• Keywords: predictor sort ; matched subjects ; blocking by a concomitant variable ; analysis of variance ; analysis of covariance ; artificial pairing

Abstract

There is a type of blocked experiment that has the potential of being poorly designed and/or analyzed. Verrill et al (1993, 1999, 2004) referred to such an experiment as a "predictor sort" experiment. David and Gunnink (1997) spoke of "artificial pairing." In text books, it is sometimes referred to as a "matched pair" or "matched subjects" design. The associated design process is also sometimes described as "forming blocks via a concomitant variable." Improperly designed and/or analyzed, predictor sort experiments can be associated with incorrect/inadequate power calculations and sample sizes, incorrect tests of hypotheses, and incorrect confidence intervals. In this paper, we review some of the results in the literature, add a section on multiple comparisons, and present results from power and confidence interval coverage simulations that emphasize the importance of proper design and analysis of predictor sort experiments.