Monday, September 17, 2012

Day 11 - Team I


Conjoint Analysis

Conjoint analysis is a statistical technique used in market research to determine how people value different features that make up an individual product or service. The objective of conjoint analysis is to determine what combination of a limited number of attributes is most influential on respondent choice or decision making. A controlled set of potential products or services is shown to respondents and by analyzing how they make preferences between these products

 Types of conjoint analysis

Most conjoint analysis studies carried out professionally use either Adaptive Conjoint Analysis (ACA) which relies on computer based interviewing, or Choice-based Conjoint. However, there are actually many "flavours" of conjoint analysis depending on the task at hand. 

Choice-Based Conjoint (CBC) is favoured academically and widely used for pricing and brand value studies, ACA for larger more marketing focused work. Other forms of conjoint incorporate Buy-your-own tasks or take the basic principles and extend them to create tailored conjoint designs for specific markets.

There are three key design elements for conjoint analysis. Firstly, breaking down a product or service into component attributes and levels. Then, the choice of how to present a series of descriptions (product profiles) and what method should be used to find out which are most preferred. And finally the method for the calculation of the value of each attribute and level to the market - the importance and utility scores, or part-worths.
The design decision of which to use depends on the number of attributes you have, the contact method and time available for the interview - for instance, on-line, face-to-face CAPI, or postal or paper-based interviewing. In addition, the different flavours of conjoint analysis each have differing characteristics in analysis. If data is available at an individual level as with Adaptive Conjoint Analysis, then it is simpler to develop market segmentation schemes. If the data needs to be more robust, perhaps because of the inclusion of attributes that may interact with each other, then Choice Based Conjoint analysis may be more relevant. If you run into conjoint analysis as a student, then it is most probably through the use of full-profile conjoint analysis. There are other options available depending on the task at hand.

Adaptive Conjoint Analysis - ACA
Adaptive Conjoint Analysis (ACA) is one of two most common methods for carrying out conjoint analysis. The benefits of ACA are that it allows for a large number of attributes (up to 30) and levels (up to 7 per attribute) to be used. However, ACA does require a computer-based interview and the large number of attributes means that it is common for an ACA interview to last 45 minutes or more. In addition, some of the methods it uses to simplify the task of working out utilities mean that some care is needed in choosing and designing the attributes in order to get reliable results.
Technically ACA is known as a hybrid technique as it contains elements of 'self-explication' followed by the trade-off tasks themselves. ACA itself is produced by Sawtooth Software and can be conducted face-to-face or on-line. Telephone use of ACA is difficult and paper-base questionnaires are not possible.

Choice Based Conjoint Analysis- CBC
The most common alternative to ACA is Choice-based conjoint (CBC). Although this uses the same over-arching principles as ACA, in design, implementation and calculation it is completely different.
Whereas ACA has respondents selecting from products described with two or three attributes, CBC shows full descriptions using all the attributes available. In addition, CBC can show more than just two "products" at the same time, together with a none-of-these option enabling more realistic choice decisions to be evaluated.
The limitation on the amount a respondent can absorb at a time, combined with the rapidly increasing number of "full-profile" combinations that are possible means that choice-based conjoint is typically limited to 5-7 attributes (in contrast to 25-30 for ACA).
An additional twist is that utilities and importances in CBC are calculated across a sample as a whole, whereas for ACA you get utilities and importances for each individual in the sample. To get at individual level estimations relies on techniques such as Hierarchical Bayes analysis to infer back individual values. However, combined with the lower number of attributes, choice-based studies require far shorter questionnaires (15-20 mins) and can be designed to be purely paper-based.
The advantages choice-based conjoint gives you are greater robustness of results - particularly for pricing work (although there are ways of getting around ACA's pricing limitations), combined with shorter and therefore less costly fieldwork. It is also favoured for its rigour academically. In applications such as pharmaceutical work it also enables comparisons with fixed products or fixed tasks enabling you to test new formulations against an existing gold standard.
The disadvantages are the lower number of attributes that are possible unless you move to more complex bespoke designs using partial profiles, and the lack of directly valued individual level utilities - although techniques such as hierarchical bayesian analysis seeks to remedy this by post-hoc simulation of individual level values. However, if you are looking to use conjoint analysis for clustering or segmentation you will need to be aware of the trade-offs needed to get individual level utility scores.

Discrete Choice Analysis
A more advanced form of choice-based conjoint is Discrete Choice Analysis (also known as "stated preference research"). DCA studies are particularly popular for transportation studies looking at modal choice - the preference between a train, car and airline for instance. The main difference from CBC is the inclusion of continuous variables such as price and time. This allows the ability to examine the varying costs of the ticket with varying times taken to travel and so to establish the value of time for the journey. This enables transport economists to make statements like "2cm extra leg room is worth 10 minutes longer journey time or £40 extra fare" or "an extra train every 15 minutes would encourage x% of car drivers to switch to the train".

Full profile Conjoint Analysis
An additional option that dates back a long time but that is still used is full profile conjoint analysis. Full-profile is the original form of conjoint and is still in use, though predominantly in the US it would appear. Like choice-based conjoint this uses a more limited number of attributes to describe the product or service, but sufficient cards or treatments are shown to one respondent to enable individual level utilities to be calculated. A fractional factorial design is used to specify a fixed set of profiles that need to be shown for analysis. The difficulty is that this does limit the number of attributes quite severely. However these old school studies are still popular for simple, non-computer-based conjoint projects and are most common for students learning about conjoint for the first time.

Other forms and formats
Recent developments in conjoint include the Adaptive Choice Based Conjoint method from Sawtooth. Which combines elements of a configurator, an adaptive element and choices. In addition we have our own dobney.com conjoint designer that allows for a range of other more bespoke research areas where traditional forms of conjoint analysis are lacking or where current designs can seem too difficult from a respondent point-of-view These include emotional association tasks, and repertoire purchasing (where someone is buying a bundle of products across a range of uses), volumetric measurement and improved choice displays such as using sliders or more interactive elements to encourage a fuller participation in the decision making process. Similarly, for international research the presentation of prices in different formats and currencies requires careful consideration.

Deciding which format of conjoint analysis to use
We often find that the choice of the type of conjoint analysis to use depends on a number factors and it's often difficult for someone new to conjoint analysis to visualise the options and in cases where there are 6-8 attributes, there can be several options and approaches. We often produce several versions and presentations for illustration to help clients see and understand which form and presentation version of conjoint to use, without compromising the final statistical quality of the survey.


Author
Pulkit Kabra
Team I

No comments:

Post a Comment