(Pieter Koele, Sponsor) Studies employing designs including a thentest were reviewed to examine to what extent they demonstrated a response shift, while controlling for subject bias by including a placebo control condition. A computer search was executed to detect articles that included the Howard, Ralph, et al. (1979) article in their reference list. Of 24 studies, 16 did not warrant any statement about the occurrence of a response-shift bias, relative to a subject bias, because they lacked a placebo control condition. Support for the subject bias hypothesis, relative to the response-shift hypothesis, was found in one of eight studies that did include a placebo control condition. With respect to these results, three supplementary considerations were taken into account. It was concluded that subject bias has not proved to be an unqualified alternative explanation of the response-shift interpretation in the reported studies.The research design most frequently employed to evaluate training effectiveness is the two-group pretestposttest design (Campbell & Stanley, 1966, Design 4) . Cronbach and Furby (1970)noted that for pretest and posttest scores to be comparable, a common metric must exist for the two sets of scores. In using self-report instruments, Howard, Ralph, et al. (1979)found that a response shift-a change in internal standard from pretest to posttest-renders pretest and posttest scores incompatible. To obtain selfreport measures of pretest and posttest levels of functioning with respect to the same internal standard, Howard, Ralph, et al. (1979) administered a retrospective pretest-a thentest-at the time of posttesting. Subjects were asked to report how they then perceived themselves to have been prior to the intervention. A significant mean difference between pre-and thenscores in the experimental condition would indicate a response shift . Since a response shift is dependent on the treatment, it should occur in the experimental and not in the control condition. If a response shift is demonstrated, a valid estimate of a treatment effect is given by the mean difference between post-and thenscores (Howard, Ralph, et al. , 1979). However, designs including a thentest may be more susceptible to subject bias resulting from subjects' perception of the demand characteristics of the experimental situation (Orne, 1969) than Design 4 of Campbell and Stanley (1966). Since the thentest is administered at the same time as the posttest, subjects may, in an attempt to indicate favorable results, adjust their thenscores in a downward direction. Consequently, a mean pre/then difference score may not reflect a shift in internal standard, but may represent subject bias instead.