Comparison Of Simple Sampling Design And Complex Sampling Design On Different Prevalence Estimation

Objective:To compare simple sampling design and complex sampling design on the accuracy of prevalence estimation and sampling errors under different prevalence conditions in large surveys.To provide evidence for researchers to choose appropriate sampling methods with prevalence estimation in surveys.Methods: The sampling population of the study was 381,499 resident population who participated in the national health examination above 15-year-old in a prefecture of Xinjiang in 2017.Three diseases with different prevalence(hypertension was 22.35%,diabetes was 13.86% and anemia was 0.95%)were selected and to make 100 times simulated samplings with simple random sampling(simple sampling design)and stratified three-stage PPS sampling(complex sampling design)based on the sample size corresponding to the three diseases.By comparing the difference between the estimations and the parameter and calculating the sampling errors and probability of 95% confidence interval including the parameter,to evaluate the feasibility of the two designs.Results:1.When the prevalence was 22.35%,the average of prevalence estimation was 22.11% by simple sampling which was significantly underestimated.The average of prevalence estimation was 21.65%,the sampling error was 0.65% and the probability of 95% CI including the parameter was 54% by complex sampling using traditional analysis method which was significantly underestimated.There was no difference between the average of prevalence estimation and the parameter by complex sampling using Taylor method and Jackknife method.There was underestimated in sampling error and the probability of 95% CI including the parameter by using Taylor method.2.When the prevalence was 13.86%,there was no difference between the average of parameter estimation and the parameter by simple sampling.The average of parameter estimation was 14.79% by complex sampling using traditional method which was overestimated and sampling error was 0.42%,the probability of 95% CI including the parameter was 38% which were underestimated.There was no difference between the average of prevalence estimation and the parameter by complex sampling using Taylor method and Jackknife method.There was underestimated in sampling error and the probability of 95% CI including the parameter by using Taylor method.3.When the prevalence was 0.95%,there was no difference between the average of parameter estimation and the parameter by simple sampling.The average of parameter estimation was 1.07% by complex sampling using traditional method which was overestimated and sampling error was 0.03%,the probability of 95% CI including the parameter was 26% which were underestimated.There was no difference between the average of prevalence estimation and the parameter by complex sampling using Taylor method and Jackknife method.There was underestimated in sampling error and the probability of 95% CI including the parameter by using Taylor method.Conclusion:In large sampling surveys,researchers should use complex sampling design when the prevalence of the disease is high.There is also good result by using simple sampling design when the prevalence of the disease is medium or low.When using complex sampling design,researchers should avoid using traditional analysis method and choose Jackknife method with variance estimation.