THEAMERICAN ECONOMIC REVIEW JUNE 1985 ⊥Mode 46 Firm Effects Only try Effects Only R2=.2644;R2=0|o6 987 <.OoOI 898 2729 0004 Fir m Industry Effects Firm 8 share Effects ndustry a Effects R-=4922;R=644 R-=.2670;R=.0B34 (氵=1523 =.2304 0035 <OOOI 9035 Firm, Industry, a share Effe 4962;R=,|702 (9=.2359) stricted models excluding one or more of the for firm effects(arrows pointing to the right three effects with which we are concerned. in Figure 1). These data imply that firm The values of the ordinary and adjusted r effects simply do not exist. In the absence of statistics are shown, 2 along with the esti- industry effects, the null hypothesis that the mates of y obtained from models with a realized as are identical can be rejected at share effect. Each arrow corresponds to the the 29.2 percent level (no share effect)or the imposition of a restriction that one of the 27.3 percent level(share effect present ). These hree effects discussed above is absent; the results might lead a Bayesian analyst with a number next to each arrow is the probability strongly managerial prior to accept the ex level at which a standard F-test rejects that istence of firm effects. But both tests con- restriction. These num ers are referred to ducted in the presence of industry effects simply as P-levels in what follows produce F-values less than unity, which pro- All the high P-levels in Figure 1, which vide absolutely no support for the existence ndicate failure to reject the null hypothesis of firm effects. Firm effects seem to approach at conventional levels, are generated by tests significance only when firm-specific dummy variables serve as proxies for industry effects When industry effects are controlled for, firm The adjusted R is equal to 1-[s-(e)/-(r). effects fade into insignificance. The absence itself, correspond to changes in an unbiased estimator of share effects indicates that firm efects do not he fraction operate through the revisionist mechanism to