Standard Error of Estimate and Coefficient of Determination
Residual Spread and Explained Variation in Linear Regression
For an observed response y and fitted response ŷ, the residual is e = y − ŷ. Residuals measure the vertical discrepancies left by a fitted regression line. Their squared total is the sum of squared errors, SSE.
In simple linear regression with an intercept, the standard error of estimate is se = √[SSE/(n − 2)]. It is measured in the same units as the response and summarizes the typical scale of residual variation after fitting two parameters. It is not automatically a prediction interval for every new observation.
The coefficient of determination is R2 = 1 − SSE/SST, where SST is the total squared variation of responses about their sample mean. In the fitted sample, R2 is the fraction of that total variation associated with the linear model.
It does not show that the predictor causes the response, and it does not mean that the stated percentage of individual observations is predicted exactly.
Compare standard errors only when response units and contexts are comparable. A lower SSE or standard error can indicate a closer in-sample fit, but residual plots and validation evidence still matter for model adequacy and prediction outside the fitted data.
A simple linear regression with an intercept is fitted to n = 20 observations and has SSE = 72. What is the standard error of estimate?
For a regression with an intercept, SST = 200 and SSE = 50. What is R2?