Search is a binary segmentation procedure used to develop a predictive model for a dependent variable. It searches among a set of predictor variables for the predictors that most increase the researcher’s ability to account for the variance or distribution of a dependent variable. The question “What dichotomous split on which single predictor variable will give us a maximum improvement in our ability to predict values of the dependent variable?” asked iteratively, is the basis for the Search algorithm.
Search can perform the following functions:
- Maximize differences in group means, group regression lines, distributions (maximum-likelihood chi-square criterion), or ranking (Kendall’s tau-b).
- Rank the predictors to give them preference in the partitioning.
- Sacrifice explanatory power for symmetry.
- Start after a specified partial tree structure has been generated.
The University of Michigan version of Search is a set of C and FORTRAN routines that can be launched from R, SAS, SPSS or Stata or run independently using data from many sources. It is available for computers running the Linux, Mac OS X and Microsoft Windows operating systems. For the independent Search implementation see the Srcware references in Documentation.
Search is also available as part of MicrOsiris, a full-featured data management and analysis software package for Microsoft Windows from Van Eck Computer Consulting at URL:
Search is freeware. The University of Michigan retains the copyright for Search and authorizes its use free of charge. See the Search License Agreement for details.
Search was developed by James N. Morgan, Peter W. Solenberger, Neal A. Van Eck, and Pauline R. Nagara.