A Critical Comparison of the Two‐sample and Triangular Binomial Designs

Abstract

The relative sensitivities of the two-sample and triangular designs, in cases where quality may be stated unambiguously, have been much discussed. From a classical standpoint, using a binomial mathematical model, the triangular technique should be more efficient than the two-sample (9). This arises from the differences in the chance probability for each design.The two-sample comparison consists in the presentation of two samples (different in intensity of quality) to the taster, who is required to denote one as being higher in intensity. The chance probability for this occurrence to he correct is one-half. The triangular test comprises three samples, two identical and one different in intensity of taste quality. The taster is required to determine the odd sample. By chance, this probability is one-third.When the tests are given in the above manner, actual experimental results have shown that the triangular design is not as sensitive as the twosample method in the determination of thresholds (2,4). This phenomenon has been explained as being due, to a large extent, to the inherent limitations of flavor perception ( 2 ) . However, this view is not unanimous. Protracted tasting is not necessarily the cause for fatigue or sensory adaptation (11). An attempt has been made to increase the efficiency of the triangular design by including the partially correct judgments in tests of significance (3).Using these designs, the levels of significance for thresholds have been established by employing only the probability of committing an error of the first kind ; i.e., the chance that the hypothesis of no real difference may be rejected when it is true. However, it has been emphasized that a consideration of the error of the second kind (the chance that a taster may overlook a real difference when it exists) is of at legst as great importance as that of the error of the first kind (1).It is the purpose of this paper to demonstrate, by experiment, inherent differences existing between the two-sample and the triangular designs.

Three-stimulus procedures in olfactory psychophysics: An experimental comparison of Thurstone-Ura and three-alternative forced-choice models of signal detection theory

“…For the triangle test, these include testing procedures such as the binomial test (e.g., Lockhart, 1951) and estimation of a dose-response relation using a model (Bock & Jones, 1968;Filipello, 1956) that was recently named the triangular constant method model (Frijters, 1979).…”

The evolution of a model: A review of Thurstonian and conditional stimulus effects on difference testing

“…This was shown for a variety of stimuli (Dessirier, Seiffermann, & O'Mahony, 1999;Filipello, 1956;Frijters, Blauw, & Vermaat, 1982;Masuoka et al, 1995;O'Mahony & Goldstein, 1987;O'Mahony & Odbert, 1985;Stillman, 1993;Tedja et al, 1994;Wasserman & Talley, 1969). The two distribution, equal variance model, as described above, cannot account for this satisfactorily.…”

Paired‐comparison and Triangle Sensory Methods Compared for Use in Product Improvement

“…However, in the triangle test, they were asked first to determine difference and then intensity of the samples. Filipello (1956) also showed that the triangle design was not as sensitive as the paired-comparison, but this was in the testing of thresholds of simple taste solutions and not complex food systems. Gridgeman (1970) showed that pairedcomparison results more sharply distinguished preferential differences between two treatments, but compared this to a more psychologically complex two-stage triangle test.…”