Hello everyone. Last time, I wrote about “Regression Analysis” which was introduced as one of the analysis methods when conducting primary research. This time, I would like to write about another analysis methods called “Conjoint Analysis”. This is the last of the four analysis methods (customer journey mapping, Kano model, multiple regression analysis, conjoint analysis) that I picked up as “relatively useful analysis methods” in my previous post.

**1. What is “Conjoint Analysis”?**

Conjoint analysis is an analysis method that is often used when developing concepts for products and services (although it can be used for various purposes other than those). As it is called “Design Of Experiment in the marketing field (I would like to write about Design Of Experiment in this blog later)” etc. , it is possible to quantitatively analyze which attributes (e.g. : PC weight, battery life, etc.) affect on consumer buying intention and customer satisfaction.

**2. How to use and view “Conjoint Analysis”**

There are a lot of specific ways to proceed with the analysis on the internet, so let me post only one link here (short video. about 8min).

If you roughly divide the process of conjoint analysis into the “survey design” stage and the “analysis” stage, I think that the **point of the survey design stage is “orthogonal table”** and the **point of the analysis stage is “multiple regression analysis”**.

**1) The point of the survey design stage – Orthogonal table**

In the survey design stage of conjoint analysis, a “conjoint card” is created. It is the **combination of attributes and levels** that make up a product or service. In the example below, **“weight”, “battery life”, “HDD capacity”, and “price” are attributes**. And, in terms of “weight”, the **levels are “1kg”, “1.5kg”, and “2kg”**.

One card is for one concept, and consumers are asked to evaluate each concept in a questionnaire survey. There are two major methods of evaluation: A) rank evaluation and B) score evaluation. A) The rank evaluation is a method in which consumers give consecutive numbers in order from the concepts that they find attractive (18 concepts, from 1st to 18th). B) In score evaluation, scores are evaluated on a scale of 3, 5, or 10 in the form of so-called customer satisfaction. A) The rank evaluation is more precise, but it has the disadvantage that the more concepts there are, the greater the loads on the respondents because they are asked to list all the concepts. B) Score evaluation is more respondent friendly. The smaller the scale, the easier it is to answer. Research should be designed with this in mind.

When making this conjoint card, “orthogonal table” is useful. As you increase the attributes and levels of products and services (eg, PC weight => 1kg, 2kg, etc.), the number of combinations becomes larger and the number of conjoint cards becomes enormous. Orthogonal table is excellent because they dramatically reduce the number of combinations. There are many explanations of the orthogonal table itself on the Internet, so let me put only one link here.

Although various types of orthogonal tables are available depending on the number of attributes and levels, as I wrote earlier, it is not practical if the number increases too much. **L4 (2 levels x 3 attributes), L8 (2 levels x 7 attributes), L9 (3 levels x 4 attributes), L18 (2 levels x 1 attributes, 3 levels x 7 attributes) are more realistic**. It is also said that L18 is the most commonly used as it is the combination of 2 levels and 3 levels and 18 is appropriate size. But I think 18 conjoint cards is enough heavy load! We can say this is the limit normally.

Attributes and levels of products and services are examined and assigned to this orthogonal table (**conversely, I think it is also necessary to generate ideas with the orthogonal table in mind**). By the way, it is okay to leave some attributes (columns) of the orthogonal table blank. You don’t have to force it to fill in all. Also, when assigning orthogonal table, the concept of “Latin square” and “interaction effects” comes up, but in practice it’s okay to put it aside for the time being (if you want to dig deeper, please google it).

**2) The points of the analysis stage – Regression Analysis**

If you use a conjoint card to conduct a questionnaire survey, I think that the consumer’s evaluation (the average value of the evaluation of that card) will come out for each card.

We will analyze this data, but the analysis logic itself of conjoint analysis is actually the same as multiple regression analysis. So it is nice to be able to analyze it on Excel as well! (Although it would be easier to use statistical tools such as SPSS)

In the analysis, the level of each attribute is replaced with “1” and “0”. At this time, if 1kg is “1” and 1.5kg is “0”, 2kg is inevitably “0”. Eliminate such inevitable columns. Any column in the attribute will do, but in the image above I’ve removed the yellow column.

Coefficients for each conjoint card attribute and level (impact on consumer evaluation, which is outcome variable) are output. Subtract the minimum value from the maximum coefficient value for each attribute to get the “impacts (also called utilities)”.

**3. How to utilize the results of Conjoint Analysis**

There are two ways to utilize the results of multiple regression analysis: **a) absolute analysis of utilities** and **b) creation of prediction formulas**.

**a) The absolute analysis of utilities** is simply a graphical comparison of the “utilities” calculated above. It would be nice to sort them in descending order (I would like to write about how to select a graph later on this blog).

**b)** Since **the prediction formula** is created by multiple regression analysis, **just apply it to the form of “Y=ax+bx+cx…+d”**. Add the maximum value of each attribute in Fig3 above.

**Satisfaction = 1.33 + 0.07 + 0.00 + 0.73 + 2.77 (intercept) = 4.9**

The questionnaire survey in this example was a 5th scale evaluation, so we can expect a high score of 4.9 out of 5.

That’s all for this time, and I would like to continue from the next time onwards. Thank you for reading until the end.