The utility of ‘tree-generating’ statistics in applied social work research1
Introduction
Because of the complex nature of phenomena examined in applied research, there continues to be a need to identify multivariate analytic techniques which are sensitive to interaction effects and which minimize resource and technical demands. Dumas (1989) identified two procedures sensitive to contextual variables and, consequently, ideal for planning efforts: structural equation modeling (e.g., LISREL) and meta-analysis. In applied program planning and evaluation settings, however, few local communities are likely to have numerous, relatively recent, local data sets that are needed for meta-analysis regarding target problems which are known to have substantial regional and temporal variability (e.g., substance abuse), have the resources to gather the large sample size preferred for structural equation modeling, or have the expertise and resources available to utilize either procedure. One solution to this dilemma involves the use of a family of statistical techniques, sometimes referred to as ‘tree-generating’ statistics, as an alternative approach.
Section snippets
Overview of ‘tree-generating’ statistics
Derived from the work of Morgan and Sonquist (1963), this family of statistical techniques is often characterized as ‘tree-based’ or ‘tree-generating’ because the typical display of results resembles a tree. Designed to detect complex statistical relationships, a number of specific programs now exist which can accommodate both categorical and quantitative dependent variables, operate in both mainframe and PC environments, and include options in addition to those addressed in this discussion of
AID example
To demonstrate these points concretely, the following example focuses on utilizing community data on substance use among women. The analysis will demonstrate the utility of ‘tree-generating’ statistics for public sector service planning.
Implications for applied research
The statistical control (i.e., removing the effects of other variables) characteristic of traditional analytic approaches becomes increasingly unrealistic the more micro the level of application. In vivo, no corresponding ability to control the effects of interrelated characteristics and conditions exist. Consequently, it may be argued that traditional methods (e.g., multiple regression) impose conditions which have no corollaries in reality, thereby introducing some degree of artificiality
References (9)
- Clayton, R. R., & Voss, H. L. (1981). Young men and drugs in Manhattan: a causal analysis (National Institute on Drug...
Client admissions for drug abuse treatment: Fiscal Year 1989–90
(1990)Primary prevention: toward an experimental paradigm sensitive to contextual variables
Journal of Primary Prevention
(1989)- et al.
Perinatal HIV prevention project: first year final report
(1989)
Cited by (1)
- 1
The data discussed in this paper were drawn from research supported by the US Department of Health and Human Services, Centers for Disease Control (Grant #U62/CCU403278) through subcontract with the Florida Department of Health and Rehabilitative Services by the Florida State University, Center for Human Services Policy and Administration. See Imershein et al. (1989) for details of the original study.