Latent-Gaussian Spatio-temporal models for complex problems

“大数据”时代的一个特点是，越来越多的大量空间信息被例行收集并储存在行政数据库中。这使得研究人员能够在比以前更精细的空间尺度上回答问题，例如检查城市内癌症风险的变化，而不是城市之间。这些新的数据来源和伴随而来的研究问题需要在空间统计领域取得进展，因为非常适合50个卫生区域模型的方法在应用于10 000个人口普查区域的数据时往往效果不佳。这项研究计划将以最近的研究为基础，推进与以下方面有关的统计方法：*- 以高空间分辨率预测健康结果病例（人口普查区、邮政区）;*- 将统计模型拟合到混合空间分辨率的空间数据（即点位置、邮政编码、普查区域）;*- 使用行政健康数据来解决目前需要临床记录的问题;以及 *- 简化用于拟合时空模型的统计软件。将制定新的统计方法，以解决上述每个领域的重要未决问题。

One feature of the `big data' era is that increasingly large amounts of spatial information are routinely collected and stored in administrative databases. This has enabled researchers to answer questions at a finer spatial scale than was previously possible, for instance examining variation in cancer risk within a city as opposed to between cities. These new data sources and the research questions which accompany them have required advancements to be made in the area of Spatial Statistics, as methods which were well suited to models for 50 health regions often work poorly when applied to data from 10,000 census regions.***This research plan will build on recent research to advance statistical methodology related to:***- forecasting cases of a health outcome at a high spatial resolution (census tracts, postal regions);***- fitting statistical models to spatial data at mixtures of spatial resolutions (i.e. point locations, postal codes, census regions);***- using administrative health data to address questions currently requiring clinical records; and***- simplifying statistical software for fitting spatio-temporal models.***New statistical methodologies will be developed to address important outstanding issues in each of these areas.**