Bootstrap方法在渔业捕捞分层抽样调查中的应用分析

Application of Bootstrap method in stratified sampling survey for capture fishery

应用Bootstrap方法,对太湖捕捞渔业抽样调查参数估计进行研究,探讨分层抽样调查在捕捞产量统计中的应用。随机抽样模拟的结果表明,有放回的Bootstrap方法适用于抽样比例较小的分层抽样调查。对于总体分布未知且样本量容量有限的情况下分层抽样调查的方差估计,重权Bootstrap方法相比其它方法具有明显的优势,适合用于渔业抽样调查工作。

Samples based on fishery survey have interference of man-made factors and cut statistical been developing in China in recent years to reduce the costs. Survey on capture fisheries tends to use stratified sampling, so the samples can be more evenly distributed among population. However, the sample size may be insufficient in some stratum after stratification and it is hard to determine the probability distribution. The conventional method could be quite cumbersome to take parameter estimation. Several Bootstrap methods are proposed to address statistical inference problem with finite samples. The application of Bootstrap method is rarely reported in fishery statistics. In this paper the Bootstrap method is used to study the parameter estimation from sample survey for the capture fishery in Lake Taihu and the application of stratified sampling in the statistics of fishing production is discussed. Based on the data of frame survey of capture fishery in Lake Taihu, the stratified sampling can be designed with fishing vessel types or combination of fishing vessel types and fishing areas. The stratified sampling for vessel types can be stratified to 6 stratums or 20 stratums. A simulation study was conducted to examine the performance of Naive Bootstrap and Rescaled Bootstrap method to obtain the empirical distribution of Bootstrap estimation and the Bootstrap variance estimation. In order to evaluate the relative merits, the relative bias was measured to evaluate the accuracy of the studied estimators and the relative root mean square error was measured to evaluate the stability. The results of sample designs can be evaluated by the designing effects. Results showed that the Bootstrap method with replacement is suitable for the stratified sampling with small size of sampling fractions. The parameter estimation accuracy was not higher than other methods if the sampling fractions were not negligible. As for the variance estimation of the stratified sampling under unknown population distribution and limited sample size the Rescaled Bootstrap method had obvious advantages compared to other methods and was suitable for sample survey of capture fishery. In view of designing effects Bootstrap method with replacement is designed for the stratified sampling with small size of sampling fractions. According to the conclusion of simulation based on the data of frame survey in Lake Taihu in 2009, the Bootstrap method has been adopted to take estimations of catch and value by species and by month in the sample survey since 2010 and it has obtained good results. The Bootstrap method will be further studied in the stratified multistage sampling so as to provide a new approach for the national large-scale fishery survey.