
There's an idea in statistics called the most parsimonious model. Say you want to generate a model to predict scores on a standardized test; you might throw IQ, socio-economic status, mother's educational attainment and GPA into your statistical model and, using those factors, hope to predict something like SAT scores with a reasonable amount of accuracy. Maybe that model gets you accurate predictions 95% of the time. If, however, you only put IQ and mother's educational attainment in the model, you get a 92% accuracy rate. You'd go with the second model because it has more parsimony; it's not worth [...]