Linear Models

A linear scoring or prediction model combines inputs with fixed weights, often with an intercept. Consistency can help repeated judgments, and simple weighting rules have performed well in studied prediction tasks. Performance still depends on inputs, measurement, scaling, context, and validation.

A weighted sum applies the same combination rule to each set of inputs. That consistency can be valuable when repeated intuitive judgments use shifting criteria, but the choice of inputs, scales, and outcome still determines what the score means. Summing dollars, minutes, and ratings without a defensible scaling rule creates an arbitrary number.

Distinguish a prediction model from a preference rubric. Predictive weights should be evaluated against outcomes on unused cases; preference weights express priorities and need sensitivity checks. Linear models can contain many predictors, transformed variables, or interaction features, so the label alone does not guarantee simplicity or protection against overfitting.

When to use it

When repeated decisions would benefit from consistency; when human judgment is being corrupted by irrelevant factors; when simple models would outperform complex intuitive assessment; when transparent, interpretable decision-making is needed.

How it can help

Define the outcome or preference the score represents, select defensible inputs and scales, and compare the model with realistic alternatives. Use unused cases for predictive evaluation and sensitivity checks for preference weights.