Hierarchical models have received wide attention in a variety of fields; for example, social and behavioral sciences, agriculture, education, medicine, healthcare studies, and marketing. A general theory of optimal design for efficient estimation of hyperparameters in hierarchical linear models is developed. A statistical design optimality criterion is proposed for this setting, and optimal design structures are investigated under various forms of the variance-covariance matrix of the random effects. It is shown that, when the variance-covariance matrix is assumed to be diagonal, that is, when the random effects are assumed to be independent, level-balanced orthogonal designs are optimal if they exist. When the random effects are not independent, however, orthogonal designs cease to be optimal. The structure of the optimal continuous design in each scenario is derived. Hierarchical models have been proven powerful in addressing a broad class of problems that involve the learning of effect sizes and the drivers of maximal or minimal effect sizes. The modeling of the "level effect" proposed in this dissertation is such an example. The "level effect" refers to the phenomenon in market research studies that consumer preference sensitivity to a product attribute is affected by the set of attribute levels displayed. Standard conjoint analysis models in marketing research do not adjust for the variety of attribute levels offered and, consequently, can not predict consumer buying behavior well in a different context. A hierarchical Bayes approach is proposed that models the individual consumer behavior and, by incorporating a number of ideas from the Psychology literature, models the level effect through hyperparameters. To evaluate the effectiveness of the proposed model and the effect of different survey designs, web-based survey studies are conducted on credit card products. The optimal design theory developed for efficient estimation of hyperparameters is applied to select the alternative survey designs. The proposed model offers a good fit to the data and predicts consumer preference well in a different context. Furthermore, designs with higher efficiency lead to better estimation and prediction accuracy in general, confirming the optimal design theory.