「MaxDiff計画」を用いたキャラメルポップコーンのおいしさ調査

[#1]

"Popcorn drizzled in caramel is so delicious!"

I still remember the excitement I felt when I tried caramel popcorn for the first time. I don't usually eat popcorn, but I enjoy caramel popcorn.

When you go to the supermarket, you'll find a wide variety of caramel popcorn on sale. After trying a few, you'll find that each has its own unique characteristics, such as some that are delicious, some that have a strong caramel flavor, and some that have a more subdued flavor.

Just around that time, a new function called "MaxDiff Design" was added to JMP. I thought, "This could be used to investigate the deliciousness of caramel popcorn!" So I asked the staff at my company (JMP Japan) for help and conducted a survey. Although it was a while ago, I would like to report the results of that survey.

What is MaxDiff design?

Conjoint designs are often used in consumer surveys, which are a survey method in which people are asked to choose the best product from multiple products.

On the other hand, the "MaxDiff design" asks you to choose the "best" and "worst" of multiple products presented to you. For example, if you are presented with three types of cake, you choose the cake you like best and the cake you like least.

MaxDiff designs provide more detailed information about product preferences than choice model designs because they include “best choice” information.

Taste survey using MaxDiff design

Evaluators : 9 employees of our company

Target products : 8 types of caramel popcorn (products A to H)

Below is a summary of the eight products. These are entered into a JMP data table, which we will refer to as the "research table" from here on.

*Product photos are intentionally blurred.

The survey assessed participants using the following procedure (see questionnaire below).

1. Each evaluator is presented with three of the eight products (set 1) and asked to taste each one.
2. Among those three,
   • Choose the one you thought was the tastiest and had the strongest caramel flavor.
   • The one that was felt to be the least tasty and the one that was felt to have the weakest caramel flavor
     Select.
3. Evaluate again with a different combination (set 2).
4. Repeat this until set 5.

Questionnaire

Each assessor was also asked to indicate how often they usually eat sweet snacks (every day, 5-6 days a week, 3-4 days a week, 1-2 days a week, rarely (less than 1 day a week), never).

In the above example, three products are selected from eight products, but if they are selected randomly, an imbalance will occur, with some products being evaluated multiple times. Therefore, it is necessary to use the concept of experimental design to design a combination that allows all products to be compared in a balanced manner.

This type of design allows us to more accurately measure and compare the importance of each product, so we use a MaxDiff design in JMP to have participants choose between different product combinations.

Creating a MaxDiff Design in JMP

To create a MaxDiff design in JMP:

1. With a study table open, select Design of Experiments (DOE) > Consumer Studies > MaxDiff Design.
2. As shown in the figure below, specify "Product Name" as [X, Explanatory Variable], then in "Plan Options", specify the number of profiles per choice as 3 and the number of choice sets as 5.
3. Click the Create Plan button.

Number of profiles per choice set : The number of products in one choice set. In this example, we are presenting three products at a time, so we enter "3" here.

Number of Choice Sets : The total number of choice sets to be presented to the respondent for selection. In this example, five choices are presented to each evaluator, so "5" is entered.

This will output a lookup table for MaxDiff like this:

This table means that rater 1 will rate the products as follows:

Choice set	Product (left)	Commodity (Central)	商品(right)
1	D	G	H
2	F	B	C
3	A	E	F
4	G	A	B
5	C	D	E

In the column "Choice" enter the following values ​​as the outcome of the evaluation:

Most preferred: 1, least preferred: -1, others: 0

In this survey, we are evaluating "taste" and "caramel flavor," so we create two columns and enter values ​​as shown below. The results of evaluator 2's evaluation of taste and caramel flavor are highlighted in red. (Evaluator 2 uses a different design table from Evaluator 1.)

Results of the MaxDiff analysis

The MaxDiff model can be fitted by running the script "MaxDiff" in the MaxDiff investigation table.

Below are the results for "taste" and "caramel flavor."

MaxDiff Results : Marginal utility and marginal probabilities for each product are displayed. The higher the marginal utility, the more favorable the product is. The marginal probabilities are calculated from the marginal utility and indicate the probability that the product will be chosen by the consumer.

Likelihood ratio test : An index for testing the effect of factors. A smaller p-value indicates that preferences differ significantly depending on the product.

Taste and caramel flavor

Taste : The products were rated in the order of E, F, and D, but the marginal probabilities did not show any significant differences. Since the p-value of the likelihood ratio test was 0.5100, it cannot be said that there was a significant difference in taste between the products.

Caramel flavor : The results were rated in the order of F, A, H. The p-value was 0.1028, which means there is a difference from the taste results. However, with a significance level of 0.05, it cannot be said that the caramel flavor differs significantly between products.

From these results, no clear differences in taste or caramel flavor could be found among the eight types of caramel popcorn. The writer tried them all himself, and found them all to be equally delicious.

However, two interesting findings emerged during the analysis:

Result 1. Differences in frequency of eating sweets and perception of caramel flavor

In this survey, we asked the evaluators how often they eat sweet snacks, and based on the results we divided them into two groups:

5 days or more a week ⇒ Many

4 days or less per week ⇒ Low

When the caramel flavor was analyzed including this frequency as a factor (subject effect), a significant effect of the factor × product interaction was shown (p-value < 0.05).

Looking at the line graph with product on the horizontal axis and utility value on the vertical axis, divided into low and high frequency, we can see that there is a difference in the utility value of the caramel flavor depending on the product between people who eat sweet snacks frequently and those who eat them infrequently.

Product C: People who often eat sweet snacks will not notice the caramel flavor as much, while people who don't eat sweet snacks very often will notice it strongly.

Product E: People who often eat sweet snacks will feel a strong caramel flavor, while people who don't eat many sweet snacks will feel a weaker caramel flavor.

These are interesting results that show that the perception of caramel flavor varies depending on how often you eat sweet snacks.

Result 2: The higher the price, the better the taste

From the above results, we were able to determine the utility values ​​(desirability) of "taste" and "caramel flavor" for each product, and then analyzed the relationship between these and the price per gram of the products.

We plotted the price of each product on the horizontal axis and the utility value on the vertical axis to calculate the correlation coefficient. The correlation coefficient for deliciousness was found to be relatively high, at approximately 0.7.

*Product H has been excluded from this analysis because it has a much larger content volume than the other products.

These results confirmed the plausible trend that the more expensive the caramel popcorn, the tastier it is.

by Naohiro Masukawa (JMP Japan)

Naohiro Masukawa - JMP User Community