It is important to note that with fixed-effects estimation, strict tests for the presence of effects are possible only for the last effects fitted. Thus, Table 2 shows that business-unit effects and industry-year effects are quite significant. However, because of the nesting in the model, it is not possible to fit industry or corporate effects after fitting business-unit effects. Put differently, in analysis-of-variance one cannot hold business unit effects 'constant' and then estimate industry or corporate effects. Thus, the significance attaching to the entry of 'corporation' in Table 2 (F-value = 6.13) does not necessarily mean that corporate effects are significantly different from zero. 16 Thus, analysis of variance or regression cannot reveal whether the apparent explanatory power of corporation is due to real corporate effects or whether it reflects differences among corporate returns induced by large uncontrolled business-unit effects. The same problem arises in interpreting the results on industry effects.