1.Whenever the number of attribute level combinations to search makes it feasible to run Exhaustive Search, this will guarantee finding the globally optimal solution (within the domain included in your research study). For example, searching across 5 attributes each with 4 levels leads to 4 raised to the 5th power number of product alternatives to simulate, or just 1024 scenarios.
2.When the number of combinations to search exceeds what is feasible with Exhaustive Search, it might make sense first to run Grid Search, starting with the default 10 passes and later with even more passes (say 50 or 100) to see if an even better result may be found. If the same answer is always obtained, that is probably the optimum. If not, then the experience obtained should permit reducing the size of unknown domain substantially, so it may become feasible to run Exhaustive Search in that reduced domain.
3.Genetic Search is a trusted algorithm that has the potential of finding good solutions when conditions limit the capabilities of the Grid approach, such as when the response surface is very irregular with multiple peaks. It takes much longer, but in certain cases achieves superior results.
4.For both Genetic and Exhaustive Search the user can specify the number of solutions that should be reported, and that many of the best solutions are displayed in the report window. This capability seems most likely to be useful when the response surface has multiple peaks. In that case the researcher may be able to use expert opinion to evaluate a number of near-optimal solutions.
5.Often, the business goal is best served by considering tradeoffs among multiple objectives, such as profit & share of preference. Ideally, one would prefer a solution that near-maximized profit while also near-maximizing share of preference. In the case of Exhaustive Search or Genetic Search, you can specify multi-objective searches (you may select multiple goals). The software will display the results both in a table and graphically so you can examine the characteristics of multiple solutions and the tradeoffs among multiple goals.
6.We recommend Randomized First Choice as the preferred method in nearly all circumstances. If a much faster method is required, the First Choice method has the advantage of not inflating shares for similar products, but its estimated market shares will probably be too extreme and their standard errors will be somewhat larger. The Share of Preference is also extremely fast, though it is subject to IIA concerns. Good compromises include: a) narrowing the range of search via Randomized First Choice operating on many fewer than the default number of iterations (such as using just 10 or 50 iterations per respondent) followed by full Randomized First Choice using the default number of iterations within a much smaller search space, or b) Share of Preference with Top-N to help reduce IIA concerns.
7.We have seen situations where searches seem to capitalize on “reversals” in part worths to produce less useful solutions. It may, therefore, make sense to use part worths that have been constrained during estimation to avoid reversals.