Have you thought about how your favourite ice cream brand manages to maintain the same taste, even with the changes in ingredients? Discrimination testing in sensory analysis plays an excellent role here. These tests are used in sensory analysis to determine if two samples are chemically different or are different to human senses. For example, an ice cream producer replaces an expensive vanilla flavour with a cheap flavour, and their consumer doesn’t notice the changes.

This fine art of ensuring consistency is done through discrimination testing, which saves costs and also maintains the consumer experience. This article provides a comprehensive guide on discrimination testing to understand how it plays a role in undetectable product changes.

## 1. Types of Discrimination Tests

The tests to perform discrimination testing are:

### 1.1 Paired comparison tests

In Paired comparison tests, trained panelists identify the sensory attributes of food and follow the guidelines on the score sheet. This test involves two randomized sequences, so panelists receive either samples A or B and ensure equal distribution. It is one tailored test because the panelist knows the extent of the sample attributes. The alternative hypothesis is drawn for the chosen sample with higher attributes (Sample A) and Sample B.

The paired comparison test identifies which sample has a high intensity of specific attributes to ensure target sensory differences between samples. Changes in one sensory characteristic affect the other. For example, reducing the sugar in sponge cake will not only reduce the sugar content but also change the texture of the cake and browning. In these types of products, a paired comparison test may not be compatible.

### 1.2 Triangle test

The triangle test is the technique for sensory evaluation and is used to determine whether the two products have differences in sensory attributes. This method involves analyzing three samples; two of them are the same, and one is different. These samples are prepared and presented to participants randomly. Then, they are asked to identify the difference in samples. This method is used in the food and beverage industries to check the difference between batches and formulations. It helps detect changes and ensure consistency in product development. It is a simple method and easy to understand.

### 1.3 Duo trio test

In the duo trio tests, panelists receive the three samples. One sample is marked as a reference and is the same formulated as the other two samples. Then, panelists select the sample that is similar to a reference sample. Null hypothesis and alternative hypothesis are drawn. A trained panelist performs the task with the guidelines. Duo-trio test allows the sensory panelist to identify if two samples are different but the direction of difference is unknown. It means it describes that samples are different but does not specify which attribute causes the difference.

There are two types of duo trio test:

### 1.4 A-not-A tests

There are two types of A-not-A-tests.

- Standard A-not-A-tests
- Alternate A-not-A-tests

An alternate test is not frequently used. It is a sequential difference paired test. The panelist receives and analyzes the two samples one by one, and then they are removed. Then, he is asked about any differences between samples. The alternate test depends on four randomized sequences. This is a one-tailed test, and the panelist is asked questions to determine whether the two samples are different or the same.

A standard test is used to determine the presence or absence of characteristics in a product.

### 1.5 Sorting Methods

The sorting method is used in sensory evaluation to organize the products based on their sensory attributes. This sorting method identifies the patterns and relations between product development and quality control. There are two types of sorting methods.

- The Harris Kalmus test
- The two-out-of-five test

The Harris Kalmus test is a discrimination test that determines individual thresholds for phenylthiocarbamide. Panelists receive an increasing concentration of PTC in eight samples (four containing water and the other four containing PTC concentration). Panelists sort the samples into four groups. The sorting continues until the panelists sort correctly. This is a shortened version of the test for the PTC and PROP (6-n propyl thiouracil) threshold used by Lawless.

In the two-out-of-five test, panelists receive five samples and sort them into two groups. One group contains two samples different from the other three. This test is used for odour threshold work where samples are weak. The probability of choosing the correct two samples is 0.1. This low probability of choosing the correct samples is an advantage. The possibility of sensory fatigue from this method is a disadvantage.

### 1.6 The ABX Discrimination Task Dual Standard Test

In the ABX discrimination test, the panelists identify the difference between two samples, which are represented as a treatment sample and a control sample. An additional sample, X, is introduced that can match either samples A or B. So, the panelists identify two reference samples that match the X sample.

## 2. Strengths and Weaknesses of Discrimination Test

- Sensory panelists should carefully consider the nature of the sample and evaluation goal. If the variation between batches is as large as the variation between formulations, then the panelist should not use the duo-trio test and triangle test. You can use the paired comparison test for this.
- The weakness of discrimination tests is that they don’t identify the magnitude of sensory differences between sample formulations. These tests identify the presence or absence of specific attributes but do not indicate the magnitude.
- The P values and statistics from tests identify that the test is detectable but does not detect the magnitude. The P value depends on the number of panelists and the difficulty of the test method, which doesn’t directly tell the magnitude of the attribute.
- A test with a correction probability of 95% is more likely to tell a significant difference than the tests with less than 50% correctness.
- There are scaling methods to estimate the size of the difference, such as Thurstonian scaling. These methods are complex and require specific assumptions for effective results.
- The strength of the discrimination tests is that the panellist performs tasks quite simply, and the panelist automatically grasps the task.
- Sensory panelists must be aware of the Replication, counterbalancing issues and power related to discrimination testing.

## 3. Data Analysis

Discrimination testing is important in sensory evaluation, in which a panelist evaluates the sensory attributes difference between samples. There are many statistical methods to identify the data, including chi-square, binomial, and normal distribution-based Z tests.

### 3.1 Binomial Distribution and Tables

The binomial distribution helps analyze the information from discrimination tests, especially when responses are binary. It establishes a fixed number of trials of success, making it ideal for sensory tests.

The panelist’s results can be termed a Bernoulli trial with a high success probability for correct differences. These trials are independent, with a constant success probability across trials. Cumulative binomial probability tables provide the probability of observing success rates in trials, which helps to determine the observed results.

For example, 20 panelists participated in a triangle test, and the probability of correctness is 1/3. By using binomial software and tables, specialists can calculate the probability of correctness in specific numbers and determine the deviation.

### 3.2 The Adjusted Chi-square test

The Chi-square test compares the frequencies of responses with the expected frequency under the null hypothesis. Use it for small sample sizes and low expected frequencies in sensory testing. For the single proportion test, the degree of freedom is 1—for example, 30 panelists in a duo trio test with a possibility of 1/2. Compare the expected frequencies with the Chi-square formula to determine the statistical result.

### 3.3 The Normal Distribution and Z-test on proportion

In discrimination testing, we use the normal distribution and Z tests for large sample sizes. The normal distribution test relies on the central limit theorem, which describes that the sampling distribution of a sample approaches the normal distribution with an increase in sample size. The normal distribution includes a large sample size. In discrimination tests, the responses are binary (correct or incorrect), making the sample size suitable for analysis.

We use Z tests to measure the proportion of correct responses in discrimination testing. They determine the observed proportion that deviates from the exact proportion under the null hypothesis.

The null hypothesis tells that the observed proportion of correct responses is equal to the expected proportion. However, the alternate hypothesis describes that the observed proportion of correct responses is different from the expected. The steps in the Z test are

- Determine the proportion of correct responses in the sample. Calculate this by dividing the number of correct responses by the total number of responses.
- Then, compare to the expected proportion.
- Estimate the deviations and errors from standard proportion and size.
- Calculate the z value that represents a number of standard deviations.
- Interpret the P value that indicates the probability of obtained results being as high as the observed one and assume the null hypothesis is true. A low P value of less than 0.5 rejects the null hypothesis.

## 4. Issues

### 4.1 The Power of Statistical Test

Various issues affect the reliability and accuracy of discrimination tests.

Statistical power is the probability that a test identifies the difference when one exists. High power means there is a low chance of a type 2 error (failing to determine a difference) occurring. There are the factors that affect the power can be:

**(i.) Sample size**

The power of the test increases with the large sample size, reducing the errors and making it easier to detect the differences.

**(ii.) Magnitude of difference **

Detecting significant differences between samples quickly can lead to an increase in power.

** (iii.) Variability**

Test power increases as the variability in the sample decreases. Sensory tests involve small samples due to logistical problems that reduce the test power. Sensory specialists must balance the limitations with the high power to detect the differences. High power in designing equipment helps to estimate the sample size.

### 4.2 Replication

Replication handling in discrimination has been an issue in sensory tests for many years. A number of people raised this issue, but very few provided solutions.

In Replication, panelists repeat the test under the same conditions to confirm the results and enhance the validity of results. It results in many challenges:

Repeating the test causes sensory fatigue and minimizes the specialist’s agility to differentiate the attributes. Researchers introduce many statistical approaches to handle the replication challenge in discrimination testing.

- A generalized linear model (GLM) helps to describe the probability of correct results while considering fixed and random effects.
- The beta-binomial model considers overdispersion in binomial data, which is common in test replication due to panelist variability.
- Brockhoff and Schlich introduced a solution to handle Replication in discrimination testing. The guideline states that you must balance the number of replicates and panelists to obtain valid results.
- Pool the data from multiple Replication for more power of test that increases viable results.
- Implement the models that adjust the number of observations based on variability.

### 4.3 Warm up Effect

Warm-up effects in the discrimination testing influence the accuracy of tests. During the initial stages of sensory testing, this factor affects the results. This factor compromises the ability to differentiate the sensory attributes between samples, resulting in inaccuracy. The issues listed below arise from the warm-up effect:

During the initial phase, the panelist may fail to differentiate the samples, leading to false negatives. Reduced sensitivity of the panelists in determining the samples affects the reliability and accuracy of the results. Warm-up effects can also cause biases in results, which involves accuracy.

Provide training sessions to panelists before the test begins to minimize these effects. Incorporate warm samples for panelists before the actual ones to reduce bias in results. Randomization and blind techniques can reduce the bias results. Regular monitoring of process performance identifies the changes and deviations and then allows the adjustment of the errors.

## 6. Common Mistakes made in Discrimination Testing

Many common mistakes can cause inaccuracy in tests. We can avoid these mistakes for valid and accurate results in discrimination testing.

**(i.) Inadequate Training **

If panelist are not provided proper Training on the testing methods and their sensory attributes, they will not identify the correct results. Comprehensive Training makes panelist familiar with the testing protocols and sensory characteristics.

**(ii.) Presenting Poor sample**

Presenting the sample in a non-randomized order fails to control variables like serving size, lighting, and temperature. Bias will affect the results, making it challenging to determine the differences. Control and randomize the standardized presentation of samples.

Tests conducted in inappropriate conditions or environments with distractions, odour, and inadequate lighting affect the sensory perception of panelists and cause inaccurate results. Conduct these tests in the lab with proper lighting and temperature or odor management.

**(iii.) Sample size**

Inadequate sample size affects statistical power. Small samples reduce statistical power, making it difficult to detect accurate differences. An adequate number of panelists and replications in tests increases the validity of results.

**(iv.) Panelist fatigue **

Time-taking tests that are too intense lead to panelist fatigue, and fatigue reduces the sensitivity of panelists towards the sensory characteristics. They should have breaks in the testing sessions and limit the test samples to avoid fatigue.

** (v.) Environmental factors **

Environmental factors like temperature changes and lighting affect sensory perceptions. Standardized and optimal environmental conditions lead to accurate and consistent results.

## 7. Conclusion

Discrimination testing helps us understand the article about factories maintaining the taste of their products. Paired comparison tests, triangle tests, duo-trio tests, and A-not-A tests represent different types of discrimination testing. The industry uses these tests in various applications, which vary with the product. We establish a sorting method to sort the samples based on their sensory characteristics. The Harris Kalmus and two out of five tests are the sorting methods that help determine the samples. Discrimination testing in sensory science has strengths that help to get more accurate data and weaknesses the inability of the test to identify the desired results.

Data analysis is an essential step in testing that involves using advanced methods or techniques. Binomial distribution and tables reduce the bias results in samples, and the chi-square test compares the frequencies of responses. Issues in the test are the power of statistical tests, as well as the replication and warm-up effects. Avoid common mistakes while conducting tests to ensure reliable results. The mistakes involved in this method are inadequate Training, poor presentation of samples, sample size and fatigue. Get more information related to sensory analysis at Grubiie.