Statistical Methodology for Multiple Events in Time and Space

时空多事件的统计方法

• 批准号: RGPIN-2018-04799
• 负责人: Horrocks, Julie
• 金额: $ 1.31万
• 依托单位: University of Guelph
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2019
• 资助国家: 加拿大
• 起止时间: 2019-01-01 至 2020-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=676947
• 关键词: Statistical; Methodology; Multiple; Events; Time

My proposed research can be divided into 4 areas. ***1. Time-to-Event Data,***2. Relative Abundance and Distribution of Animals,***3. Text Analysis,***4. Count Data with Excess Zeros.******1. Today we collect masses of data almost continuously on patients in intensive care units, plants in precision agriculture, and drivers participating in insurance surveillance programs, etc. The purpose can be to monitor patients for adverse events, to determine optimal watering regimes for trees and agricultural plants, to set insurance rates for drivers based on their driving patterns. As the frequency of data collection increases, we have the opportunity to carefully model the effects of explanatory variables. My work will develop novel models for the effects of explanatory variables that change over time on the risk and cost of an event and investigate optimal timing of interventions. ******In other situations, we are unable to monitor continuously. For instance consider progression of a disease such as cancer. One approach is to use multi-state models with states representing stages of cancer. We do not know exactly when the progression from stage 1 to stage 2 occurred, but have information only at intermittent monitoring times, so that misclassification at intervening time points is likely. The monitoring times may be informative about progression, as patients may make an appointment when feeling unwell. My work will develop novel multi-state models with misclassification, which allow for new distributions, clustering of individuals in families, and informative monitoring times and dropout. This work is critical as better modelling of disease history will allow better treatment of patients. ******2. Statistical models for species distributions typically focus on one species at a time. I propose to develop novel multiple-species models, which exploit the correlation and interaction between species to get better estimates of distributions, and to develop methods for modelling changes over time and simultaneous selection of important explanatory variables from a large number of candidates. Modelling species accurately is important for conservation and crucial in the face of climate change.******3. Methods currently exist for classification of corpora (bodies of text) into one of several groups. However methods for testing for statistical differences between groups are not well-developed. I will develop methods for testing for differences between word frequencies. This is challenging because of the large number of candidate words. ******4. Excess-zero count models. Data on counts are ubiquitous, including number of seizures in a two week period, number of near-misses in traffic, etc. Sometimes the data have an excess of zeros over that predicted by common models such as the Poisson and Negative Binomial. I will develop novel models for gamma and Weibull interarrival times with excess zeros.**

我提出的研究可以分为四个方面。*1.到事件的时间数据，*2.动物的相对丰度和分布，*3.文本分析，*4.用过多的零计算数据。*1.今天我们几乎不间断地收集关于重症监护病房的患者、精准农业中的植物和参与保险监控计划的司机等的大量数据。其目的可以是监控患者的不良事件，确定树木和农业植物的最佳浇水制度，根据司机的驾驶模式为司机设定保险费率。随着数据收集频率的提高，我们有机会仔细模拟解释变量的影响。我的工作将开发新的模型，用于解释变量随时间变化对事件风险和成本的影响，并调查干预的最佳时机。*在其他情况下，我们无法持续监控。例如，考虑一种疾病的发展，比如癌症。一种方法是使用多状态模型，状态代表癌症的不同阶段。我们不知道从第一阶段进展到第二阶段的确切时间，但只有在断断续续的监测时间才有信息，因此在中间的时间点很可能发生错误分类。监测时间可能会提供有关进展的信息，因为患者可能会在感到不适时预约。我的工作将开发具有错误分类的新的多态模型，该模型允许新的分布、家庭中的个人集群以及提供信息的监测时间和辍学。这项工作至关重要，因为更好地模拟病史将允许对患者进行更好的治疗。物种分布的统计模型通常一次只关注一个物种。我建议开发新的多物种模型，利用物种之间的相关性和相互作用来获得更好的分布估计，并开发方法来模拟随时间的变化和从大量候选对象中同时选择重要的解释变量。准确地建立物种模型对于保护和应对气候变化至关重要。*3.目前存在将语料库(正文)归类为若干类别之一的方法。然而，检验不同群体之间统计差异的方法并不完善。我将开发测试词频差异的方法。这是具有挑战性的，因为候选单词的数量很大。*4.超零计数模型。关于计数的数据无处不在，包括两周内的癫痫发作次数、交通中险些错过的次数等。有时，这些数据比泊松和负二项等常见模型预测的数据多出零。我将开发伽马和威布尔到达间隔时间的新模型，这些时间带有多余的零。**