The ANOVA table
| Source | Sum of squares | df | Mean square |
|---|---|---|---|
| Regression | RSS | k | MSR = RSS / k |
| Error | SSE | n − k − 1 | MSE = SSE / (n − k − 1) |
| Total | SST = RSS + SSE | n − 1 |
From it:
- R² = RSS ÷ SST
- Adjusted R² = 1 − [(n − 1)/(n − k − 1)] × (1 − R²)
- F = MSR ÷ MSE
- SEE = √MSE
Enter your own table:
The overall F-test
H0: all slope coefficients = 0. Ha: at least one ≠ 0. Reject if F exceeds the critical value with k and n − k − 1 degrees of freedom. Here F = 15.9, far above the 5% critical value of about 2.77.
Joint tests of a subset
To test whether q variables jointly matter, estimate the model with them (unrestricted) and without them (restricted):
F = [(SSE_restricted − SSE_unrestricted) ÷ q] ÷ [SSE_unrestricted ÷ (n − k − 1)]
Example: SSE rises from 540 to 600 when two variables are dropped. F = (60 ÷ 2) ÷ (540 ÷ 56) = 3.11, just below the 5% critical value of 3.16: the two variables are not jointly significant.
Comparing models
- Adjusted R²: higher is better, but it doesn’t tell you whether variables are significant.
- Akaike information criterion (AIC) = n ln(SSE/n) + 2(k + 1). Preferred for forecasting.
- Schwarz Bayesian criterion (BIC) = n ln(SSE/n) + ln(n)(k + 1). Larger penalty; preferred for goodness of fit.
Lower AIC or BIC is better. Use them only to compare models of the same dependent variable and sample.
Exam focus
Fill in an ANOVA table, compute adjusted R², the overall F and a joint F, and choose between models using AIC and BIC.