multiple regression

A single straight line through age-versus-cost is useful, but real outcomes depend on many things at once. Medical cost rises with age, but also with smoking, region, and chronic conditions. Multiple regression lets all these predictors act together in one equation, so you can ask the powerful question: holding everything else fixed, what does each factor add on its own?

The model extends the simple line to several predictors: y = a + b1*x1 + b2*x2 + b3*x3 + error. Each coefficient b is the change in y for a one-unit change in its predictor while the other predictors are held constant — the all-important phrase 'holding the others fixed.' Least squares still chooses the coefficients that minimize the total squared error, now in many dimensions at once. For instance, cost = 400 + 70*age + 1,200*smoker says age adds 70 per year and being a smoker adds 1,200, each measured after accounting for the other. This ability to isolate one factor's effect from the tangle of others is exactly why the technique is so valued.

Multiple regression is the direct ancestor of the GLMs actuaries use to set rates, where dozens of rating variables — age, vehicle, territory, prior claims — jointly determine a premium relativity. The honest caveats sharpen here. When predictors are themselves strongly correlated (multicollinearity), their individual coefficients become unstable and hard to interpret, even if the overall predictions stay fine. And piling in more variables always nudges the in-sample fit upward, which tempts overfitting; that is why adjusted R-squared and the model-selection criteria penalize complexity rather than rewarding raw fit.