Description
Models each alternative's share of a category as its attraction divided by the total attraction of all alternatives, with attraction a multiplicative function of its own attributes.
Shares lie between 0 and 1 and sum to 1 across competitors, so they cannot be modelled one at a time with ordinary regression. The attraction model gives each alternative an attraction equal to a constant times its explanatory variables (price, distribution, advertising, features) raised to alternative-specific powers, with dummy variables entering as exponentials, and sets its share equal to its attraction over the summed attraction of all alternatives. This guarantees logically consistent shares and lets a change in one competitor's mix redistribute share across all. Estimation picks a benchmark alternative, divides each share by the benchmark's share and takes logs, giving J - 1 linear equations estimated jointly with correlated errors; only differences from the benchmark's parameters are identified. Unlike discrete choice models, which work on individual choices, the attraction model works on aggregate share time series.
Attributes
- Discipline
- Product Analytics & Strategy
- Loss type
- Unmigrated
- Stage
- Strategy
- Data
- Quantitative
- Effort
- Weeks
Steps
- Assemble time series of shares and of positive-valued explanatory variables for every alternative (exponentiate dummies; avoid rates of change that can be negative).
- Choose a benchmark alternative.
- Form the log ratio of each alternative's share to the benchmark's share and the corresponding log ratios of explanatory variables.
- Estimate the J - 1 linear equations jointly, allowing correlated errors across equations.
- Interpret the identified parameters as differences from the benchmark, and simulate shares under changed prices, features or promotion for one or more alternatives.
Sources
- Quantitative Models in Marketing Research (Franses & Paap, 2001)
Also known as
- Attraction model
- MCI-type market share model
- Multiplicative competitive interaction model
Categories
- Analytics & Behaviour
- Product Analytics & Strategy