The adjusted R-squared adjusts for the degrees of freedom to better support comparison across models with different numbers of parameters. The best model will have the largest R-squared. $R^2_a = 1 - \frac{RSS/(n - (p+1))}{TSS/(n-1)} = 1 - \Big (\frac{RSS}{TSS} \Big ) \Big (\frac{n-1}{n-p-1} \Big )$