Local model uncertainty and incomplete-data bias

摘要: Summary. Problems of the analysis data with incomplete observations are all too familiar in statistics. They doubly difficult if we also uncertain about choice model. We propose a general formulation for discussion such problems and develop approximations to resulting bias maximum likelihood estimates on assumption that model departures small. Loss efficiency parameter estimation due incompleteness has dual interpretation: increase variance when an assumed is correct; incorrect. Examples include non-ignorable missing data, hidden confounders observational studies publication meta-analysis. Doubling variances before calculating confidence intervals or test statistics suggested as crude way addressing possibility undetectably small from The problem assessing risk lung cancer passive smoking used motivating example.