Researchers have developed new formulas to estimate the expected out-of-sample R-squared and correlation for predictive models using ordinary least squares and ridge regression. These formulas, detailed in a paper submitted to arXiv, relate predictive performance to design-stage quantities such as sample size, number of predictors, assumed population R-squared, and a design efficiency factor. The methodology accounts for how covariates are arranged, moving beyond traditional formulas that only consider sample size and dimension. An R package named 'predpower' has been created to implement these calculations, which were applied to a study predicting personality from smartphone behavior to demonstrate their utility in assessing a study's predictive power before analysis. AI
IMPACT Provides researchers with tools to better plan predictive modeling studies, potentially improving the efficiency and reliability of AI research.
RANK_REASON The cluster contains a research paper detailing new statistical methodology. [lever_c_demoted from research: ic=1 ai=0.7]
- arXiv
- D*
- ordinary least squares
- leave-one-out cross-validation
- number of predictors
- population R^2
- predpower
- R
- sample size
- ridge regression
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
