Designing a Conjoint Study – A 9-Step Checklist
Ready to run a conjoint but not sure where to start? Use this nine-point roadmap as your pre-flight checklist. It keeps the project realistic—and keeps you from drowning in attribute overload.
The process
To be certain that the results will be as useful as they can be, take your time to prepare. You may need to collaborate with other departments within your organization to have all the information needed to build all the inputs necessary for an effective conjoint. Here is a step-by-step guide:
Identify product attributes/benefits/features to be included: These could include size, color, flavor, price etc. This should focus on feasible options that can be produced! Though the list can be long, a longer list of potential attributes makes the conjoint design much more complicated. We suggest a list with up to 7 attributes.
Identify the levels within each attribute: This would include all the variations within each attribute, for example flavors could be vanilla, chocolate, strawberry etc. At this point again think about what is feasible for your product or service. Though you could include many variations, we recommend no more than 5 levels or variations per attribute.
Determine the acceptable pricing: We separate pricing from the other attributes as it’s so critical. Select price levels that make sense – going too low or too high could skew the results. We suggest doing some research before deciding on the acceptable price range and also accounting for production costs, marketing costs and the desired margin. All these inputs should be considered before designing your conjoint research.
Identify any exclusions: Are there any combinations of product features, levels or pricings that just do not make sense or are not realistic? For example, you would not want to show consumers a product that includes premium features at the lowest possible price. Make sure you identify these combinations and exclude them from the design. However, keep in mind that having too many exclusions throws off the results as it reduces how many times a level can be shown, so we recommend limiting exclusions to 2 or 3.
Create visuals: Pictures can tell a thousand words and the same is true with product features! If a product feature is easier to understand and more compelling if displayed visually, create a visual for that. Consumers will deem a feature less important if they do not understand it!
Establish number of tasks to be shown: As mentioned previously, in a conjoint, consumers are exposed to a few full product profiles simultaneously and are asked to select their most preferred. The question is, how many times should they go through this exercise? There is a formula you can use to determine the number of tasks, but a good rule of thumb is between 8-12 tasks.
Determine how many full product profiles to show in each task: At this step think how many full product profiles respondents can see at the same time and be able to internalize and make a thoughtful decision as to which one is the most preferred. This is not just a matter of the amount of information presented and whether one can realistically read it and understand it, but it is also a matter of how it would look visually on the screen respondents are using. This may look ok on a laptop, desktop or tablet, but how would it look on a smartphone? As a rule of thumb, we recommend between 3-5 profiles per task.
Consider adding a None of these option for each task: Consumers may not find any of the product profiles appealing, so realistically, they may walk away from such a purchase, so we should give them that option as well.
Sample size: The rule of thumb is that for each group you want to analyze, you need a minimum of about 150 respondents. So, if for example, you want to see your conjoint results by males and females, you would need a minimum of 300 respondents. The number of attributes and levels also impacts sample size.
Follow the checklist and you’ll collect data that’s sturdy enough to guide pricing, feature bundles, and launch plans. In the next blog post, we’ll dig into what the outputs actually mean and how to run quick simulations.
