死因监测整群抽样设计方案的比较研究

发布时间：2018-11-14 19:50

【摘要】： 死因监测工作是了解居民死亡水平和死因顺位,掌握居民健康影响因素,为政府制定卫生政策、评价卫生工作质量和效果的科学依据,也是研究人口自然变动规律的一个重要内容。其中死因监测点的确定是一个首要的问题。本研究以陕西省死因监测点的确定为例,以陕西省107个县(市,区)级单元作为抽样框架,进行以下几部分的研究: 1.制定不同的抽样设计方案。结合中国实际情况,引入了不等概率抽样方法,建立可行的四种抽样方法,即:完全随机整群抽样、分层整群抽样、不等概率整群抽样、不等概率分层整群抽样,分别以总人口的5%、10%、15%作为抽样比例,组合成为十二种抽样方案,每个方案都进行计算机重复抽样100次。 2.计算不同抽样方案的抽样精确度,结果表明:完全随机整群抽样抽样比例大于10%就可对总体有较好的代表性;分层整群抽样抽样比例大于15%就可对总体有较好的代表性;不等概率整群抽样抽样比例大于15%就可对总体有较好的代表性;不等概率分层整群抽样抽样比例大于15%就可对总体有较好的代表性。 3.计算并比较不同抽样方案的平均抽样标准差。结果表明:地区生产总值均数的标准差:1)抽样比例除5%外,随着抽样比例的增大(从10%到15%),地区生产总值标准差的均数反而减小。2)不同的抽样方法中,地区生产总值标准差的均数以不等概率分层整群抽样最小。死亡率均数的标准差表明:不同的抽样方法中,以不等概率分层整群抽样的标准差最小。 4.计算并比较不同抽样方案的设计效率,结果表明:地区生产总值:1)完全随机整群抽样中:不同抽样比例的地区生产总值设计效率的均数是相同的,都是1;除抽样比例为5%,随着抽样比例的增大(从10%到15%),地区生产总值设计效率的均数是减小的。2)不同的抽样方法中,以不等概率分层整群抽样的地区生产总值设计效率的均数最小。死亡率:1)完全随机整群抽样中,不同抽样比例的死亡率设计效率的均数是相同的,都是1;除抽样比例为5%,随着抽样比例的增大(从10%到15%),死亡率设计效率的均数是减小的。2)不同的抽样方法中,以不等概率分层整群抽样的死亡率设计效率的均数最小。 5.提出最佳抽样方案是不等概率分层整群抽样。本研究的主要创新点主要包括以下三点:(1)提出并将不等概率抽样应用于有明确抽样框架的总体的抽样研究中。(2)重复抽样计算并比较了每种抽样方案的平均抽样误差、抽样精确度及设计效率。(3)评价并提出了有明确抽样框架的总体的最佳抽样方案是不等概率分层整群抽样。
[Abstract]:The cause of death surveillance is the scientific basis for understanding the level and order of death, mastering the influencing factors of residents' health, making health policy for the government, and evaluating the quality and effect of health work. It is also an important content of studying the law of natural change of population. Among them, the determination of the cause of death monitoring point is a primary problem. In this study, the determination of cause of death monitoring points in Shaanxi Province as an example, with 107 county (city, district) units as sampling frame, the following parts of the study: 1. Develop different sampling schemes. According to the actual situation in China, this paper introduces the unequal probability sampling method, and establishes four feasible sampling methods, that is, complete random cluster sampling, stratified cluster sampling, unequal probability stratified cluster sampling. Taking 5 / 10 / 15% of the total population as the sampling ratio, it is combined into twelve sampling schemes, each of which is sampled 100 times by computer. 2. The sampling accuracy of different sampling schemes is calculated. The results show that the complete random cluster sampling ratio is more than 10%, the stratified cluster sampling ratio is more than 15%, and the stratified cluster sampling ratio is better than 15%. When the proportion of unequal probability cluster sampling is more than 15%, it can be better representative of the whole population, and the proportion of unequal probability stratified cluster sampling more than 15% can be better representative of the whole population. 3. The average sampling standard deviation of different sampling schemes is calculated and compared. The results show that: 1) with the increase of sampling ratio (from 10% to 15%), the mean of standard deviation of regional GDP decreases with the increase of sampling proportion except 5%. 2) in different sampling methods, The mean of standard deviation of regional GDP is minimum by stratified cluster sampling with unequal probability. The standard deviation of the mean of mortality shows that the standard deviation of stratified cluster sampling with unequal probability is the smallest among the different sampling methods. 4. The design efficiency of different sampling schemes is calculated and compared. The results show that: 1) in the complete random cluster sampling, the average of the design efficiency of the regional GDP with different sampling proportions is the same, all of which are 1; Except for the sampling ratio of 5, as the sampling ratio increases (from 10% to 15%), the average of the design efficiency of the region's gross domestic product decreases. 2) in different sampling methods, The mean of design efficiency of regional GDP with stratified cluster sampling with unequal probability is the smallest. Mortality: 1) in complete random cluster sampling, the mean of mortality design efficiency of different sampling proportions is the same, all of them are 1; With the exception of a sampling ratio of 5, the average of the mortality design efficiency decreases as the sampling ratio increases (from 10% to 15%). 2) in different sampling methods, The mean of mortality design efficiency of stratified cluster sampling with unequal probability is the smallest. 5. The best sampling scheme is stratified cluster sampling with unequal probability. The main innovations of this study are as follows: (1) unequal probability sampling is proposed and applied to the sampling study with a clear sampling frame. (2) repeated sampling calculation and comparison are made for each sampling scheme. Average sampling error, (3) the best sampling scheme with definite sampling frame is stratified cluster sampling with unequal probability.
【学位授予单位】：第四军医大学
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：R311

【参考文献】