Comparative Study On The Cluster Sampling Designs Of Sentinel Sites Death Surveillance Monitoring

Sentinel Sites for Death Surveillance is to see the level of death and the death causes, to master factors of residents'health, for the government to formulate health policies, to evaluate the quality and effectiveness of health scientific basis, also is to study the natural population change. The Extraction of the Sentinel Sites for Death Surveillance is a primary issue. In this study, we take the Selection of the Sentinel Sites for Death Surveillance in Shaanxi Province for example, make sampling unit of 107 County (city, district) in Shaanxi Province as the sampling frame, the study were the following:1. Formulate different sampling designs. Combing with the actual situation in China, we extract the application of unequal probability sampling method. To establish four viable sampling methods which are completely random cluster sampling, stratified cluster sampling, unequal probability cluster sampling, unequal probability stratified cluster sampling. We make 5%, 10%, 15% respectively of the total population as the sampling ratio, which are combined into 12 different kinds of sample programs, which are simulated by computer for 100 times.2. We calculate the accuracy of the sampling programs, results showed that: Completely random cluster sampling ratio which is greater than 10% of the sample can be representative of the overall; stratified random cluster sampling ratio which is greater than 15% have better representation of the overall; unequal probability cluster sampling ratio which is greater than 15% can have a better representation of the overall; unequal probability stratified cluster sampling ratio which is greater than 15% have good representation of the overall.3. We calculate and compare the sample average of sample standard deviation of the sample programs. The results showed that: GDP mean standard deviation: 1) Expected of 5% sampling ratio, with the sampling ratio increased from 10% to 15%, the mean standard deviation of GDP reduced instead. 2) Different sampling methods have different GDP mean standard deviation which is the lowest under unequal probability of stratified cluster sampling. Mortality mean standard deviation showed that: Among the different sampling methods, unequal probability stratified cluster sampling standard deviation is the smallest.4. We calculate and compare the design effect of different sampling design, results showed that: GDP: 1) Different sampling GDP mean sampling effect is the same in completely random cluster samples, is one; In addition to sampling ratio of 5%, with the sampling proportion increased from 10% to 15%, GDP mean sampling effect is reduced. 2) In different sampling methods, the unequal probability stratified cluster GDP mean sampling effect is the smallest. Mortality: 1) Different sampling mortality mean sampling effect is the same in completely random cluster samples, are one; In addition to sampling ratio of 5%, with the sampling proportion increased from 10% to 15%, mortality mean sampling effect was reduced. 2) In different sampling methods, the unequal probability stratified cluster mortality mean sampling effect is the smallest.5. We explore the best sampling design plan which is the unequal probability stratified cluster.The main innovation of this study includes following three points: (1) We state unequal probability sampling method which is used in a clear finite sampling frame. (2) We simulate and compare the average of sampling error, sampling precision and design effect of each sampling design plan. (3) We evaluate and put forward the best sampling plan for a clear finite sampling frame.