Determination of the Most Appropriate Number of Multiple Imputations for the Schistosomiasis Surveillance Data in China

Zhao Fei,Zhang Zhijie,Liu Jianxiang
DOI: https://doi.org/10.3969/j.issn.1002-3674.2009.05.002
2009-01-01
Abstract:Objective To determine the most appropriate number of multiple imputations(MI) on the schistosomiasis surveillance data of China.Methods Markov Chain Monte Carlo(MCMC) method of MI was used to impute the missing values in the dataset of schistosomiasis surveillance from 1990 to 2003.The indices of variance between imputations,degree of freedom(DF),relative degree of variance increasing,evaluation of missing information for the population paramters and relative efficiency were adopted to assess the effects of imputation from different aspects and then the best imputation number was determined.Results There are 200 observations and 28 variables in the whole dataset.Five variables had missing values,the infection rate in the total population(6%),the livestock's infection rate(6.5%),the density of infected snail in spring(17%),the density of live snails in spring(16.5%) and the contaminated rate of cattle dejecta(53%).The pattern of missing data was arbitrary missing pattern and the most appropriate number of imputations was 31 for this study dataset.Conclusion MCMC method was suitable for the pattern of missing data in the schistosomiasis surveillance data of China and the best imputing result was obtained when the number of imputations was 31.
What problem does this paper attempt to address?