问卷分割设计与分析技术研究

Split Questionnaire Design (SQD) is an efficient technique in survey study for respondent burden reduction by splitting the questionnaire into subsets of components for different individuals randomly. Intensive existing discussions have proven that the SQD can significantly decrease the survey cost and enhance the quality of data. Many challenging topics of SQD are still remained, such as the design of assignment procedure for components, the treatment for planned missing data, the computation issue for massive dataset, and so on. In this study, we propose an adaptive balanced Split Questionnaire Design (abSQD) by simultaneously introducing the balance of feature covariates among subsamples for different components and the prior knowledge of importance of components into the probability of respondent assigning procedure. The proposed randomization method can increase the representativeness of survey result by making the subsamples of different components comparable on the feature covariates. Additionally, it also can assign more samples adaptively to the components with larger variation which is good for the efficiency of estimator in inference. To deal with the planned missing data in the data collected by abSQD, we study estimating methodologies and their corresponding assumptions and properties by either doubly robust estimating equations and weighted model averaging. Two important extensions of the proposed estimating methods are discussed for the increasing volume of massive survey and the high dimensionality of dataset. Firstly, two statistical distributed algorithms for the increasing volume of massive surveys, the Bag of Little Bootstrap and the Subsample Double Bootstrap, are discussed for modification against the data structure in abSQD. To clarify the advantages on computational performance and the corresponding data conditions, numerical analysis would be conducted for the comparisons of the two algorithms in different scenarios. Secondly, structural variable selection and robust variable selection modeling are discussed for the data issues in practice. Both theoretical and numerical analysis would be discussed. Finally, two illustrative examples of "Study on the Customer Satisfaction Index for Insurance Industry" and "Study on the Online Video Customer" are discussed for the reference to applied researchers about the proposed abSQD and its analytical methodologies.

问卷分割设计解决由问卷长度导致的调查负担过重问题，达到降低调查成本、提高数据质量之目的。本研究针对其当前研究“问题组样本代表性缺乏保障”、“随机化过程中先验知识利用不足”、“缺失数据处理方法单一”、“未考虑海量、高维数据特征”的挑战，提出自适应平衡分割问卷设计及分析方法。本研究在问卷分割基础上利用特征变量信息和先验知识构建自适应平衡随机化设计，讨论其在样本自适应调节、特征变量平衡等方面的应用效果与理论性质；针对该机制下由设计产生的缺失数据，分别从基于“碎片化”观测数据的参数估计方法、大规模“碎片化”数据的统计分治算法和高维“碎片化”数据的变量选择模型三个角度展开研究，分别探索模型构建、参数估计、算法设计与有效性评价问题。本项目选择“人身险行业满意度调查”和“视频用户调查”作为示例探讨上述方法的适用性、可行性与解释性，提炼在实证研究中的研究思路与步骤，为管理科学研究者的量化研究提供有效工具。

问卷分割设计解决由问卷长度导致的调查负担过重问题，达到降低调查成本、提高数据质量之目的。本研究针对其当前研究“问题组样本代表性缺乏保障”、“随机化过程中先验知识利用不足”、“缺失数据处理方法单一”、“未考虑海量、高维数据特征”的挑战，提出自适应平衡分割问卷设计及分析方法。本研究在问卷分割基础上利用特征变量信息和先验知识构建自适应平衡随机化设计，讨论其在样本自适应调节、特征变量平衡等方面的应用效果与理论性质；针对该机制下由设计产生的缺失数据，从“碎片化”观测数据的参数估计方法出发展开研究，分别探索模型构建、参数估计、算法设计与有效性评价问题。本项目选择多个调查数据作为示例探讨上述方法的适用性、可行性与解释性，提炼在实证研究中的研究思路与步骤，为管理科学研究者的量化研究提供有效工具。..