Conjoint Analysis Report
SurveyMars now supports Conjoint Analysis model analysis for survey research.Conjoint analysis is a multivariate statistical analysis method used to study consumer product preference situations. For example, consumer preferences for mobile phone products, computer products, or automobile products. Conjoint analysis helps researchers understand which product attributes and levels are most important to consumers.
What is Conjoint Analysis
Conjoint analysis is used to study consumer product selection preferences. The analysis involves several key terms:
Attribute: Characteristics of a product, such as mobile phone CPU, phone size, camera pixel, etc.
Level: Specific values of attributes, such as CPU grade (high-end, mid-range, low-end), phone size (5 inches, 6 inches, 7 inches), camera pixel (12MP, 20MP, 30MP), etc.
Profile: Also called a product, formed by combining various attribute levels, such as a phone with "high-end CPU, 7 inches, 20MP camera".
The conjoint analysis process typically consists of 4 steps:
Step 1: Determine attributes and levels
- First, determine the required attributes and their corresponding level values
Step 2: Orthogonal experiment
- Based on the number of attribute levels, design to obtain potential "profiles" (products)
Step 3: Design questionnaire
- Design a questionnaire and collect data for each "profile" using "rating method/ranking method/choice method"
Step 4: Conjoint analysis
- Perform conjoint analysis after data collection
Creating Conjoint Analysis Questions
1. Navigate to the Surveymars platform to create a Conjoint Analysis question type.
2.Set up the conjoint analysis question and upload the attributes and levels to be researched. View a Conjoint Analysis example.
3. After designing the questionnaire and collecting responses, you can directly view the conjoint analysis report on the statistical analysis page.
Understanding Conjoint Analysis Theory
The data principle of conjoint analysis is to use each attribute as X and the utility scores (ratings) as Y for OLS regression. Since all attributes are categorical data, all attributes are processed as dummy variables, with the first level of each attribute as the reference level. After obtaining the regression coefficient values (regression coefficients are the utility values of levels), the utility value of the reference level is calculated (this value = 0 - sum of utility values of other levels). The larger the utility value of a level, the more important it is.
After obtaining the utility values for each attribute level, how is the importance of an attribute determined? The "maximum range" method is used: the maximum utility value of an attribute's levels minus the minimum utility value of that attribute's levels, and this range is the importance value of the attribute. After normalizing the importance values, the relative importance of each attribute is obtained.
This allows analysis of the relative importance of each attribute and the relative importance of each attribute level. The importance ranking of each profile can also be analyzed, and SPSSAU provides parameters to output profile utility values.
Interpreting Conjoint Analysis Results
1. Conjoint Analysis Results Summary:
- This table displays the importance values and percentages for each attribute
- It also lists the utility values for each level
2. Conjoint Analysis Estimation Results:
- This table displays the OLS regression model output results
- It provides Pearson correlation coefficient and Kendall correlation coefficient for model fit assessment
- The first level of each attribute is the reference level, so no regression coefficient is displayed, but the reference level's utility value can be calculated as: 0 - sum of utility values of other levels in the same attribute
Key Interpretation Points:
- Attribute Importance: Higher importance values indicate that the attribute has a greater impact on consumer preferences
- Level Utility Values: Positive utility values indicate preference for that level, while negative values indicate less preference
Data Format Requirements
The data format for conjoint analysis should be structured as follows:
- Each row represents one respondent's evaluation of one profile (product)
- For example, if one respondent evaluates 9 candidate products (profiles), and you have 100 respondents, the data will have 100 × 9 = 900 rows
- The dependent variable (Y) should contain the utility scores (ratings, rankings, or choices)
- The independent variables (X) should contain the attribute levels for each profile
- Attribute levels can be represented as numbers (1, 2, 3, etc.) and can be labeled using SPSSAU's data labeling function
Important Notes
- Conjoint analysis requires prior use of orthogonal experiment design to obtain experimental profiles
- All attributes in conjoint analysis are categorical data and will be automatically processed as dummy variables
- The first level of each attribute is used as the reference level in the regression model
- The utility value of the reference level is calculated as: 0 - sum of utility values of other levels in the same attribute
- Attribute importance is calculated using the "maximum range" method: maximum utility value - minimum utility value for that attribute
Frequently Asked Questions (FAQs)
Q1: What should I do if the system prompts "Insufficient valid samples for conjoint analysis"?
A: When this prompt appears, it means that relative to the number of attribute levels, the analysis sample is insufficient to perform OLS regression model analysis. It is recommended to increase the experimental sample size and analyze again.
Q2: What is the difference between rating method, ranking method, and choice method?
A: The rating method asks respondents to rate candidate products, with higher scores indicating greater preference. The ranking method asks respondents to rank candidate products, with better rankings indicating greater preference. The choice method asks respondents to choose yes or no for candidate products. Typically, the rating method and ranking method are used more frequently.
Q3: How is attribute importance calculated?
A: Attribute importance is calculated using the "maximum range" method: the maximum utility value of an attribute's levels minus the minimum utility value of that attribute's levels. This range is the importance value of the attribute. After normalizing the importance values, the relative importance of each attribute is obtained.
Q4: How is the utility value of the reference level calculated?
A: The first level of each attribute is used as the reference level in the regression model. The utility value of the reference level is calculated as: 0 - sum of utility values of other levels in the same attribute.
Q5: What do the Pearson and Kendall correlation coefficients represent?
A: These coefficients represent the model fit of the conjoint analysis. They are calculated as the correlation between the actual utility scores and the OLS regression fitted values. Higher coefficient values indicate better model fit.
Q6: Do I need to use orthogonal experiment design before conjoint analysis?
A: Yes, orthogonal experiment design is required before conjoint analysis. Based on the number of attribute levels, design to obtain potential "profiles" (products) for evaluation.
