Conjoint Analysis and Choice Modelling
(Note that in USA and some other countries, the spelling is choice modeling)
These are techniques for decomposing market preferences into the contribution of each of a number of attributes. They can be used for prediction in many situations where cross-section or time series regression models are ineffective.
Imagine trying to understand the relative contribution to airline choice of factors such as brand, price, meal quality, entertainment, reward points, facilities at airport lounge, and so on. Often, there is little difference between airlines on these factors so cross-sectional modelling yields imprecise estimates of the impact of only some factors. Even if we include time series data (and this brings with it additional confounding factors) there is not enough variation in the data to use regression analysis. And what if we want to quantify the impact of a new initiative that has never been tried - such as in-flight internet access or a lower fare than has been available in the past?
Conjoint analysis and choice modelling address these problems by using an experimental design to vary the attribute levels across hypothetical products which can be described to survey respondents.
Two such products are set out below as a simplified example.
In a conjoint analysis study, each respondent ranks a set of such products into preference order. This can be used to infer each respondent's decision rule - or the "part-worth" of each attribute level. This can also be used for segmenting respondents on the basis of benefits sought.
In a choice modelling study, each respondent evaluates a "set" of products in indicates which would be chosen. This is repeated several times for different choice sets. The aggregate choice frequencies can be modelled (usually multinomial logit) to infer the relative impact of each attribute level on choice. Market shares can be predicted across a range of scenarios using this model.
Interested in forecasting? See my book Forecasting: the essential skills