Robustness Considerations in the Choice of a Method of Inference for Regression Analysis of Survey Data

SUMMARY This paper follows the research of Holt, Smith and Winter (1980) and Nathan and Holt (1980) and compares the performance of model based and design based procedures for the regression analysis of survey data, when some of the model assumptions are violated. The comparisons are both analytical and based on simulations and indicate the sensitivity of the model based procedure to correct specification of the relationships between the regression variables and the design variables. The conclusion from the paper is that the randomization distribution can not be ignored in the inference process and model based estimators should be modified to be design consistent estimators of the finite population regression coefficients.