INTERNAL REPLICATION STRATEGIES FOR (MODERATELY) LARGE SAMPLES: CROSS-VALIDATION TECHNIQUES IN PROJECT TALENT. (11th November 2018)

• Record Type: Journal Article
• Authors: Chapman, B; Huang, A; Horner, E; Peters, K; Sempeles, E; Roberts, B; Lapham, S
• Is Part Of: Innovation in aging. Volume 2(2018)Supplement 1
• Page Start: 786
• Page End: 786
• Publication Date: 2018-11-11
• Subjects: Aging -- Periodicals; Gerontology -- Periodicals; 612.67
• DOI: 10.1093/geroni/igy023.2911
• Languages: English
• ISSNs: 2399-5300

Abstract

Abstract: Integrative analysis of several distinct studies has become the gold standard for replicability. At times however, researchers may have only one data source. When it is sufficiently large, however, single studies permit the deployment of a range of cross-validation methods that permit one to examine the consistency, stability, or replication of parameter estimates such as regression coefficients. Using an example from Project Talent, a large cohort study, k-fold, jackknife, and bootstrap cross-validation are demonstrated in a survival analysis predicting 49-year mortality from adolescent personality traits. Both simple graphical and formal measures of cross-validation consistency are then reviewed, and guidelines for quantifying and presenting the stability of results are considered. The promises and perils of internal cross-validation in smaller samples are also discussed.