Statistics Glossary → Regression through the origin (no intercept)
Regression through the origin (no intercept)concept
A model that drops the intercept and forces the regression line or plane through the origin (zero response when all predictors are zero).
The no-intercept model is y = b₁x₁ + … + bₚxₚ + e. It is defensible only when theory requires the response to be exactly zero when the predictors are zero (for example, the variable part of cost is zero when nothing is produced). If the true intercept is not zero, the slope estimates are biased, and the residuals no longer sum to zero.
In a no-intercept model R² is computed from the squares of the raw y values instead of deviations from the mean (the uncentred definition), so it usually comes out very high and cannot be compared with the R² of a model with an intercept. If there are no observations near x = 0, forcing the intercept to zero imposes a strong assumption outside the observed range. When in doubt, one can fit the model with an intercept and test whether it differs from zero.
An everyday example
A painter studies how paint use relates to the area painted. Painting nothing uses no paint, so a model through the origin seems sensible. But if every job uses a fixed 0.3 litres just to wet the brushes, forcing the intercept to zero overstates the paint used per square metre.
In MerQur
The 'No-intercept model' option in Multiple Linear Regression; R² is computed with the uncentred definition and cannot be compared with the model that has an intercept. It is recommended only when theory requires y = 0 at x = 0.