Measures of Fit

For the Measure of Fit, we usually report R2R^2

R2=ESSTSS=1SSRTSSR^2=\frac{E S S}{T S S}=1-\frac{S S R}{T S S}

where we have that:

  • TSS=1in(YiYˉn)2T S S=\sum_{1 \leq i \leq n}\left(Y_i-\bar{Y}_n\right)^2: total sum of squares

  • ESS=1in(Y^iYˉn)2E S S=\sum_{1 \leq i \leq n}\left(\hat{Y}_i-\bar{Y}_n\right)^2: explained sum of squares

  • SSR=1inU^i2S S R=\sum_{1 \leq i \leq n} \hat{U}_i^2: residual sum of squares

  • TSS=ESS+SSRTSS = ESS +SSR

Note that, for Yˉn\bar{Y}_n, we have that Yˉn=1ni=1nYi=1ni=1nY^i\bar{Y}_n = \frac{1}{n} \sum_{i=1}^n Y_i = \frac{1}{n} \sum_{i=1}^n \hat{Y}_i

  • R2=1R^2=1 if and only if SSR=0S S R=0, i.e., U^i=0\hat{U}_i=0 for all 1in1 \leq i \leq n.

  • R2=0R^2=0 if and only if ESS=0E S S=0, i.e., Y^i=Yˉn\hat{Y}_i=\bar{Y}_n for all 1in1 \leq i \leq n.

For the interpretations of these measures:

  • View 1n1in(Y^iYˉn)2\frac{1}{n} \sum_{1 \leq i \leq n}\left(\hat{Y}_i-\bar{Y}_n\right)^2 as an estimator of Var[Yi]\operatorname{Var}\left[Y_i\right]

  • View 1n1inU^i2\frac{1}{n} \sum_{1 \leq i \leq n} \hat{U}_i^2 as an estimator of Var[Ui]\operatorname{Var}\left[U_i\right]

  • R2R^2 may be then viewed as an estimator of 1Var[Ui]Var[Yi]1-\frac{\operatorname{Var}\left[U_i\right]}{\operatorname{Var}\left[Y_i\right]}

  • Replacing these estimators with their unbiased counterparts yields "adjusted" R2R^2, which is

Rˉ2=1n1nk1SSRTSS\bar{R}^2=1-\frac{n-1}{n-k-1} \frac{S S R}{T S S}

As R2R^2 always increases with the inclusion of an additional regressor, to deal with this problem, we should use "adjusted" R2R^2: Rˉ2\bar{R}^2.

Last updated