SOP: Multiple Sample Difference Test

GIFSQ — Global Institute for Food Safety & Quality

Principle

More than one test variable is evaluated per session, with reduced reliability compared with the two-sample test. Each panelist is served 3–6 samples depending on the number of test variables. One sample is a known standard; the panelist compares each coded sample with the known standard. One coded sample is a duplicate of the standard. The score the panelist assigns to the blind standard is subtracted from the score assigned to the test variables (or the results are analysed without this adjustment). The panelist is not to guess. Both the direction and the degree of difference are judged.

Apparatus required

• Individual judging booths (identical, uniformly illuminated), each fitted with drinking water, a glass, clean towels and a basin or receptacle for expectoration
• Coloured lighting (where needed to mask small colour differences between samples)
• Stainless steel, glass or china serving dishes and cups; plain serving trays; small stainless steel forks and spoons
• Prepared evaluation cards for the multiple sample difference test

Sample preparation

Prepare the standard (R) and all test samples in exactly the same manner. Include one coded duplicate of the standard among the test samples. Code every test sample with a three- or five-digit number. Serve 3–6 samples per session (trained panel: 6–25 judges; semi-trained: 10–25 judges).

Procedure

Give the panelist the standard or reference sample marked R; ask the panelist to taste it carefully for the quality characteristics to be evaluated

↓

Present the coded samples (including the blind duplicate of the standard)

↓

For each coded sample, rate the degree of difference from the standard: 0 none, 1 slight, 2 moderate, 3 large

↓

Rate the direction of quality: E equal, I inferior, S superior

↓

Comment separately on odour and on flavour for each sample

Interpretation of results

Assign numerical values to the ratings (in the book’s worked example, no difference from the standard = 5; large superior to R = 1; large inferior to R = 9). Analyse the results by the analysis of variance (Tables 19-6, 19-7, 19-8 and 19-9), and examine differences between samples with Duncan’s Multiple Range Test and the Dunnett Test.

Duncan’s Multiple Range Test: compute the standard error of the sample mean SE = √(mean square of error / N), where N is the number of panelists. Calculate the shortest significant range Rp = p × SE, where p is read from Table 19-9 at the chosen significance level for the error degrees of freedom. Arrange the sample means in ascending order; compare the range between pairs of means with the Rp for their relative position. Begin with the largest against the smallest, then the largest against the second smallest, and so on; underline means in a grouping to indicate they are not significantly different. Means within a non-significant grouping need not be tested further.

Dunnett Test (comparison of treatments against a pre-determined control): when the number of observations N is equal for all samples,

A = t × s × √(2 / N)

• A = the amount of the difference between a sample mean and the control mean required for significance
• t = value from Table 19-12
• s = standard deviation (square root of the average variance) of the p+1 distributions
• p = number of samples excluding the reference sample
• N = number of observations on which each mean is based
• degrees of freedom df = (p + 1)(N − 1)

Worked illustration from the book: a tomato ketchup manufacturer compared ketchup made with spices (R, reference 0) against spice oil A (1), spice oil B (2), oleoresin A (3) and oleoresin B (4), with 10 panelists. The analysis of variance gave: samples — df 4, SS 144.12, MS 36.03, F = 35.78 (significant at the 1% level; tabulated 2.64 at 5% and 3.91 at 1%); panelists — F = 1.448 (no significant difference); error — df 36, MS 1.007. Duncan’s test: SE = √(1.007/10) = 0.317; Rp at the 5% level = 0.91, 0.96, 0.99, 1.01. Mean scores: 4 (2.8), 3 (4.9), 0 (5.0), 2 (6.1), 1 (8.0). Sample 3 (oleoresin A) does not differ significantly from the reference; sample 4 (oleoresin B) is significantly superior to the reference and all others; samples 1 and 2 differ significantly from the reference and are inferior. Dunnett test (samples 0, 3, 4): average variance s² = 1.13, s = 1.0646, df = 27; t (Table 19-12) = 2.00 at 5% and 2.745 at 1%; A = 0.95 at 5% and 1.31 at 1%. Mean differences: 0 − 3 = 0.1 (not significant); 0 − 4 = 2.2 (significant). Sample 4 differs significantly from the reference; sample 3 is comparable to the reference.

Specimen evaluation card fields: instructions; rating scale (degree: 0 none / 1 slight / 2 moderate / 3 large; direction: E equal / I inferior / S superior); table of sample code with degree, direction and comments for odour and for flavour; signature.

Precautions

• The panelist must not guess; the blind duplicate standard checks guessing.
• Conduct judging in individual booths so that judgements are independent; allow no communication between panel members except consultation with the panel leader on any point of doubt.
• Test in the hour after a normal meal (e.g., 10.00 a.m. to 12 noon, or 3 to 5 p.m.).
• Panel members should not smoke or chew pan or supari for at least half an hour before the test.
• Judge quickly but not hurriedly; make odour observations by sniffing before tasting.
• Allow retesting, but avoid prolonged testing; tasters may swallow or spit out samples, and should rinse between samples (or use bread or puffed rice) to remove a strong flavour.
• Serve 3–6 samples per session; more than one test variable per session is evaluated with reduced reliability.
• Keep the evaluation card simple, clearly typed or printed, with unambiguous terms and directions in the desired sequence of action.

References

Ranganna, S., Handbook of Analysis and Quality Control for Fruit and Vegetable Products, 2nd edition.

Larmond, E., Methods for Sensory Evaluation, Publication 1284, Canada Department of Agriculture, Ottawa — cited in the book.