285 – 285 - New Advances in Sample Design and Adjusting for Survey Nonresponse
A Comparison of Two CHAID Packages for Modeling Survey Nonresponse
Tien-Huan Lin
Westat
When computing survey weights for use in analysis of complex sample survey data, it is common practice to mitigate bias due to unit nonresponse by modeling response propensity and adjusting weights to account for different response propensities. The CHAID (Chi-square Automatic Interactive Detector) algorithm is commonly utilized to produce weighting classes for this purpose. We review two popular software packages that implement the CHAID algorithm: SI-CHAID and HPSPLIT. SI-CHAID is an interactive stand-alone GUI that is easy to manipulate and produces informative graphical images of the decision tree, but requires manual intervention and additional effort to incorporate into a code-based environment. HPSPLIT is a SAS code-based procedure. However, manipulation of the tree is less intuitive and the graphical representation of the tree is less informative than SI-CHAID. We empirically evaluate the two packages in terms of the resulting bias and variance of the weighted estimates using simulations of a complex survey sample design to examine the interchangeability of the two software packages such that users can determine the software that best meets the analysis needs of the study.