PubAg

Main content area

Paired Preference Tests: A signal detection based analysis with separate d′ values for segmentation

Author:
Zhang, Xiaotian, Halim, Jeremia, Wichchukit, Sukanya, O'Mahony, Michael, Hautus, Michael J.
Source:
Journal of sensory studies 2016 v.31 no.6 pp. 481-491
ISSN:
0887-8250
Subject:
consumer acceptance, models, placebos, potato chips, sensory properties, statistical analysis
Abstract:
Consumers gave graded preference responses to potato chips in a paired preference test. The graded responses were given to both the target pair under consideration and putatively identical “placebo” pairs of chips. From these data, a novel Signal Detection analysis was used. A model was developed giving a “magnitude of preference” distribution, for those consumers who preferred the first type of chip in the target pair and a second distribution for those who preferred the second type of chip. A second pair of distributions was generated for the two placebo pairs that had also been presented to the consumers. Using a signal detection paradigm, a value of d′ was computed for each chip, representing the difference between the preference distribution for the target pair (Signal + Noise) and its corresponding placebo pair (Noise). The analysis has the advantage that these dʹ values are not distorted by the responses of those consumers who had reported preferences for the placebo pair. This advantage is not a feature of the regular computation of d′ values based on the 2‐AC test. PRACTICAL APPLICATIONS: Paired preference tests are an important part of the measurement of consumer acceptance. Unfortunately, they are prone to response bias whereby consumers report preferences for putatively identical products; they are responding to the test conditions rather than the sensory properties of the products under assessment. Using identical products as a control, the effects of this on the results for the two different products to be assessed can be analyzed. There are various statistical approaches and models for solving this problem. This paper introduces an improved form of analysis based on signal detection/Thurstonian modeling. The method provides more meaningful information regarding segmentation.
Agid:
5719701