Fitting Lines to Data

Straight-line regression estimates a response as an intercept plus a slope times a predictor. Ordinary least squares minimizes squared response residuals. Fit, predictive performance, and causal explanation are distinct questions; each requires appropriate checks.

A fitted straight line summarizes how an observed response varies with a predictor within a dataset. Ordinary least squares chooses its intercept and slope to minimize squared vertical residuals: observed response minus predicted response. The line compresses observations; it does not reveal why the relationship exists or guarantee that future observations follow it.

Inspect what the compression leaves behind. Curved residual patterns suggest a missed shape, changing spread affects uncertainty, and a single unusual observation can strongly influence the line. Evaluate predictions on observations excluded from fitting and report the relevant input range. A simpler line and a more flexible curve should compete on the actual prediction task, not on visual elegance.

When to use it

When data patterns need quantification; when variable relationships need estimation; when distinguishing signal from noise; when overfitting risk needs management.

How it can help

Use a fitted line to summarize or predict a measured relationship over a supported range. Inspect residuals and influential observations, and assess prediction on data excluded from fitting. Add complexity when it improves the relevant task with adequate evidence.