Missing values,frequently seen in survey,can be imputed under normal linear model when variables with missing values correlate with variables without missing values.Compared with singular imputation,multiple imputation can efficiently estimate the population variance and is widely used.And the application of Bayesian multiple imputation makes the population variance more accurate because both difference and residuals estimate originate from the random selection of posterior distribution.A large number of simulation tests find that Bayesian multiple imputation,compared with the singular imputation and multiple imputation,could construct a wider confidence interval so as to achieve a higher parameter coverage,which is more convincing when missing data account for a great proportion