Improved meta-analysis for truncated data

This paper proposes an improved meta-analysis for truncated data to decrease the bias from file drawer problem for meta-analysis. It shows the construction of confidence intervals and test after parameters estimate approached by maximum likelihood estimate with Newton-Raphson iteration, or by way of bootstrap method or jackknife method which beyond the tradition method under the supposition that parameters are normally distributed. This model is available when trials are heterogeneous and contains traditional meta-analysis that with no truncation as special cases. In the case analysis we show that both the confidence intervals constructed by way of bootstrap t method, bootstrap quantiles or the best confidence intervals constructed by bootstrap sampling and jackknife sampling is far more better than the result of traditional meta-analysis.