Increasing the Efficiency of Consumer Tests Using Robust Statistical Procedures

• Activity Number: 323
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Consulting
• Author(s): Jason Parcon*
• Companies: PepsiCo
• Keywords: Least Squares ; Robust Regression ; Consumer Studies ; Experimental Designs

Abstract

An important step in product development (PD) is to get the pulse of the consumer to determine consumer-preferred products. Consumer data are typically modeled using the least-squares (LS) method, which is known to be inefficient when data deviate from normality or contain outliers. Unfortunately, most consumer data hardly fit the normal distribution and have numerous outliers, which may put into question decisions reached regarding consumer preference. To (partially) compensate for these hurdles, consumer studies utilize a large number of respondents, making them expensive.

This presentation will discuss the use of Wilcoxon (WIL) procedures for analyzing consumer data as alternative to LS. WIL is known to be more efficient than LS if data distribution is not normal and robust against outliers. Real (coded) data sets will be used to compare LS and WIL results for studies whose goals are to (1) select from a best set of treatments (ANOVA, multiple comparisons) and (2) optimize ingredient levels to maximize consumer acceptability (response surface experiments). The impact of differences in LS and WIL results on a consumer test's structure and PD decision-making will be addressed.