Techniques and tools
Conjoint Modelling

Conjoint Modelling is an important technique for predicting the choices of products and services made by decision makers. The theory is set out below – this is an over-simplification but sets out the principles.

A product or service can be broken down into its various components or benefits (known as attributes in the conjoint terminology). Each of these attributes can take on discrete values (levels). For example, a laptop might be broken down in terms of:

• Processor Clock Speed (4 speeds – modest through to high))
• RAM (3 alternatives – modest, slow speed though large, high speed)
• Chassis type (4 models)
• Graphics Card (3 possible graphics card specs, plus no dedicated card)
• Battery life (2.5 hours, 3.5 hours, 4.5 hours)
• Brand (Brand A, Brand B, Brand C, Brand D)
• Price (5 levels)

Research is conducted which asks potential buyers to evaluate different combinations of these attributes as a complete product. Only a small subset of the possible product combinations (there are 8640 possible combinations for our example) need to be presented in order to give us enough information to predict the preference for any product, making this exercise practical from a research point of view. Typically, no more than 16 exercises are conducted. Note that this type of exercises is usually conducted using visual stimulus – either face-to-face or On-line. The data can also be collected over the telephone as long as the visual stimulus is available for viewing on-line during the interview.

Products can be presented one at a time, but more commonly several are presented in the same exercise and the respondent is asked to choose which one they would buy. There are other question types as we will see later.

The objective of the study might be either of the following:

• Understand the trade-offs made by potential buyers between the product attributes and establish the average likelihood of each attribute level being chosen
• Determine which product has maximum share of preference among buyers. In other words find the optimal product which is most likely to be the first choice for the larger group of buyers, within practical constraints (i.e. a product is unlikely to be offered with the maximum processing speed, RAM, battery life at the cheapest price)
• Determine actual likelihood of buying from a selection of different products, showing the share for each product
• Simulate the revenue demand for a product (consisting of any combination of the attribute values above) assume certain competitor scenarios

These objectives can be met via a conjoint model, which infers the relatively value of the various product attributes and levels to consumers from the choices they make in the exercises. It uses this information to make predictions about the choices that would be made between other hypothetical product combinations not presented in the exercises. In other words the model fills in the gaps for product choices which are not covered directly in the research.

The perceived value of different product attributes and levels to buyers is likely to differ in the real world across various subgroups of the population being sampled. The conjoint method employed by Logit Research – Latent Class Choice Modelling – allows these differences to be incorporated into the model. This gives the following additional benefits:

• A “needs based segmentation” can be derived which groups potential buyers according to their relative preferences for different attributes and levels. This occurs as part of the modelling process and provides a much more robust segmentation than post-hoc segmentation procedures performed using other methods
• A multilevel needs based segmentation can potentially be performed if respondents fall within a large enough numbers of groups e.g. regions
• Continuous factors can be included in the model which allow certain attributes and levels to vary in a continuum either over the sample or within specific needs segments
• Attributes which have the same value across the needs segments can be constrained to be equal, providing a simpler model (simpler models tend to be better at making predictions for new cases)
• The objectives spelt out earlier can be met for any subgroup of interest, using a combination of needs segments and continuous factors.

Structuring the Exercises

There are different ways available to collect the information for modelling. The method chosen depends on the objectives and nature of the product or service. A very efficient and increasingly popular method is to ask decision makers to choose which product they are most likely to buy and which they are least likely to buy. This gives an additional piece of information which can be used to help predict their first choice, giving a more robust model. However, if one of the objectives is to simulate real market share in a hypothetical competitor scenario then we would allow a “none of these” option and would typically only ask for “first choice” only.

For a frequently purchased service with different service levels such as a courier service, it might make more sense to ask decision makers to allocate 100 points across the available options, to represent the likely distribution of their volume of purchases.

In other instances we might prefer respondents to rate each individual product or service separately on a scale, rather than choose between competing options. In this case the model will predict the rating for a product, rather than whether it will be chosen over a competing product.

All of these options are possible when using the Latent Class approach to Conjoint. Note that no proprietary software is required at the data collection stage. In other words traditional data collection methods are employed – Face-to-face, CATI with on-line stimulus, Self Completion or On-line survey – without the need for a computer assisted conjoint interview.

Conjoint Outputs

All models are summarised in presentation ready format using PowerPoint and Excel. A custom-built excel based simulator for testing various product and competitor scenarios is also provided. Some examples of Conjoint output are shown below:

Please contact Gary Bennett for further information (garyb@logitresearch.com).

< Back to Techniques and Tools Home Page