6. Concluding remarks
Phillip S. Kott and Dan Liao
In Section 4, we noted two reasons to prefer calibration weighting in two steps: to make implicitly fitting a logistic response model easier and to incorporate nearly quasi-optimal calibration. A side benefit of two-step calibration is more efficient estimation of the response model in step one since there is no sampling error to confound the estimation. This is useful when one wants to analyze the causes of unit nonresponse for its own sake.
We must concede, however, that the reduction in mean squared error using two steps was modest in our simulation experiments in Section 5. Moreover, the practical appeal of the simplicity of calibrating in a single step cannot be denied.
When calibration-weighting is used to adjust for nonresponse that is not missing at random as described in Chang and Kott (2008) and Kott and Chang (2010), the efficiency gains from a second step involving only calibration variables and functions of calibration variables model variables is likely to be sizeable.
When the finite population correction factors can be ignored, replication offers a much simpler approach to variance estimation than equation (3.7) even though the second summation on the right-hand side can be dropped in this situation. A different attractive alternative is the “collapsed” version of equation (4.2) that ignores the impact of the first calibration step:
This estimator clearly estimates the prediction-model variance if that model holds. A version of it − with the second summation removed − fared well in our simulation experiments (not shown). Some caution is needed before one draws too strong a conclusion from that result since the linear model was never too far from holding in our investigations.
Finally, a number of assumptions were made to simplify the exposition. The interested reader can extend the results to unbounded more general and not-necessarily-bounded weight-adjustment functions, or to allow the prediction-model errors to be correlated within primary sampling units. When grows faster than the assumption that can sometimes be dropped. See, for example, Kott (2009, page 69).
Acknowledgements
This paper was prepared for the Symposium on the Analysis of Survey Data and Small Area Estimation, in honour of the 75th Birthday of Professor J.N.K. Rao sponsored by the Fields Institute for Research in Mathematical Sciences. The authors would like to thank the organizers of the conference for the invitation to present this paper and the Institute for its generous funding of the conference without which this paper would never have been written. They would also like to thank a number of editors and referees for their helpful comments.
References
Bang, H., and Robins, J.M. (2005). Doubly robust estimation in missing data and causal inference models. Biometrics, 61, 962-972.
Bankier, M. (2002). Regression estimators for the 2001 Canadian Census. Presented at the International Conference in Recent Advances in Survey Sampling.
Chang, T., and Kott, P.S. (2008). Using calibration weighting to adjust for nonresponse under a plausible model. Biometrika, 95, 557-571.
Deming, W.E., and Stephan, F.F. (1940). On a least squares adjustment of a sample frequency table when the expected marginal total are known. Annals of Mathematical Statistics, 11, 427-444.
Deville, J.-C. (2000). Generalized calibration and application to weighting for non-response. In COMPSTAT: Proceedings in Computational Statistics, 14th Symposium, Utrecht, The Netherlands, (Eds., J.G. Bethlehem and P.G.M. Van der Heidjen), Heidelberg: Physica Verlag, 65-76.
Deville, J.-C., and Särndal, C.-E. (1992). Calibration estimators in survey sampling. Journal of the American Statistical Association, 87, 418, 376-382.
Deville, J.-C., Särndal, C.-E. and Sautory, O. (1993). Generalized raking procedures in survey sampling. Journal of the American Statistical Association, 88, 1013-1020.
Estevao, V.M., and Särndal, C.-E. (2002). The ten cases of auxiliary information for calibration in two-phase sampling. Journal of Official Statistics, 18, 233-255.
Estevao, V.M., and Särndal, C.-E. (2000). A functional form approach to calibration. Journal of Official Statistics, 16, 379-399.
Folsom, R.E. (1991). Exponential and logistic weight adjustments for sampling and nonresponse error reduction. Proceedings of the American Statistical Association, Social Statistics Section, 197-202.
Folsom, R.E., and Singh, A.C. (2000). The generalized exponential model for sampling weight calibration for extreme values, nonresponse, and poststratification. Proceedings of the American Statistical Association, Survey Research Methods Section, available online at http://www.amstat.org/sections/srms/Proceedings/, 598-603.
Fuller, W.A., Loughin, M.M. and Baker, H.D. (1994). Regression weighting in the presence of nonresponse with application to the 1987-88 Nationwide Food Consumption Survey. Survey Methodology, 20, 1, 75-85.
Kim, J.K., and Haziza, D. (2014). Doubly robust inference with missing survey data. Statistica Sinica, 24, 375-394.
Kim, J.K., and Park, H. (2006). Imputation using response probability. Canadian Journal of Statistics, 34, 1-12.
Kim, J.K., and Shao, J. (2013). Statistical Methods for Handling Incomplete Data, London: Chapman and Hall/CRC.
Kott, P.S. (2006). Using calibration weighting to adjust for nonresponse and coverage errors. Survey Methodology, 32, 2, 133-142.
Kott, P.S. (2009). Calibration weighting: Combining probability samples and linear prediction models. In Handbook of Statistics 29B: Sample Surveys: Inference and Analysis, (Eds., D. Pfeffermann and C.R. Rao), New York: Elsevier.
Kott, P.S. (2011). A nearly pseudo-optimal method for keeping calibration weights from falling below unity in the absence of nonresponse or frame errors. Pakistan Journal of Statistics, 27, 391-396.
Kott, P.S., and Chang, T.C. (2010). Using calibration weighting to adjust for nonignorable unit nonresponse. Journal of the American Statistical Association, 105, 1265-1275.
Kott, P.S., and Liao, D. (2012). Comparing weighting methods when adjusting for logistic unit Nonresponse. Presented at Federal Committee on Survey Methodology Research Conference, available online at http://www.fcsm.sites.usa.gov/files/2014/05/Kott_2012FCSM_III-B.pdf.
Little, R.J., and Rubin, D.B. (2002). Statistical Analysis with Missing Data (2nd Ed.), New York: John Wiley & Sons, Inc.
Lundström, S., and Särndal, C.-E. (1999). Calibration as a standard method for the treatment of nonresponse. Journal of Official Statistics, 15, 305-327.
Oh, H.L., and Scheuren, F.J. (1983). Weighting adjustment for unit nonresponse. In Incomplete Data in Sample Surveys, (Eds., W.G. Madow, I. Olkin and D.B. Rubin), New York: Academic Press, 2.
Rao, J.N.K. (1994). Estimation of totals and distributing functions using auxiliary information at the estimation stage. Journal of Official Statistics, 10, 153-165.
Robins J.M., Rotnitzky A. and Zhao L.P. (1994). Estimation of regression coefficients when some regressors are not always observed. Journal of the American Statistical Association, 89, p. 846-866.
- Date modified: