In One-Factor ANOVA, Both Mean Square for Treatments (MSTr) and Mean

In one-factor ANOVA, both mean square for treatments (MSTr) and mean square for error (MSE) are unbiased estimators for estimating the common population variance $\sigma ^ { 2 }$ when __________, but MSTr tends to overestimate $\sigma ^ { 2 }$ when __________.