Eyeball tests are often misleading, or fail to detect weak correlations (or deviations from model assumptions such as heteroskedasticity). That's why we check with more formal methods.