Choice Base Conjoint Analysis
Purpose
This analysis method can be used to analyse data collected using the Choice Based Conjoint question type in EyeQuestion.
Background
Choice-Based Conjoint Analysis is a sophisticated market research technique used to decipher consumer preferences. It simulates real-world buying scenarios by presenting respondents with a series of choices, each representing different combinations of product attributes. By analyzing how individuals make choices within these scenarios, businesses can derive valuable insights into the relative importance of various product features and their impact on purchasing decisions.
Data Constraints
This analysis method requires at least one data point for each possible display location in the design. In other words, should your design present three product concepts per trial, on each trial a different product will be presented in each of three possible display locations. In the resulting dataset any panellist must have chosen any product at least once in all three of the possible display locations. If the dataset does not meet this constraint the analysis will not run in EyeOpenR and an error message explaining this issue will be presented to the user in EyeOpenR in the ‘Information’ table.
Options
You are presented with the following options before the choice based conjoint analysis is performed:
- Choice Based Conjoint Question: You can select from a drop down list the conjoint question you wish to analyse (the drop down list contains all the questions within your project)
- Number of decimal places: Defines the number of decimal places returned in the results tables (excl. Utility Estimate Distributions)
- Do your results include a none option?: You can indicate whether the ‘None’ option is included in yours results and therefore was available to panelists when selecting from their choice of product concepts.
Results and Interpretation
The analysis performed on your conjoint data within EyeOpenR generates, for each Judge|Panelist in your dataset, a utility estimate (the average of MCMC draws of unit-level model coefficients) corresponding to each attribute level within your design. This utility estimate can be viewed as an estimate of the weight applied by an attribute level to a given choice decision, with positive values increasing the likelihood of choosing a product with the given attribute level, and conversely negative values reducing the likelihood of choosing a product with the given attribute level.
The results of the above analysis are presented on four separate tabs within EyeOpenR, with each tab presenting distinct measures, each derived from the utility estimates but offering different insights into the results of your study. The information presented on each of these tabs is described below
Attribute Importance
This records the relative importance of each attribute within the design. Attribute importance is returned as two measures (in the below description the utility estimate is the summed utility estimate across all Judges|Panelists):
- Importance: This can be interpreted as representing the relative importance attributed to a given attribute when panelists made their choices in your study. It reflects the distance between the largest and smallest utility estimate for levels within an attribute.
- Percentage: This is simply the relative importance of the given attribute presented as a percentage relative to all other attributes in your study. This is derived as follows: (Importance for given attribute / (sum of importance of all attributes)) * 100.
Utility Estimate Distributions
The chart displayed here indicates the level to which your panelists agree or differ in the influence of each attribute. A distribution is displayed for each attribute level in your study. A distribution with a single high peak indicates your panelists were largely influenced in their choices in the same way by the given attribute level. Whereas if the distribution is spread widely across the parameter space then this indicates there was little similarity across you panelists in how this attribute level influenced their choice. It is also possible that you see multiple peaks for a given attribute, this suggests that within your study distinct groups of panelists exist in which the same attribute level influenced each group differently. More specifically, the table returned here, from which the distributions are derived, contains the utility estimate for each Judge|Panelist for each level of each attribute. The chart displays a plot for each attribute level, reporting how the various Judge|Panelist estimates are distributed across the Utility parameter space (distributions are constructed using Gaussian Kernel Density Estimation). Across all plots increased utility indicates a Judge|Panelist is more likely to choose a product with this attribute level.
Sensitivity: Attribute Level Utilities
The chart displayed here shows the influence of each attribute level on the choices made by all panelists in your study. A larger attribute level utility indicates that panelists were more likely to chose a product containing this attribute level. While conversely a smaller utility value indicates panelists were less likely to chose a product containing this attribute level. The attribute level utilities are simply the sum across all Judges|Panelists of the utility estimates for the given attribute level.
Market Performance by Product Concept
This table can be interpreted as representing all product concepts in terms of most likely to least likely to be chosen. This table contains all possible product concepts within the project design ranked from largest to smallest in relation to a product’s Total Utility where Total Utility is simply the sum of attribute level utility estimates for all levels that define the given product concept.
Information
This table records any error messages and/or warning messages generated when analyzing your data.
If there are no data points within the data set for any one of the display locations (see above NOTE) the analysis will not run and therefore the following error message will appear in the information table providing explanation “The following display locations were never selected: [DISPLAY LOCATIONS]. In your dataset each location must have been selected at least once in order to run this analysis.”.
If you have indicated that a none option is present in your design [see 'Options within EyeOpenR' section above] but no none option is detected in the data set then the following warning message is displayed in the information table "WARNING: None option not recorded in dataset and therefore not included in analysis". Conversely if you have indicated that there is no none option present in the design yet a none option is detected in the dataset the following warning message will appear in the information table “WARNING: None option was recorded in dataset and therefore included in analysis".
Technical Information
Within EyeOpenR the analysis of results from choice based conjoint studies is performed using Hierarchical bayes multinomial logit models with MCMC used to generate model estimates. This is implemented using the choicemodelR R package (Sermas, 2012).
References
- Sermas, R. (2012) ChoiceModelR: Choice Modeling in R. https://CRAN.R-project.org/package=ChoiceModelR
Related Articles
Conjoint Choice Based
Introduction Choice-Based Conjoint Analysis is a sophisticated market research technique used to decipher consumer preferences. It simulates real-world buying scenarios by presenting respondents with a series of choices, each representing different ...How To Built A Choice Based Conjoint Questionnaire Using Eyequestion
What is Choice-Based Conjoint Analysis? Choice-Based Conjoint Analysis is a sophisticated market research technique used to decipher consumer preferences. It simulates real-world buying scenarios by presenting respondents with a series of choices, ...Free Choice Profiling
Introduction Free-choice profiling (FCP) is a quick and inexpensive method in which consumers are asked to both identify attributes in the sample and rate the liking and/or intensity of those attributes. They should be provided with adequate ...Penalty Analysis
Purpose To provide a penalty analysis of a consumer data set, that is to investigate how liking or acceptability of product decreases when product attributes are not at the optimal intensity. Data Format Consumer.xlsx Note: for EyeOpenR to read your ...Multiple Factor Analysis (MFA)
Purpose Performs a multiple factor analysis on a data set. This is not ‘Factor Analysis’ in the classic statistical sense, but rather a method for handling multiple groups of variables that are all measured on the same samples, and from this ...