How to read this analysis
This is an ordinary least squares (OLS) multiple linear regression. It estimates price while holding every other selected feature constant. That “all else equal” qualification is what makes these coefficients different from the single trend line above.
- Coefficient: the estimated dollar change in price for a one-unit increase in that feature, with the other selected features held constant. For mileage, one unit is one mile; multiplying its coefficient by 1,000 gives the estimated change per 1,000 miles. The intercept is the model's predicted price when every feature is zero and may not describe a realistic vehicle.
- Standard error: how uncertain the coefficient estimate is. Smaller values relative to the coefficient indicate a more precise estimate.
- t statistic and p value: test the null hypothesis that the feature's true coefficient is zero. A small two-sided p value (commonly below 0.05) is evidence of an association after accounting for the other selected features; it is not the probability that the coefficient is true.
- R²: the share of observed price variation explained by the model. Adjusted R² penalizes adding features that do not improve the model enough and is more useful when comparing models with different numbers of features.
- Important limits: association does not prove causation. Listing prices can also depend on trim, condition, seller, accident history, options, and other omitted factors. Results may be unstable with a small sample, highly related features, unusual listings, or a non-linear relationship. “Age” and “model year” encode the same information, so selecting both makes the model mathematically singular.