Mallows’s $C_p$

Mallows’s $C_p$ is used to assess model selection in regression analysis using ordinary least squares (OLS).

$$ C_p = \frac{1}{n} \left( \text{RSS} + 2p\hat{\sigma}^2 \right) $$

where:

• $\text{RSS}$: Residual Sum of Squares
• $p$: number of parameters used in the model
• $\hat{\sigma}^2$: estimated variance of the error term

Lower values of $C_p$ indicate better models.