Respondents’ reports about the frequency of everyday behavior are often found to differ considerably when either low- or high-frequency response scales are used to record the answers. It has been hypothesized that the susceptibility to this type of response effect is determined by the cognitive accessibility of the respective target information in respondents’ memories. The first aim of the present paper is to test this hypothesis using two alternative, individual level indicators of the cognitive accessibility of information. These measures are the subjects’ self-reported response certainty and the time needed to answer the question under consideration. A second issue is how response certainties and response latencies should be transformed prior to data analysis in order to maximize their predictive power for response effects. Accordingly, the ability of untransformed measures to predict scale effects is compared with that of logarithmic, square-root and reciprocally transformed versions. The empirical results show that untransformed response certainties and response latencies are equally valid predictors of whether and to what extent subjects’ answers are affected by the presentation of response options. A square-root transformation is found to have no effect on both measures, whereas a logarithmic transformation slightly improves the validity of response certainties. In contrast, a reciprocal transformation proves to have a substantially positive effect on both measures and improves their ability to predict the reliability of respondents’ survey reports.