METHODS OF CASE CONTROL SAMPLING FOR THE COX REGRESSION MODEL (SURVIVAL ANALYSIS, SIMULATION, PROPORTIONAL HAZARDS).

Résumé du livre

SS2 was found to have a significantly smaller mean squared error than RS by about 13% at large values of (beta) and 5% overall. Using the full Cox model estimates as the minimum MSE attainable, SS2 was 30% closer to the Cox MSE than was RS. SS2's performance was superior to RS on every criterion investigated. SS1 gave dramatic reductions in variance at H(,o): (beta) = 0, but had serious problems with bias at large (beta). SS2 is recommended as a favorable alternative to simple random sampling providing