The Conjoint Design

In order to allow accurate estimation of the preference associated with each conjoint item (i.e., conjoint utilities), the conjoint statistical design must be properly prepared and must satisfy some balance criteria.

While it is possible to import and use experimental designs prepared with alternative tools/software (R-sw Conjoint can run analyses based on any type of design, such as orthogonal, d-efficient, full profile, partial profile, alternative-specific, etc.), R-sw Conjoint includes powerful functions to generate and check full profiles designs suitable for Rank-Based Conjoint (RBC), Allocation-Based Conjoint (ABC) and Choice-Based Conjoint (CBC) models. In a full profile design, a scenario is described in terms of all attributes included in the study (one level is chosen for each attribute).

The criteria currently implemented within R-sw Conjoint are:

  • high heterogeneity: scenarios within a task (CBC only) and across tasks are as heterogenous as possible;
  • no within-task duplication (CBC only): the same scenario is not repeated within the same task;
  • no within-version duplication: the same scenario is not repeated within the same version (this is not possible if the number of scenarios within a version is higher than the number of possible profiles);
  • maximum within-task balance (CBC only): attributes levels are not repeated within the same task. This is not possible if the number of levels for an attribute is lower than the number of scenarios per task; in this case attributes levels appear with a similar frequency across the scenarios of any given task to ensure level balance. 
  • maximum within-version balance (ABC/RBC only): attributes levels appear with a similar frequency across the tasks of any given version to ensure level balance. For a CBC design, within-version balance can be improved by setting a high number of iterations, which dictates how many times the design generation is repeated; the final design is the one with the highest balance;
  • one-way balance: attributes levels are included in the design a similar amount of times; One-way balance can be improved by setting a high number of iterations, which dictates how many times the design generation is repeated; the final design is the one with the highest balance;
  • two-way balance: the levels of any attribute appear together with the levels of any other attribute in the design a similar amount of times (low priority criteria);
  • positional balance (CBC only): the levels of any attribute appear in each task position (first scenario, second scenario, third scenario, etc.) a similar amount of times (low priority criteria).

The function generate.design available in R-sw Conjoint generates a RBC/ABC or a CBC design. It is possible to specify prohibitions, i.e., combination of two or more attributes levels that should not appear in any scenario.
The function check.design produces summary statistics for any RBC/ABC or CBC design, so that it is possible to assess how a design performs on the various balance criteria.

If you would like to use a design other than a full profile (e.g., partial profile) and you do not have access to alternative design software, we can help:

  • if you are a licensed user of R-sw Conjoint, we can prepare for you the experimental design free of charge (up to 10 designs per license period) according to the project objectives and your specifications (we can provide our recommendations if required);
  • if you are not a licensed user of R-sw Conjoint, we can still support you with the conjoint design, data conversion/editing, data analysis and simulator development. The cost for our support depends on the project requirements. For further information, please email Demia: consulting@demia.it.