SAFETYLIT WEEKLY UPDATE

We compile citations and summaries of about 400 new articles every week.
RSS Feed

HELP: Tutorials | FAQ
CONTACT US: Contact info

Search Results

Journal Article

Citation

Hoggarth PA, Innes CR, Dalrymple-Alford JC, Severinsen JE, Jones RD. Accid. Anal. Prev. 2010; 42(6): 1759-1768.

Affiliation

Van der Veer Institute for Parkinson's and Brain Research, 66 Stewart Street, Christchurch 8011, New Zealand; Department of Psychology, University of Canterbury, Private Bag 4800, Christchurch 8140, New Zealand.

Copyright

(Copyright © 2010, Elsevier Publishing)

DOI

10.1016/j.aap.2010.04.017

PMID

20728627

Abstract

This study compared the ability of binary logistic regression (BLR) and non-linear causal resource analysis (NCRA) to utilize a range of cognitive, sensory-motor, personality and demographic measures to predict driving ability in a sample of cognitively healthy older drivers. Participants were sixty drivers aged 70 and above (mean=76.7 years, 50% men) with no diagnosed neurological disorder. Test data was used to build classification models for a Pass or Fail score on an on-road driving assessment. The generalizability of the models was estimated using leave-one-out cross-validation. Sixteen participants (27%) received an on-road Fail score. Area under the ROC curve values were .76 for BLR and .88 for NCRA (no significant difference, z=1.488, p=.137). The ROC curve was used to select three different cut-points for each model and to compare classification. At the cut-point corresponding to the maximum average of sensitivity and specificity, the BLR model had a sensitivity of 68.8% and specificity of 75.0% while NCRA had a sensitivity of 75.0% and specificity of 95.5%. However, leave-one-out cross-validation reduced sensitivity in both models and particularly reduced specificity for NCRA. Neither model is accurate enough to be relied on solely for determination of driving ability. The lowered accuracy of the models following leave-one-out cross-validation highlights the importance of investigating models beyond classification alone in order to determine a model's ability to generalize to new cases.


Language: en

NEW SEARCH


All SafetyLit records are available for automatic download to Zotero & Mendeley
Print