In this work the primary objective is to maximize the precision of the maximum likelihood estimate in a linear regression model through the efficient design of the experiment. One common measure of precision is the unconditional mean square error. Unconditional mean square error has been a primary motivator for optimal designs; commonly, defined as the design that maximizes a concave function of the expected Fisher information. The inverse of expected Fisher information is asymptotically equal to the mean square error of the maximum likelihood estimate. There is a substantial amount of existing literature that argues the mean square error conditioned on an appropriate ancillary statistic better represents the precision of the maximum likelihood estimate. Despite evidence in favor of conditioning, limited effort has been made to find designs that are optimal with respect to conditional mean square error. The inverse of observed Fisher information is a higher order approximation of the conditional mean square error than the inverse of expected Fisher information [Efron and Hinkley (1978)]. In light of this, a more relevant objective is to find designs that optimize observed Fisher information. Unlike expected Fisher information, observed Fisher information depends on the observed data and cannot be used to design an experiment completely in advance of data collection. In a sequential experiment the observed Fisher information from past observations is available to inform the design of the next observation. In this work an adaptive design that incorporates observed Fisher information is proposed for linear regression models. It is shown that the proposed design is more efficient, at the limit, than any fixed design, including the optimal design, with respect to conditional mean square error.
An Online Meta-Level Adaptive Design Framework with Targeted Learning Inference: Applications to Evaluating and Utilizing Surrogate Outcomes in Adaptive Designs
Division of Biostatistics, University of California, Berkeley(加州大学伯克利分校生物统计系);Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Center(弗雷德·哈钦森癌症中心疫苗与传染病部);Department of Biostatistics, University of Washington(华盛顿大学生物统计系)
Efficient Inference after Directionally Stable Adaptive Experiments
University College London(伦敦大学学院);Gatsby Computational Neuroscience Unit, University College London(伦敦大学学院计算神经科学单元);Cornell University(康奈尔大学);Netflix;Google DeepMind(谷歌DeepMind);Rutgers University(罗格斯大学)