able citing it, as it was being revised (and thus superseded) to appear as Greenwaid (1997) and Greenwald and Gillmore (1997a) alongside Marsh and Roche (1997). We believe that our reference to Haskell (1997) was a sufficient, appropriately documented, and scholarly representation of his interpretation of the grading-leniency literature, reliant as his summary was on the Greenwald material. Our intentionally inclusive reference to "such dubious but popular interpretations" ( Marsh & Roche, 1997, p. 1191 i.e., like those mentioned in the preceding sentence) encompasses Haskell's (1997) summary of the grading-leniency literature and Greenwald's grading-leniency work .that he cited.We subsequently outlined major weaknesses in the field studies and described other research that undermines claims that grading leniency represents a serious threat to the validity of SETs. In his revised version, Greenwald (1997) acknowledged that the critical reviews of the grading-leniency field studies "deserve serious consideration" (p. 1183), but avoided answering specific published criticisms, instead characterizing such critiques rather dismissively as "speculation." We maintain that these various weaknesses, and extensive contrary empirical evidence both new and old (see Marsh & Roche, 1997, 1998 provide a strong basis for characterizing as mere "speculation," any conclusion that reported group differences, where significant, actually represent effects of grading leniency. Following from brief comments in our postscript, Marsh and Roche (1998) document in greater detail our concerns (which have not diminished in relation to the revised versions) about Greenwald (1997), Greenwald and Gillmore (1997a), and Greenwald and Gillmore (1997b, the published version of Haskell's Greenwald & Gillmore reference). These studies draw causal conclusions from correlational data by comparing three (or four) straw-person theories with a pet theory. They are hardly "sophisticated" as Haskell (1997) claimed, and provide no "strong evidence" of grading-leniency bias.Haskell 's (1997, 1998) attempts to sideline the issue of SET validity are contradicted by his footnote, which acknowledges its centrality to his main thesis. His article relies on the dubious assumption that students in his anecdotal reports are, for whatever mysterious reason, not providing valid SETs, but are instead either exacting revenge or just plain wrong about what's good for them. It is not the nonstatistical or contextual nature of such views that defines them as "popular," but the fact that they find an uncritical, accepting audience among a segment of academics, despite flying in the face of the best available evidence and the most comprehensive and authoritative reviews.