Joint Modeling of Multiple Data Sources

点击翻译按钮获取中文摘要

In recent years, joint statistical analyses of sources of data have gained traction in numerous fields of application, including animal population dynamics and human population health. This research program aims to advance the statistical methodology and applications for the joint modeling of multiple data sources. The first objective (O1) of this research is to address unexplored issues in integrated population modeling for the study of animal population dynamics. My team and I will develop continuous time integrated population models and address the questions of how to optimally allocate survey effort between the different surveys, how to deal with conflicting evidence between data sources and other assumptions breakdown. The second objective (O2) is to develop novel statistical methodologies for the analysis of disconnected networks in network meta-analysis where the purpose is to synthesize evidence across randomized studies to assess the relative efficacy or safety of treatments. With standard methods, the treatment effects between disconnected treatments cannot be estimated. To achieve this, my team and I will develop new modelling strategies including parametric and non parametric models for study-specific baseline treatment effects and a joint model that will use correlation between outcomes. The third objective (O3) is to develop statistical methodologies for the joint modeling of data sources in two new application areas. Specifically, my team and I will develop a novel methodology for estimating the size of yearly smolt runs when some of the daily data is missing due to trap failures on some days. Information will be shared across years through the use of rainfall covariates. We will also develop a novel methodology for estimating the size of homeless populations when the data is collected through a plant-capture scheme, and planted homeless may be unsure whether they were counted by the survey or not. We will develop a new model that shares evidence across multiple cities. In addition to addressing several important methodological issues for the joint analysis of multiple data sources, this research will allow, for the first time, to make important connections between the various literatures on joint analysis methods and highlight differences. By making the methodology readily available through software implementations, this work is expected to significantly promote more complex and realistic joint analyses in the natural sciences. The applications of the novel methodologies will have real world impact by influencing policymaking in Canada and worldwide, from wildlife management and conservation to drug review and more. The knowledge gained from this research will be crucial for conservation managers to understand and react to the effects of climate change. This research will undertake considerable training of HQP (8 students) which will significantly contribute to capacity building in scientific and statistical modeling in Canada.