基于多阶模型理论的非抽样误差研究

发布时间：2019-02-14 07:36

【摘要】：非抽样误差是统计调查中除抽样误差以外,由于各种原因而引起的误差,对调查结果的影响非常大。本文结合中国统计调查的实际,全面研究了概率抽样调查中非抽样误差产生的原因、测度方法及估计与调整方法。文章探讨了非抽样误差产生的制度、文化和应用等方面的原因,设计了非抽样误差在统计调查总误差中所占比重的测度方法,研究了定量测度非抽样误差的方法路径,应用多阶模型理论设计了概率抽样调查中出现测量误差和无回答误差时的估计量以及对传统估计结果的调整方法。本文的研究初步形成了研究概率抽样调查中非抽样误差的内容体系和方法体系。在理论研究方面,本文以多阶模型为方法体系的核心,并辅以路径分析等其它研究方法。本文探索了将多阶模型应用于非抽样误差研究的思路。由多阶段抽样调查方式调查获得的数据具有多阶特征,适合采用多阶模型方法进行研究。当出现测量误差时,本文吸收了多阶模型中空模型的建模思想,设计了分层抽样层均值和总均值方差的估计量。这部分研究还引入了测量可靠性指标,研究了可靠性对估计结果的影响。对于多阶段调查数据中的多变量关系分析,本文讨论了利用多阶模型对存在测量误差时多变量关系的估计与调整方法。当出现无回答误差时,本文利用了多阶模型分析多阶段抽样调查数据的原理,采用经验加权方法估计分层抽样中各层均值的估计量。本文还尝试应用虚拟变量和多阶模型结合的方法研究出现无回答时的估计问题。同时,本文还采用了其它研究方法。如在研究基于设计的非抽样误差测度方法时,本文以均方误差的定义式为基础,论证了非抽样误差在统计调查总误差中所占比重的测度方法,得出了该比重所在区间的下限,并编制了该比重与回答率和测量可靠性不同取值的对应关系表；在研究基于模型的非抽样误差测度方法时,文章运用路径分析法从非抽样误差产生的根源入手研究了包括多指标和单指标统计调查中非抽样误差的测度。此外,本文还推导了测量误差方差的定量测度公式,设计了存在测量误差时分层抽样中各层均值的方差估计量。在应用研究方面,本文采用2007年广东省城镇住户调查的11市和7县、区的1600个家庭的消费数据对主要理论进行了实证检验,形成了一套利用多阶模型研究非抽样误差的应用体系。实证结果表明：当组间差异显著时,应该运用多阶模型进行数据分析。在恰当设计的程序中,多阶模型能够比传统方法更好地“拟合”样本数据的特征,实现非抽样误差的测度与调整。实证分析展示了多阶模型视角下非抽样误差的研究路径,给出了定量测度非抽样误差的模拟案例。
[Abstract]:Non-sampling error is the error caused by various reasons except sampling error in the statistical survey, which has a great influence on the investigation result. In this paper, the causes, measurement methods, estimation and adjustment methods of non-sampling errors in probabilistic sampling surveys are studied. This paper discusses the causes of the system, culture and application of the non-sampling error, designs the measurement method of the proportion of the non-sampling error in the total error of statistical investigation, and studies the method path of quantitative measurement of the non-sampling error. Based on the theory of multi-order model, the estimation of measurement error and no response error in probabilistic sampling survey and the adjustment method of traditional estimation results are designed. In this paper, the content system and method system of non-sampling error in probabilistic sampling survey are preliminarily formed. In the theoretical research, this paper takes the multi-order model as the core of the method system, and complements other research methods such as path analysis. In this paper, the idea of applying multi-order model to the study of non-sampling error is explored. The data obtained from multi-stage sampling survey have multi-order characteristics and are suitable to be studied by multi-order model method. When the measurement error occurs, this paper absorbs the modeling idea of the multi-order model hollow model, and designs the estimators of the stratified sampling layer mean and the total mean variance. In this part, the reliability index is introduced and the influence of reliability on the estimation results is studied. For the multivariable relation analysis of multistage survey data, this paper discusses the estimation and adjustment method of multivariable relationship in the presence of measurement error by using multi-order model. When there is no response error, the principle of multi-stage sampling data analysis is used in this paper, and the empirical weighting method is used to estimate the mean value of each layer in stratified sampling. This paper also attempts to use the method of virtual variable and multi-order model to study the estimation problem when there is no answer. At the same time, this paper also adopts other research methods. For example, in the study of the non-sampling error measurement method based on design, based on the definition of mean square error, this paper demonstrates the measurement method of the proportion of non-sampling error in the total error of statistical investigation, and obtains the lower limit of the proportion between the regions in which the proportion is located. At the same time, the corresponding relation table between the specific gravity, the response rate and the measurement reliability is worked out. In this paper, the non-sampling error measurement method based on the model is studied, and the non-sampling error measurement including multi-index and single-index statistical survey is studied by the path analysis method from the root of the non-sampling error. In addition, the quantitative measurement formula of measurement error variance is derived, and the variance estimator of the mean value in stratified sampling with measurement error is designed. In the aspect of applied research, this paper uses the consumption data of 1 600 households in 11 cities and 7 counties and districts of Guangdong Province in 2007 to make an empirical test on the main theories. A set of application system is formed to study the non-sampling error using multi-order model. The empirical results show that when the differences between groups are significant, multi-order model should be used for data analysis. In the properly designed program, the multi-order model can better "fit" the characteristics of the sample data than the traditional method, and realize the measurement and adjustment of the non-sampling error. The empirical analysis shows the research path of non-sampling error from the perspective of multi-order model, and gives a simulation case of quantitative measurement of non-sampling error.
【学位授予单位】：暨南大学
【学位级别】：博士
【学位授予年份】：2011
【分类号】：C811

【参考文献】