Model Misspecification in Multiattribute Parameter Estimation
指出联合分析中常用的主效应模型在效用函数复杂时预测效果差，并提出折中设计可同时估计主效应和选定的双向交互作用，提供更通用的实验方案。
Most applications of conjoint analysis have emphasized main-effects models, largely because fewer data points are needed to fit that type of model at the individual level. The authors suggest that such simplifications can lead to poor predictions when the underlying utility functions depart from the simplicity of a main-effects model. They also show how compromise designs, which allow orthogonal estimation of selected two-way interactions (as well as main effects), can provide a more general experimental design in cases where a specified set of two-way interactions is suspected.