Editor's note: Seth D. Fowler is vice president at Harman Atchison Research, Overland Park, Kan. The author wishes to thank John Howell for his guidance on the technique and review of this document. Additionally, thanks go to Joe Harman, Sherri Atchison, Jake Lee and Joe Curry for their review of this document and their helpful comments.
Choice-based conjoint is a research technique, popularized in the 1990s, that allows researchers to understand the utility of different product features and benefits. The technique is very good at determining the relative utility of individual product attributes, preference for sets of attributes (product/service concepts) and the magnitude of differences in preference. The technique has been improved in various ways since its introduction but current standardized methods force a single response and lack the ability to appropriately include volume-based responses. Companies want to understand preference but, more importantly, how changes to a product or service might impact purchase volume and, ultimately, sales and revenue. This article examines the application of volumetric data in conjoint research.
Choice-based conjoint is one of the few techniques that allows us to break down the utility of individual product attributes and build a model that can simulate not only preference for a current design but also understand the relative impact on preference of multiple changes to that design. This, in itself, has provided valuable insights to companies trying to understand how they can improve their offerings. In many instances, the information gained from choice-based conjoint can be used in concert with other information, such as the cost of the different attributes tested, to develop models that optimize the offering for consumers while minimizing costs to the business. For example, two attributes might generate similar levels of preference but one might cost less to supply. In this case, the product can be optimized for the consumer and the business. Consumers are just as happy with either option but the company can reduce its cost to provide the product.
Need more information
Regardless of the benefits current choice-based conjoint techniques provide, companies want and need more information. One key disadvantage of choice-based conjoint, mentioned previously, is that it does not measure the volume of items a consumer might purchase; it only measures preference for the items tested. This requires the addition of questions outside of the conjoint exercise to help determine how much more or less of a product consumers would actually purchase based on potential changes. Researchers have to make assumptions about the degree of impact on purchase volume created by changes based on both conjoint- and non-conjoint-related data from the survey. This has been a stumbling block for choice-based conjoint research in the past.
But what if the technique could produce volumetric data to model changes in purchase behavior? This would allow for a whole new level of analysis. Instead of only being able to calculate preference, we would be able to calculate changes in purchase frequency or volume. This information would be of great value and allow for decision tools that are more robust and able to optimize products and services based on potential changes in revenue and profit.
A logical next step
With the application of advanced statistical methods, this is becoming a reality. The inclusion of volumetric information in conjoint-based research is an area of current academic research (John R. Howell, Sanghak Lee, Greg M. Allenby 2015 and John R. Howell and Greg M. Allenby 2012). These techniques can be incorporated in choice-based studies to allow for the usage of many types of volumetric data such as the number of items that would be purchased or the number of store visits within a particular time frame. The application of these statistical techniques represents a logical next step in the evolution of conjoint analysis.
Current choice-based conjoint techniques frequently use what is referred to as a “dual-response none” method of collecting responses in the respondent task. Respondents are typically shown two to four combinations of attributes (product/service concepts) and first asked which one of the choices they prefer. On the same screen, they are also asked if they would actually purchase the product they indicate they prefer, or not. This allows for the estimation of individual attribute-level utilities and the calculation of shares of preference. However, this forced-choice setup is not always ideal, especially in cases where purchasing multiple items is the norm or even required.
With this new approach, respondents can input any number to reflect the amount they would choose as opposed to selecting only the most preferred option. This allows for understanding preference as in traditional choice-based techniques and also allows for determining the degree of preference since respondents can increase or decrease the volume according to their level of preference. The volumetric element is very flexible and can take many different forms, such as the number of purchases or store visits over a certain period of time or even the specific number of an item that consumers would purchase. The volumetric variable used is highly dependent on the type of information needed and how the researcher plans to size the opportunity.
Case study example
To illustrate this point, we will use an example of a rewards program for a fictional hotel chain, Traveler Hotels. Those who travel for business or pleasure have many choices for accommodations and hotel loyalty programs can have a positive impact on a traveler’s share of bookings. Traveler Hotels is interested in maximizing its share of annual bookings among its current base of rewards program members and enticing non-members to enroll in the program. To meet this objective, Traveler Hotels wishes to develop a compelling rewards program that builds loyalty and commitment to the brand. Its current rewards program is basic and consists of earning points for every dollar spent at Traveler Hotel properties. These points can be accumulated and exchanged for free nights at any Traveler Hotel property.
Traveler Hotels has developed a list of potential rewards to add to the program but it first needs to understand how these different possible rewards will impact the number of times rewards members and prospective rewards members are willing to stay at its hotels. The list of benefits under consideration includes:
- earning and exchanging points for free-night stays (current rewards program);
- double-points earning for a length of time after signing up for the program;
- allowance of free food and beverages for the first calendar year after signing up;
- the ability to combine points and cash to pay for nights;
- free room upgrades after earning a certain number of points;
- exchanging points for airline miles at partner airlines;
- point-earning bonuses after reaching point earning milestones;
- access passes to airline partner airport lounges after reaching point earning milestones.
Each of these potential rewards (attributes) has an associated cost. Further, Traveler Hotels needs to understand which one(s) will increase share of annual nights booked and by how much. Traditionally, we would show the respondent three or four possible program concepts on the screen, ask them which they prefer and if they would stay at Traveler Hotels more often if the particular program was offered. With this revised method, the dual-response none is replaced with a volumetric measure. Instead, we ask how many times they are likely to stay at Traveler Hotels and two competitors in the next 12 months if the programs detailed were offered. Competitive brands are included in the exercise to allow for the element of competitive pressure. Figure 1 is an example of what the exercise might look like. As with traditional choice-based designs, the respondent completes several of these tasks indicating how many nights they would stay during the next 12 months at each hotel if the concepts displayed were the rewards programs available.
These volume estimates are then broken down. Just as in traditional methods where part-worth utilities are calculated, volume estimations are calculated for each attribute tested. Using a custom-designed model, we can then calculate how many nights respondents estimate they will stay annually for any given combination of the attributes (rewards programs). Using a volumetric measure adds a new dimension to the model. This information, along with the cost of the different rewards program options and per-visit spend can be modelled to determine the relative increase (or decrease) in revenue and profit for any given reward option tested. The current Traveler Hotels base rewards program is included in the model, allowing us to understand the impact changes to the program will make relative to the current program. The inclusion of competitive brands in the exercise creates a more realistic choice task for the respondent and helps alleviate some of the problems encountered with overestimating nights stayed if only one brand were present in the exercise.
A wide variety of situations
The example above uses a fictional hotel rewards program to demonstrate how this volumetric component can overcome some of the limitations of current choice-based conjoint methods. The technique adapts currently available commercial software to collect the data but the analysis is a custom statistical application. It can be applied in a wide variety of situations for different products, programs and services. It should be noted that these volume estimations are subject to the same limitation as traditional choice-based conjoint techniques, one such limitation being that an estimation of future purchase behavior may be impacted by any number of intervening variables. Product optimization should be the goal rather than the prediction of actual market performance in terms of revenue and profit. As with current choice-based conjoint techniques, it is important to minimize the complexity of the task respondents complete. Researchers should be aware of these issues as they formulate and report estimations in changes of volumetric purchase behavior.
This new technique represents an exciting development in conjoint analysis, broadening its ability to give researchers key market insights. This new tool allows for the use of volumetric information where previously only preference was established. As described, it does this by allowing respondents to indicate volume changes in their purchase behavior, based on changes to the product or service. Since this is a newer technique, little additional literature is available on the subject. For additional reading, please refer to the papers referenced below. This new technique represents a strong enhancement to current choice-based conjoint applications. As its adoption grows, companies will be keen to make use of its ability to predict consumer response to product changes.
Howell John R., Sanghak Lee, Greg M. Allenby. (2015) “Price promotion in choice models.” Marketing Science.
John R. Howell and Greg M. Allenby. (2012) “Volumetric conjoint analysis, incorporating fixed costs.” Proceedings of the Saw-tooth Software Conference, Sequim, Wash.