Articles in this Special Section have important implications for researchers wishing to contribute to scientific bases for decision making. As noted, existing statistical decision rules are often sufficiently compromised by their basis on non optimal predictors and inadequate data-combination procedures as to render them of limited value. Indeed, of the 93 studies cited by that used hit rates as the accuracy statistic, in fewer than half of these (44 of 93) did the hit rates for mechanical prediction exceed the accuracy of clinical prediction by more than 5 percentage points.
One implication of this finding is the need for researchers to identify better sets of predictors by relying less heavily on post hoc analyses of data sets of convenience and by formulating prior to data collection those constructs most likely to be related both empirically and theoretically to the criterion (or criteria) of interest. A second implication of this finding, noted by Garb, is the need for researchers to pursue expertise in new statistical techniques affording increased power for analyzing complex predictor-criterion relationships.
The article by price at all on artificial neural networks (ANNs) poses a classical paradox in this regard. On the one hand, Price et al. demonstrated across two data sets that ANN analysis offers a powerful and flexible technique for predicting categorical criterion variables derived from complex diagnostic processes as well as discrete phenomena (eg, development of a disorder, attrition from treatment, or relapse) in which conditional probabilities of the event change over time.
They noted that ANNs are more capable of detecting complex relationships than are linear models, and they are particularly advantageous when the underlying functional relationships between predictors and criteria are unknown. On the other hand, ANN analysis is sufficiently new to psychological assessment and sufficiently complex that in the near future only a handful of researchers (let alone clinicians) are likely to pursue the special expertise necessary for using this procedure. (Anderson 2006)
This leads to a third implication, which is that researchers bear a responsibility for making mechanical aids for decision making more accessible to the typical practitioner. When mechanical methods of processing data involve complex multivariate procedures, this objective may be met by making computerized algorithms available on micro diskettes – much as CBTIs are distributed currently – along with relevant documentation regarding specific operations for use, underlying theoretical and empirical rationale, and appropriate cautions regarding contexts of application and means of implementing results.
Alternatively, researchers can translate findings from complex multivariate methods into simpler algorithms readily adopted by most practitioners. Garb noted that linear prediction models have worked surprisingly well in many prediction contexts, and there is some evidence that unit weighting of predictors frequently works as well as or better than differential weights.
Examples of successful application of linear unit-weighting models are as diverse as the detection of neurological impairment, differential diagnosis of neurosis versus psychosis, and predicting couples’ divorce 4 years following marital therapy. Assistive technology (AT) devices and services promote a child’s ability to move, communicate, and interact while participating in everyday activities and routines and may also help families support their children’s learning and development.
For infants and toddlers (i. e. , children ages birth through 2 years) with disabilities, AT offers a set of tools to help achieve desired family outcomes and promote acquisition of children’s developmental goals. AT, which consists of both low-tech (eg, switches, head pointers, picture boards, crutches) and high-tech (eg, computers, power wheelchairs, augmentative and alternative communication [AAC]) devices, can increase young children’s options and facilitate their physical and social inclusion in various natural settings. (Simon 2004) http://www.imediaconnection.com/content/10679.asp