In surveys, customers declare that everything matters to them: price, quality, delivery, service, sustainability, and brand. Such a result does not support any decision, because an offering cannot be improved on ten fronts at once. A MaxDiff preference study solves this problem by forcing respondents to make real choices and ranking attributes according to the importance revealed through those choices rather than simple declarations.
How does MaxDiff force customers to make real choices?
MaxDiff (short for Maximum Difference Scaling, also known as best-worst scaling) is a preference measurement method in which respondents do not evaluate attributes in isolation but compare them with one another. In a single task, they see a set of several attributes – most often three to five – and select two extremes: the one that is most important to them and the one that is least important. The task is repeated multiple times, each time with a different combination of attributes selected according to a planned experimental design.
The key difference from traditional methods is that a MaxDiff preference study removes the respondent’s ability to say, “everything is important.” A rating scale allows respondents to select “5” for every attribute, resulting in flat, unhelpful findings. MaxDiff works differently: because respondents must choose one most important and one least important attribute from the presented set, the actual trade-offs emerge. These customer trade-offs are at the heart of the method and distinguish it from simple ranking.
Why does this matter right now? Offerings in most categories are becoming increasingly complex, while the list of potential improvements almost always exceeds available budgets and implementation capabilities. A product manager, marketer, or analyst needs a solid basis for prioritization – information about which elements of an offering actually influence audience decisions and which are merely nice additions. MaxDiff provides such a hierarchy of product attributes on a single, comparable importance scale.
The result of the analysis is a ranking of all examined attributes from most important to least important, expressed as numerical indicators. Because all attributes are measured on the same scale, it is possible to compare the size of the gap between individual attributes. This is a significant advantage over methods that generate isolated ratings that cannot be compared directly.
How should a MaxDiff analysis be designed and conducted?
Conducting the study correctly requires several structured steps. Below is the typical course of a project carried out using a workshop-based approach:
- Defining the attribute list. The starting point is a list of offering attributes to be ranked. They should be phrased clearly, be comparable in their level of generality, and not overlap with one another. It is worth preceding this stage with qualitative research to avoid overlooking attributes that matter to the target audience.
- Designing the experimental design. An algorithm distributes attributes across successive tasks so that each attribute appears with similar frequency and in different contexts. This ensures balance and makes it possible to reliably estimate the importance of each element.
- Conducting the measurement. The respondent completes a series of screens requiring them to select the best and worst attribute. The number of tasks is selected to obtain stable results without overburdening the participant.
- Analysis and modeling. Individual-level and aggregate importance scores are estimated based on the choices, often using hierarchical Bayes models. This makes it possible to obtain a hierarchy of product attributes at the individual respondent level as well, opening the way for segmentation.
In practice, the most important aspect of a project is often not the calculation itself, but the selection and wording of attributes. As Hume’s Institute experts point out, when declarative results show that “everything is important,” in reality nothing is a priority – and MaxDiff forces a choice and reveals the actual hierarchy of attributes that can be translated into operational decisions. This observation goes to the heart of the method: the value of the study increases precisely where traditional surveys fail, namely when distinguishing between attributes that all seem “fairly important.”
The results of a MaxDiff analysis are often presented, after rescaling, as an ordered list of attributes with assigned importance values that add up to a fixed total. This makes attribute prioritization clear: it shows not only the order, but also the gap between individual attributes. It often turns out that several attributes account for most of the “weight,” while the long tail of remaining attributes is of marginal importance to the audience – a valuable operational insight when designing an offering or communications.
It is worth adding that MaxDiff combines well with segmentation. Because, with appropriate modeling, the method allows preferences to be estimated at the individual respondent level, it is possible to identify groups with different attribute hierarchies – one values price most, another reliability, and yet another after-sales support. This extends the usefulness of the study beyond a single averaged ranking.
What mistakes should be avoided, and how does MaxDiff differ from conjoint analysis?
Although the method is resistant to many of the weaknesses of rating scales, it has its own limitations and design pitfalls. The most common factors that reduce the quality of results are listed below:
- An overly long attribute list. The more attributes there are, the more tasks a respondent must complete for each one to appear sufficiently often. An excessively extensive list tires participants and reduces reliability. It is better to filter out obvious or redundant attributes in advance.
- Inconsistent levels of attribute generality. Comparing a very broad attribute with a very specific one distorts the comparisons. Attributes should be formulated at similar levels of specificity.
- Confusing importance with willingness to pay. MaxDiff measures the relative importance of attributes, not their price or their impact on the choice share of a specific product variant. This limitation should be kept in mind when interpreting the results.
- Treating the result as a strategic recommendation. An attribute hierarchy is an input into decision-making, not the decision itself. The method describes audience preferences; it does not replace feasibility or cost analysis.
The most common question, however, is how MaxDiff differs from conjoint analysis. Both methods are based on choices and reveal customer trade-offs, but they answer different questions. A MaxDiff preference study ranks individual attributes by importance – it answers the question, “what matters most?” Conjoint, by contrast, examines how specific attribute levels (for example, a price of 99 zł versus 129 zł, or delivery within 24 hours versus 48 hours) combine into complete product variants and affect choices between them.
In other words, MaxDiff operates at the level of attributes themselves and their importance, while conjoint operates at the level of configured offering variants with specified levels of those attributes. In Hume’s Institute projects, MaxDiff works well as a first step when there is a large number of attributes and the list needs to be narrowed down. Conjoint comes into play when the set of attributes is already limited and the goal is to model purchase decisions and simulate choice shares.
When is it worth using MaxDiff?
MaxDiff is not the answer to every research question, but in certain situations it is clearly a better method than the alternatives. Typical use cases are listed below:
- When a large number of offering attributes, messages, or benefits need to be ranked by their importance to the audience.
- When earlier rating-scale studies produced flat results in which almost everything received a high rating.
- When attribute prioritization is needed before modeling product variants using conjoint analysis.
- When comparability of results across different markets or segments is important – MaxDiff is less susceptible to differences in response styles than rating scales.
- When the goal is to identify audience segments with different need hierarchies.
If, on the other hand, the goal is to estimate willingness to pay for a specific attribute level or simulate choice shares between variants, conjoint analysis will be the more appropriate tool. Method selection should always begin with the research question, not the technique.
Frequently asked questions
How does MaxDiff differ from a rating scale?
A rating scale allows respondents to evaluate each attribute independently, which often leads them to select high values for all attributes and produces flat results. MaxDiff requires respondents to choose the most and least important attribute from a set, thereby revealing actual trade-offs. The result is a single comparable importance scale rather than a collection of isolated ratings that are difficult to compare.
How many attributes can be studied using MaxDiff?
The method can handle lists ranging from several to several dozen attributes, although the longer the list, the more tasks the respondent must complete for each attribute to appear sufficiently often. In practice, the aim is to keep the list as concise as possible by first filtering out obvious or conceptually overlapping attributes. The key is to balance the number of attributes with participant comfort and attention.
When does MaxDiff work better than conjoint?
MaxDiff works better when the goal is to rank individual attributes by importance, especially when there are many of them and the offering is at an early stage of being organized. Conjoint is selected when it is necessary to model choices between complete product variants with specified attribute levels, for example in price analysis. The two methods are often used sequentially – first MaxDiff to narrow the list, then conjoint to model decisions.
Want to organize your offering’s attributes according to their actual importance to customers? Ask about a MaxDiff study of offering priorities – Hume’s Institute will help select the scope of attributes and design a measurement tailored to your research question.