Supplement to “ Sensitivity to missing data assumptions : Theory and an evaluation of the U . S . wage structure ”

In this model, the parameter ρ indexes the degree of nonignorable selection in the outcome variable Yi. We chose μ = 0 6745 to ensure a missing fraction of 25%, which is approximately the degree of missingness found in our analysis of earnings data in the U.S. Census. We computed the distributions of missing and observed outcomes for various values of ρ by simulation, some of which are plotted in Figures A.1 and A.2. The resulting values of S(F), which correspond to the maximum vertical distance between the observed and missing CDFs across all points of evaluation, are given in Table A.1.