Inference from survey data and small area estimation

抽样调查被广泛用于提供各种重要主题的可靠统计数据(估计数)(例如，每月失业率和人们健康状况的衡量标准)。在评估其质量时，还需要衡量与这些统计数据相关的可变性。常用的测量方法包括标准误差和可信区间。调查数据也被广泛用于研究变量之间的关系和检验假设的有效性。调查越来越多地被用来提供制定政策和计划所需的小范围统计数据。我的研究计划将重点放在抽样调查的所有方面。我将特别研究以下问题：(A)基于特定地区数据的传统小地区估计不可靠，因为小地区的样本量较小。因此，利用相关地区的信息进行间接估计是为了增加可靠性。我在这个主题上做了大量的工作，特别是基于模型的估计和相关的标准误差。我建议继续这项工作，重点放在现实模型、测量误差、缺失数据和其他特征上。(B)调查中经常出现项目无反应，填补是处理缺失数据的常用方法。(C)纵向调查被广泛用于研究潜在的动态和获得更有效的估计。过去，我曾研究过考虑到复杂调查设计特征的横断面调查方法，如聚类法和不等权重法。我将把这些方法推广到纵向调查。这些结果将应用于加拿大统计局进行的纵向调查数据。

Sample surveys are widely used to provide reliable statistics (estimates) on a variety of important topics (e.g., monthly unemployment rates and measures of health status of people). Measures of variability associated with these statistics are also needed in assessing their quality. Commonly used measures include standard errors and confidence intervals. Survey data are also extensively used to study relationships among variables and to test the validity of hypotheses. Surveys are increasingly used to provide small area statistics which are needed in formulating policies and programs.  My research program will focus on all the above aspects of sample surveys. In particular, I will study the following problems:(a) Traditional estimates for small areas, based on area specific data, are not reliable due to small sample sizes in small areas.  Indirect estimates that utilize information from related areas are therefore used to increase the reliability. I have done extensive work on this topic, particularly model-based estimates and associated standard errors. I propose to continue this work, focussing on realistic models, measurement errors, missing data and other features.(b) Item non-response often occurs in surveys and imputation is commonly used to handle missing data. Methods of making valid inferences under imputation will be studied.(c) Longitudinal surveys are widely used to study underlying dynamics and to get more efficient estimates. In the past, I have studied methods for cross-sectional surveys that take account of complex survey design features such as clustering and unequal weights. I will extend these methods to longitudinal surveys. These results will be applied to data from longitudinal surveys conducted by Statistics Canada.