DEVELOPEMTN OF A TIME-SERIES ANALYSES SOFTWARE PACKAGE FOR EPIDENIOLOGIC DATA

流行病学数据与环境数据是紧密联系在一起的。因此，有必要对它们进行深入的分析，而不是孤立的分析，需要对它们进行系统的收集和整理。由于这些数据的收集和整理是该项目最重要的一部分，我们必须建立良好的流行病学数据库。我们收集了许多疾病的每周数据，例如，横滨的哮喘，东京的风疹和流感等疾病，并将它们输入到每个数据库中。对几年的自然环境数据、空气污染物浓度数据、交通量数据等进行了校正。对采集的数据进行了相关分析、多元分析等时间序列分析。结果，我们了解了几种疾病的病原体数量的一些统计特性，并对某些疾病与环境因素的关系获得了一些有趣的发现。由于流行病学时间序列不一定是平稳的，因此数据处理的理论研究往往比较困难。因此，我们需要发展一些新的理论。我们致力于发展分析非平稳时间序列的新方法。本文提出了非平稳过程自由度的概念，并指出自由度的概念与熵的概念密切相关。进一步，我们推导出了时变有效带宽，它是时间序列非平稳谱扩展的定量度量。谱带宽度对评价时变光谱的性能起着重要的作用。

Epidemilogic data and environmental data are closely connected with each other. Therefore, it is necessary to analyze them simultaneouslt, not independently, and we need to collect and arrage them systematically. Since this data collection and arrangement are the most important part of the project, we have endeavored to maker good databae on epidemiolgic data. We collected weekly data of many disseses, for example, asthma in Yokohama, rubeola and some disease like influenza in Tokyo, etc., and put them into each database. We also corrected some natural environmental data of several years, air contaninant concetraion data, traffic volume data, etc. Collected data were analyzed with time series analysis programs, for example, correlation, analysis programs multivariate analysis programs, etx. As a result, some statistical properties of the number of partients of several diseases were known and we obtained some intersting findings on the relation between some dfiseases and environmental factors. Since epidemiologic time series is not necessarily stationary, the theoretical study of the data processing is often difficult. Therefore, we need to develop some new thories. We have made efforets to develop new methods to analyze a nonstatinary timesseries. We introduced new idea on the number of degrees of freedom of a nonstationary process and showed that this number of degrees of freedom is closely connected with the idea of ENTROPY. Further, we deduced time varying effective bandwidths which are quantitative neasures of the spread of ninstationary spectrum of o the time series. The banwidths play importnat rolls for evaluation of the property of the time verying spectrum.