🔗 Regression Calculator

The Line of Best Fit, and How to Judge Whether It Actually Fits

Linear regression finds the single straight line that best represents the relationship between two variables across an entire dataset, using a method called 'least squares' that has a clean, intuitive goal: minimize the total squared vertical distance between the line and every actual data point.

Squaring the distances before summing them, rather than simply summing the raw distances, serves the same purpose it does in standard deviation calculations: it prevents points above and below the line from canceling each other out, and it penalizes large deviations more heavily than small ones, pushing the line toward a genuine overall best fit rather than one that happens to average out to zero net error by coincidence.

The resulting equation, y = a + bx, has two components with distinct practical meaning. The slope (b) is the more informative number in most applications — it says exactly how much the dependent variable is expected to change for every one-unit increase in the independent variable, which is the actual predictive relationship being modeled. The intercept (a) is the model's predicted value of y when x equals zero, which is sometimes meaningful (a company's fixed costs before any units are produced) and sometimes not meaningful at all if x=0 falls outside any realistic range of the data (predicting a person's weight at age zero, for instance, would not be sensible).

R², the coefficient of determination, is the single most useful number for judging whether the regression line is actually a good model of the relationship, separate from whether a line could technically be drawn at all. An R² of 0.9 means 90% of the variation in y can be explained by its linear relationship with x — a strong fit. An R² of 0.2 means only 20% is explained, meaning 80% of the variation in y is driven by factors the model does not capture at all, and a straight-line relationship with x alone is a weak predictor.

Linear regression's applications span forecasting next quarter's sales from historical trend data, estimating a house's price from its square footage, and testing scientific hypotheses about how one measured variable affects another. Its core assumption — that the relationship between the two variables is genuinely linear — should always be checked before trusting the result, since forcing a straight line onto a fundamentally curved relationship can produce a misleadingly confident-looking equation that does not actually describe the data well.