A typical discrete-choice project consists of a number of phases, as shown in the following diagram.
Phase 1. Describe product attributes and levels.
In this phase, research practitioners define the scope of the research effort. They do this by agreeing on a list of attributes* that characterize products in the marketplace, including their proposed product but also any product against which they might compete. Attributes can include a range of things, from brands, pricing and features to advertising messages, promotional activities and bundling. The Design Wizard looks for attributes as a row of text cells in Excel, as in the range A1:F1 below.
*Attributes: Characteristics of a product or service, such as price, brand, feature, advertising message, etc.
This example involves cellular phones and carriers, the objective being to understand how consumers choose among mobile phones with differing brands, features, styles and prices. Together these differing characteristics form a set of attributes, where each attribute has a number of levels* associated with it. Levels represent the range of values attributes can take on. In our example below, we have added levels to each attribute, so for example the attribute Brand contains the levels Motorola, Ericsson, Nokia and Sony. The levels appear in blue text to make them easier to distinguish.
*Levels: Values an attribute can take on, such as red, blue and green for the attribute color.
Putting together an attribute list may sound like a trivial exercise, but in practice this can take considerable time and effort. Developing attribute lists for new products may require a combination of brainstorming, focus groups or other research activities to refine the list to a point where it is useful.
A subtle and difficult part of this exercise involves defining the competitive space for the product from the point of view of the customer. Often this is not the same as the universe of competitors that product manufacturers use. Rather, this is the economic definition of competition, which includes all substitute products of any type. Focus groups or other marketing techniques may be required to uncover the appropriate set of competitors.
Phase 2. Construct experimental design.
Most, but by no means all, discrete-choice studies involve the construction of hypothetical marketplaces. These are role-playing shopping experiences in which respondents are presented with made-up products in pretend stores. These hypothetical products are assembled and arranged on pretend shelves according to an experimental design , which is carefully constructed so that combinations of attributes are not correlated across products and the appearance of levels is roughly balanced across products. StatWizards' Experimental Design Builder automates and simplifies this task for you.
Phase 3. Recruit sample and generate questionnaire.
Once the shopping exercises are ready for insertion into a questionnaire, most projects add a set of standard questions to collect demographic and other relevant data. Often the questionnaire is divided into three parts, pre-DCM, DCM and post-DCM, where the before and after sections collect demographic and other information.
The questionnaire development process often includes a pre-test, during which the instrument is tested for clarity, length and efficacy. Generally this is followed by an editing and revision cycle that ultimately produces the field survey instrument.
At some point before phase 4, the researcher needs to develop a sampling frame* profiling the individuals to be recruited. For business purposes, this frequently represents potential customers, often stratified according to demographic or other characteristics. Be advised that sampling theory is a complex subject that lies beyond the scope of this document. Before drawing a sample, you should consult an expert and/or one of the many texts on the subject.
*Sampling Frame: The complete list of all sampling units (i.e., individuals, families, businesses) in the population to be studied.
Phase 4. Send to field.
After thorough testing and revision, a discrete-choice survey gets launched. Collection media can include any combination of written, fax-back or web-based procedures, as well as personal interviews or focus groups. Telephone interviews can also be used but are recommended only for relatively simple shopping exercises. Each of these approaches has different costs, response rates and efficacies associated with it. The choice between them depends on a number of considerations, not the least of which is the realism with which the data-collection scheme mimics how consumers make choices in the real world.
Phase 5. Set up data.
When data return from the field, the researcher must convert it to a form that a statistical computer program can accept. This is often a non-trivial task spanning hours or days, including time spent vetting the data and choosing variables to include in the modeling process. Once a data set has been created, the model estimation phase can proceed.
Phase 6. Estimate model.
A variety of very good statistical packages for estimating discrete-choice models exist, and advances in the state of the art continue at an astonishing pace. For this reason, the StatWizards Choice Suite does not provide a model estimation capability. Rather, the Data Setup Wizard organizes survey data to facilitate estimation using many available packages. That way, the analyst is free to choose which approach to employ.
Phase 7. Build simulator.
The final phase in a discrete-choice project is the one business people find most useful. That involves building a market simulator based on the statistical model estimated in phase 6. Excel makes a powerful platform for such a model, as the example below shows. StatWizards' Simulator Builder can build even a model as complicated as this one for you automatically.