Efficiency and uncertainty measurement in microsimulations

微观模拟中的效率和不确定性测量

• 批准号: 465374470
• 负责人: Professor Dr. Ralf Münnich
• 金额: --
• 依托单位: Deutsche Statistische Gesellschaft (DStatG)
• 依托单位国家: 德国
• 项目类别: Research Units
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/465374470?language=en
• 关键词: Efficiency; uncertainty; measurement; microsimulations

The MikroSim base population is a (partially) synthetic, close-to-reality population of Germany, containing more than 80 million individuals in approximately 40 million households in the year 2011. Using individual transition probabilities, the whole population can be projected into the future. The projection of such a large number of individual observations enables detailed analyses, i.e.\ analyses with a close regional or contextual focus. At the same time, it causes an enormous computational effort. A more wide-spread use of this important analysis tool can be fostered by reducing the necessary computation time.The complexity of dynamic microsimulation models, that originates from the nature of the investigated research questions, leads to a multitude of potential influences on the precision of analysis results. While it is common practice in the social sciences to state measures of precision with estimators, like confidence intervals, this is typically neglected in the context of microsimulations. The reasons for this are the lack of available methods for the systematic measurement and combination of all relevant sources of uncertainty on the one hand, and the additional computational burden associated with such a measurement on the other hand. However, in order to be able to judge the validity of results, the measurement of their precision is also needed in microsimulations.Therefore, the main research focus within this project is on efficiency and uncertainty measurement in microsimulations which are closely related to each other. First, the potential for efficiency improvements in microsimulations by using samples from the base population for projections is investigated. Special attention is paid to the suitability of different complex sampling designs for the projection of target values on different levels of regional and content-related granularity. Then, the diverse sources of uncertainty in the simulation process are identified. Using different statistical methods, these sources are implemented into the simulation structure as an aggregated uncertainty process. The latter can then be used to construct quasi confidence intervals for simulation results that facilitate the judgement of their validity.The methods that are developed within this project are directly applied to microsimulations of the other projects within the Research Unit. Additionally, they are provided to researchers in an application-oriented guideline and implemented as a tool within the prototype of a simulation data centre.

MikroSim基础人口是德国的（部分）合成的、接近现实的人口，2011年包含约4000万个家庭中的8000多万人。使用个体转移概率，可以预测整个种群的未来。如此大量的单个观测的投影使得能够进行详细的分析，即分析与密切的区域或上下文的重点。与此同时，它会导致巨大的计算工作量。通过减少必要的计算时间，可以促进这一重要分析工具的更广泛的使用。动态微观模拟模型的复杂性源于所调查研究问题的性质，导致对分析结果精度的许多潜在影响。虽然在社会科学中，用估计量（如置信区间）来衡量精度是常见的做法，但在微观模拟的背景下，这通常被忽略。其原因一方面是缺乏系统测量和综合所有相关不确定性来源的可用方法，另一方面是与这种测量相关的额外计算负担。然而，为了能够判断结果的有效性，在微观模拟中还需要测量其精度，因此，本项目的主要研究重点是相互密切相关的微观模拟中的效率和不确定性测量。首先，微观模拟的效率提高的潜力，通过使用样本的基础人口的预测进行了研究。特别注意的是，不同的复杂的抽样设计的目标值的预测在不同层次的区域和内容相关的粒度的适合性。然后，在模拟过程中的不确定性的各种来源进行了识别。使用不同的统计方法，这些来源作为聚合不确定性过程被实施到模拟结构中。后者则可以用来构建准置信区间的模拟结果，便于判断其validity.The的方法，在这个项目中开发的直接应用到微观模拟的研究单位内的其他项目。此外，他们提供给研究人员在一个面向应用的准则，并作为一个工具内的模拟数据中心的原型。