Selection of sentinel sites for death surveillance,using cluster or unequal probability sampling / 中华流行病学杂志

ABSTRACT

To compare the sampling errors from cluster or unequal probability sampling designs and to adopt the unequal probability sampling method to be used for death surveillance.Taking 107 areas from the county level in Shaanxi province as the sampling frame,a set of samples are drawn by equal probability cluster sampling and unequal probability designs methodologies.Sampling error and effect of each design are estimated according to their complex sample plans.Both the sampling errors depend on the sampling plan and the errors of equal probability in stratified cluster sampling appeares to be less than simple cluster sampling.The design effects of unequal probability stratified cluster sampling,such as πPS design,are slightly lower than those of equal probability stratified cluster sampling,but the unequal probability stratified cluster sampling can cover a wider scope of monitoring population.Conclusions:Results from the analysis of sampling data can not be conducted without consideration of the sampling plan when the sampling frame is finite and a given sampling plan and parameters,such as sampling proportion and population weights,are assigned in advance.Unequal probability cluster sampling designs seems to be more appropriate in selecting the national death surveillance sites since more available monitoring data can be obtained and having more weight in estimating the mortality for the whole province or the municipality to be selected.