Sawtooth Software: The Survey Software of Choice

What is MBC (Menu-Based Choice)?

Over the past decade, there has been increasing interest in designing and analyzing menu-based choice questionnaires, as they often more realistically reflect real-world buying situations than standard Choice-Based Conjoint (CBC) or ratings-based conjoint. Examples in the literature include articles by Liechty et al. (2001), and Cohen and Liechty (2007).

The main difference between MBC and other discrete choice methods from Sawtooth Software (MaxDiff and CBC) is that MBC can handle a variety of menu choice situations in which respondents make from one to multiple choices in the process of building their preferred selection.

In conjoint analysis, we consider multiple factors (attributes), where each attribute has at least two levels. Menu-based choice problems also involve multiple factors, each having multiple levels. Whereas we often think of a CBC question as being composed of multiple product concepts (cards), we should think of the entire MBC menu question being represented by a single card. This allows researchers to use the familiar tools for conjoint design with MBC experiments (CBC or CVA Software, Warren Kuhfeld's SAS routines), except that the number of factors for MBC experiments will often be much larger than for traditional conjoint or CBC.

Questionnaire development and data collection may be done with any questionnaire instrument (even paper-and-pencil) or web survey tool of your choice, including Sawtooth Software's Lighthouse Studio program. We generally recommend using randomized design strategies, as they are robust and convenient to use for MBC studies. If you use Lighthouse Studio, you will need to customize the questions using Free Format and typically some customized Javascript code that you must write on your own.

Analysis includes commonly known tools of Counting, logit, and HB analysis. The statistical routines used in MBC are the same as used in CBC. However, the model specification and sheer number of inter-related models is more complex and greater in MBC. MBC software manages a complicated process with elegance and simplicity. During the model specification process, the researcher can preview the design matrix, so the process is transparent and never becomes "black-box." Own-effects, cross-effects, availability effects, conditional dependencies, linear, log-linear, and part-worth terms are supported. Despite the flexibility, MBC automatically handles so many of the coding aspects that can occupy so much time if attempted manually.

The final deliverable of an MBC project is usually the market simulator. MBC software provides a simulator that can project what percent of the respondents are likely to pick each item from a menu, given a set of menu prices. If using HB estimation of the parameters, MBC can also simulate expected combinations of items that respondents are most likely to pick.

The most stringent test of how well MBC performs hinges on the ability to predict individual-level combinatorial choices of items. Across multiple studies, MBC software has proven adept at predicting individual combinatorial choices. In our most stringent test, MBC combinatorial predictions approached (92% of) the level of test-retest reliability. This means that the MBC simulator was able to predict respondents' exact choices on a menu 92% as successfully as when respondents were asked to repeat the same menu task at a later point in the survey.