 |
|||
|
GfK Healthcare | February 2011 | |
|
|||
Extending the Use of MaxDiff in Health Care Marketing Research  By Andrew Douglas, Senior Vice PresidentThanks in part to seminal work by Jordan Louviere, MaxDiff (maximum difference) scaling has come to be one of the most utilized methods within market research to understand self-explicated preference and importance. While past Pipeline articles have pinpointed the benefits and issues in using MaxDiff over typical ratings and rankings, this article will outline three techniques developed to extend the utility of the MaxDiff within health care research. Using ‘Splits’ to Find Targets Because the MaxDiff method uses iterative responses, one of its benefits is that the modeling can be done at the individual level through hierarchical Bayes methods. These methods generate better predictability in comparison to aggregate models (finding the “average” importance across individuals) and more robustly define the heterogeneity of response variability as signal versus noise per respondent. The MaxDiff is an “ipsative” measure (a nonnormative forced choice), yet it is iterative so that item level specific utility can be inferred. Ipsative measures are valuable indicators of attitudinal differences between items within respondents; thus individuals who have similar views can be grouped. From a practical standpoint the MaxDiff method allows you to understand what is important and the magnitude of the items’ importance for each respondent; we must remember that this is done from a simple choice task so the respondent is not overburdened. When you simply look at the individual “importance” data it becomes clear that there are distinct differences in what individuals prefer; a closer look confirms that patterns emerge across these preferences. As common sense would dictate, these patterns indicate that groups of respondents similarly find items more or less preferred. Across most projects we find these patterns are less related to the a priori or demographic splits (e.g., specialties, patient types), but rather are more related to the underlying attitudinal segment of the respondent. By using an optimized clustering algorithm and nonparametric statistical testing, significant differences between the segmented groups can be identified. As an example, some physicians may be more concerned with compliance and safety, some more interested in immediate impacts, while others may focus on disease progression (example below). By using this simplistic attitudinal segmentation, we have found that the differences between groups are relevant and targetable for marketing and sales thrusts. Acknowledging that preference/importance can differ depending on the segment can lead to more effective strategies to reach and engage more of your total population’s needs. This is a form of benefit segmentation that can be done at a fraction of the cost of a full market segmentation exercise. Simple cluster analysis identifies physicians whose importance scores reliably differ from each other.  The key differences differentiating the groups of physicians lie in the contrasting importance of these three benefits:  The Adaptive MaxDiff From a client perspective, during the course of product development the number of requests and inputs from stakeholders can be seemingly unmanageable. When a marketing team pulls together an exhaustive list of benefits for a new drug, or when a device manufacturer is given an overabundance of 50+ attributes from the engineers, the initial push back is (and should be) to reduce the list. Many methodologists would constrain the absolute number of items to test to be less than 30 when utilizing a MaxDiff exercise. The issue is not the ability to parameterize or model more data, but rather in the inability of the respondent to react in an engaged manner to upward of 20 to 25 screens. In most cases, stakeholders try to gnaw away at the list to achieve a more reasonable expectation, but in some cases the need to understand importance/preference for the majority of the list remains a necessity. When we as vendors ask for attribute lists to be shortened or nontactical messages to be excluded, the task of actually shortening the lists can be a no-go. Within the past four years GfK Healthcare has pioneered a response task and analytic method that gives a robust solution to this problem. In this case, we use a similar process to an adaptive conjoint analysis where a two-step choice allows for optimizing the top end of importance or preference list, while those that fall to the bottom are less important/preferred, gain less utility and are less differentiated in the model. The task initially presents the entire list to the respondent and asks him or her to subset the more important items from the less important. For example, from a list of 50, the respondent selects the 20 most influential items. The task then utilizes an experimental design to trade off the selected items in a typical MaxDiff exercise. Akin to partial profile design across respondent, the analytics allow for item utility to be generated for each attribute. Those items that are excluded initially per individual are assigned a null utility while those that are chosen are analyzed per individual for their expressed utility. Although simple in execution, the algorithm has been optimized to isolate specific utility for upward of 50+ attributes. Although the items that are selected less frequently in the set do suffer from collective lack of response, this in itself is informative, and those items are probably less important to strategists. The findings can focus marketing and sales strategy from the almost infinite to a more select and actionable subset of items. Strategic Messaging GfK Healthcare uses an analogous method in identifying influential attributes/messages across several topics. From a marketing perspective, tactical promotion should cover a range of topics so the reach of any given item is not duplicated by another, and the full range of important topics is covered for the given product. It would not necessarily make sense to promote five similar efficacy messages and not cover safety, dosing and access. It would be nice if, out of 15 messages, the six efficacy messages were determined to be the most influential, then the two safety messages come next, followed by four access items and finally the three dosing messages bring up the rear. However that is atypical. The utility of any one topic can span from lacking to high influence. Typically, our clients want to understand what the best item within the topic is, how those items relate to each other and how they relate to items across topics. This task can become insurmountable as the number of topics and the number of items within topic grows. Using a similar dual-step approach, we have created proprietary algorithms to calculate all cross utilities. Respondents are presented the original list items that have been categorized into their corresponding topic area (buckets). The respondent is asked to select the most and least influential items within each topic. Therefore, if there were 50 items across 10 topics, 20 items would be selected (it is even useful to have the respondent select the two most and the least influential for each topic). The original list of 50 items has now been reduced by more than half. This allows those 20 items to be traded off in a typical MaxDiff exercise without concern for fatigue. Those items that are traded off get their expressed utility. For the remainder of items not chosen, item utility is calculated by giving partial credit to those that were not selected, but were in the same topic as a chosen item during each MaxDiff portion of the task, keeping in mind this is an iterative process that allows associated items to accrue points across entire task. The results from this technique and analyses are quite robust: 1). Attribute-specific utility can be calculated; 2). the range of importance for any topic can be determined; 3.) and the specific item within topic utility can be compared. Marketing and sales decisions depend on understanding what attributes influence and motivate consumers to act. We have been extending and modifying MaxDiff to help make importance and preference data more actionable for our clients’ decision making. Want to learn more on this topic? Please contact:  |