How to Tell When Simpler, More Unified, or Less Ad Hoc Theories will Provide More Accurate Predictions

By Malcolm R. Forster, Elliott Sober. 1994

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

The Promise and Peril of Human Evaluation for Model Interpretability (2017, background): “The problem is also discussed in the philosophy of science under "scientific realism" and "simplification" [6, 7].”
Instrumentalism, Parsimony, and the Akaike Framework (2002, background): “They can’t be better in Bayesian terms, because they can’t be more probable (Popper 1959; Forster and Sober 1994).”
Predicting Wimbledon 2005 tennis results by mere player name recognition (2007, unknown): “Because of this, heuristics ignore the noise inherent in many data sets, and are thus more robust to overfitting, a problem that impairs the predictive power of many complex and highly parameterized models (Forster & Sober, 1994; Hertwig & Todd, 2003).”
Curve Fitting, the Reliability of Inductive Inference, and the Error‐Statistical Approach (2007, methodology): “…of models, one can use a formal model selection procedure, such as the Akaike Information Criterion (AIC) and related procedures based on the likelihood function, to select a fittest model by trading goodness-of-fit against simplicity (see Forster and Sober 1994 and Kieseppa 1997, inter alia).”
Predictive Analytics in Information Systems Research (2011, unknown): “Yet, contrary to common belief, explanatory power does not imply predictive power (Dawes 1979; Forster and Sober 1994).”
Kullback-Leibler information as a basis for strong inference in ecological studies (2001, background): “Parsimony enjoys a featured place in scientific thinking in general and in modelling specifically (see Forster and Sober 1994; Forster 2001) for a strictly science philosophy perspective).”
‘All models are wrong...’: an introduction to model uncertainty (2012, background): “Statistica Neerlandica © 2012 VVS. taking model selection as a trade-off between the goals of accommodating old data and predicting new (Forster and Sober, 1994; Hitchcock and Sober, 2004) or, equivalently, of balancing bias and variance, or, again equivalently, of weighing complexity against fit.”
Model Selection and the Multiplicity of Patterns in Empirical Data (2007, methodology): “The best model of the data set is the model corresponding to the best-fitting curve that is a member of the family with the highest expected predictive accuracy (Forster and Sober 1994).”