Advanced Analytics: Understanding stakeholder preferences through conjoint analysis
Conjoint analysis is a statistical method that can be used to evaluate and measure the relative value that patients and healthcare professionals place on different attributes or features. Whether your objective is to identify to the optimal features for your new delivery device, or evaluate which features of your drug are most important to highlight in an upcoming launch campaign; conjoint offers a more robust alternative to traditional rank/ rate questions, and ultimately enables more informed decision-making
For example, if you ask an asthma patient how important it is that their inhaler can be stored at room temperature, they may respond with an 8 or 9 on a 10-point scale. Ask them how important a dose tracking feature is, or that the device is small and portable – you are likely to find that the patient continues to answer at the top end of the scale, suggesting that all attributes are equally important … not particularly useful when you are trying to decide which delivery device to proceed with for your new drug development!
Another potential issue with rating scales is that when conducting research across multiple markets, cultural differences in how people provide feedback means that ratings may not be directly comparable (e.g. an average rating of 8 in the UK may have a very different interpretation to a rating of 8 in Sweden). Do certain markets, or specific sub-segments in your target population have a higher preference for a certain attribute, or are we just seeing scale-meaning bias?
A ranking question can go some way to overcoming these issues, but there are also other factors to consider. What if you have around 30 attributes to test for preference/ importance? A direct ranking question is likely to place a high level of cognitive demand and burden on respondents (especially if the attributes are quite wordy, e.g. with message testing), leading to fatigue and unreliable results. Similarly, you may have two features that both rank highly, but that in reality are incompatible with one another (e.g. inclusion of a smart dose tracking feature and cost).
Conjoint analysis helps solve these problems by using trade-off exercises to mimic the stakeholder decision-making process. In real life, we are constantly forced to make choices based on the importance we attach to different features of a product or service, even though neither of them may be perfect. By using conjoint we can determine the optimal features for a product or service, and estimate the relative weight stakeholders will give to various factors that underlie their decisions.
Importantly though, conjoint is not a one size fits all methodology, instead encompassing a number of different analytical techniques, such as choice-based, adaptive or menu-based. Two of the most popular methods are explained in more detail below.
MaxDiff (Maximum Difference Conjoint)
MaxDiff (also known as best-worst scaling) is a technique used for obtaining relative preference/ importance scores for multiple attributes. Respondents are typically shown 3-5 attributes at a time and asked to indicate their least and most preferred. The task is then repeated, each time showing a different set of attributes. The number of sets shown, and the contents of each set are statistically derived, enabling you to estimate the relative importance of each attribute.
A common application for this technique is message testing. For example, you may be looking to launch a new campaign for your drug, and you want to know what specific messaging/ product claims resonate best with your patients, how these messages stack up against the competition, and whether there are any sub-segment nuances. In this scenario, a MaxDiff has several advantages over a traditional rating/ ranking exercise:
- Stronger discrimination power
- Allows you to test a large number of messages without respondent burden (around 30, or potentially more with a Sparse MaxDiff design)
- Provides relative scores (allowing you to distinguish the relative distance between yours and your competitors messages)
- Removes scale-meaning bias, so that comparisons between markets or other sub-segments can be more effectively conducted
Considerations: MaxDiff estimates the preference of attributes relative to each other but does not tell us if we have a good or bad set of messages from an absolute perspective (although inclusion of competitor messages can help with this). In addition, from a MaxDiff alone, we are unable to tell which attributes or messages work best in combination.
ACBC (Adaptive Choice-Based Conjoint)
ACBC is one of the most advanced approaches to preference modelling, which combines the strongest elements of Choice-Based Conjoint (CBC) with Adaptive Conjoint Analysis (ACA). The model learns from respondents as they answer each question, allowing for a more dynamic approach that is very effective for testing different versions of the same product or service.
For example, you may be wanting to understand prescribing preferences for breast cancer treatments, taking into account different attributes (such as side effects, cost, mode of action and route of administration), with various levels to be tested for each attribute (e.g. for side effects: risk of addiction, risk of heart attack etc.) ACBC would allow us to identify the most important factors for oncologists in driving prescription choice and also generate a comparative market share for any combination of attributes.
The methodology incorporates three major sections – build your own (where the respondent builds their optimal product/ service configuration), screening (where ‘must-haves’ and ‘unacceptables’ are assessed), and choice tasks (where the tradeoff tasks with the remaining product configurations are tested).
Considerations: More complicated and longer than traditional CBC. However, respondents tend to find the adaptive nature of the survey more engaging, and it also allows you to probe deeper into each respondent’s choice hierarchy. In addition, this methodology can be used on smaller sample sizes versus CBC.
In conclusion, next time you need to understand stakeholder preferences consider using a conjoint methodology, for greater discriminatory power and improved insights to support your decision-making.