Caren M. Rotello

Integrating Text and Pictorial Information: Eye Movements When Looking at Print Advertisements

Viewers looked at print advertisements as their eye movements were recorded. Half of them were told to pay special attention to car ads, and the other half were told to pay special attention to skin-care ads. Viewers tended to spend more time looking at the text than the picture part of the ad, though they did spend more time looking at the type of ad they were instructed to pay attention to. Fixation durations and saccade lengths were both longer on the picture part of the ad than the text, but more fixations were made on the text regions. Viewers did not alternate fixations between the text and picture part of the ad, but they tended to read the large print, then the smaller print, and then they looked at the picture (although some viewers did an initial cursory scan of the picture). Implications for (a) how viewers integrate pictorial and textual information and (b) applied research and advertisement development are discussed.

Sum-Difference Theory of Remembering and Knowing: A Two-Dimensional Signal-Detection Model.

In the remember-know paradigm for studying recognition memory, participants distinguish items whose presentations are episodically remembered from those that are merely familiar. A one-dimensional model postulates that remember responses are just high-confidence old judgments, but a meta-analysis of 373 experiments shows that the receiver operating characteristic (ROC) curves predicted by this model have the wrong slope. According to the new sum-difference Theory of remembering and knowing (STREAK), old items differ from new ones in both global and specific memory strength: The old-new judgment is based on a weighted sum of these dimensions, and the remember- know judgment is based on a weighted difference. STREAK accounts for the form of several novel kinds of ROC curves and for existing remember-know and item-recognition data.

Associative recognition: A case of recall-to-reject processing

Two-process accounts of recognition memory assume that memory judgments are based on both a rapidly available familiarity-based process and a slower, more accurate, recall-based mechanism. Past experiments on the time course of item recognition have not supported the recall-to-reject account of the second process, in which the retrieval of an old item is used to reject a similar foil (Rotello & Heit, 1999). In three new experiments, using analyses similar to those of Rotello and Heit, we found robust evidence for recall-to-reject processing in associative recognition, for word pairs, and for list-discrimination judgments. Put together, these results have implications for two-process accounts of recognition.

“Remembering” emotional words is based on response bias, not recollection

Recent studies have demonstrated that emotional stimuli result in a higher proportion of recognized items that are “remembered” (e.g., Kensinger & Corkin, 2003; Ochsner, 2000), leading to greater estimates of recollection by the dual-process model (Yonelinas, 1994). This result suggests that recognition judgments to emotional stimuli depend on a recollection process. We challenge this conclusion with receiver operating characteristic (ROC) curve data from two experiments. In both experiments, subjects studied neutral and emotional words. During the recognition test, subjects made old-new confidence ratings as well as remember-know judgments. Four models of remember-know judgments were fit to individual subjects’ data: two versions of a one-dimensional signaldetection-based model (Donaldson, 1996; Wixted & Stretch, 2004), the dual-process model (Yonelinas, 1994), and the two-dimensional signal-detection-based model known as STREAK (Rotello, Macmillan, & Reeder, 2004). Consistent with the literature, we found that emotion increases subjective reports of “remembering.” However, our ROC analyses and modeling work reveal that the effect is due to response bias differences rather than sensitivity change or use of a high-threshold recollection process.

Theremember response: Subject to bias, graded, and not a process-pure indicator of recollection

Recognition memory judgments have long been assumed to depend on the contributions of two underlying processes: recollection and familiarity. We measured recollection with receiver-operating characteristic (ROC) data and remember-know judgments. Under standard remember-know instructions, the two estimates of recollection diverged. When subjects were told they might need to justify theirremember responses to the experimenter, the two estimates were more likely to agree. The data support the conclusion thatremember responses are generally based on a continuous underlying process but that specific task instructions can produce data that appear consistent with a high-threshold recollective process. Models based on signal detection theory provide a better account of these data than does the dual-process model (Yonelinas, 1994) or process-pure interpretations.

Sources of bias in the Goodman-Kruskal gamma coefficient measure of association: implications for studies of metacognitive processes.

In many cognitive, metacognitive, and perceptual tasks, measurement of performance or prediction accuracy may be influenced by response bias. Signal detection theory provides a means of assessing discrimination accuracy independent of such bias, but its application crucially depends on distributional assumptions. The Goodman-Kruskal gamma coefficient, G, has been proposed as an alternative means of measuring accuracy that is free of distributional assumptions. This measure is widely used with tasks that assess metamemory or metacognition performance. The authors demonstrate that the empirically determined value of G systematically deviates from its actual value under realistic conditions. A distribution-specific variant of G, called G-sub(c), is introduced to show why this bias arises. The findings imply that caution is needed when using G as a measure of accuracy, and alternative measures are recommended.

Memory strength and the decision process in recognition memory

We investigated the role that memory strength plays in the decision process by examining the extent to which strength is used as a cue to dynamically modify recognition criteria. The study list consisted of strong and weak items, with strength a function of study duration or repetition. The recognition test list was divided into two consecutive blocks; strong items appeared in one block, weak items in the other. If the change in item strength across blocks leads to a shift in criterion, the false alarm rate should change accordingly. In four experiments, the false alarm rates did not change across blocks, even when the difference between the strong and the weak items was magnified and marked with semantic cues. However, the strength of the items in the first test block affected the false alarm rate. Thus, strength cues influence initial criterion placement but fail to induce criterion shifts following permanent and even dramatic changes in item strength. These null findings are contrasted with those in a fifth experiment, in which accuracy feedback produced dynamic criterion shifts.

Measures of sensitivity based on a single hit rate and false alarm rate: The accuracy, precision, and robustness of′,A z, andA’

Signal detection theory offers several indexes of sensitivity (d’,Az, andA’) that are appropriate for two-choice discrimination when data consist of one hit rate and one false alarm rate per condition. These measures require simplifying assumptions about how target and lure evidence is distributed. We examine three statistical properties of these indexes: accuracy (good agreement between the parameter and the sampling distribution mean), precision (small variance of the sampling distribution), and robustness (small influence of violated assumptions on accuracy). We draw several conclusions from the results. First, a variety of parameters (sample size, degree of discriminability, and magnitude of hits and false alarms) influence statistical bias in these indexes. Comparing conditions that differ in these parameters entails discrepancies that can be reduced by increasing N. Second, unequal variance of the evidence distributions produces significant bias that cannot be reduced by increasing N—a serious drawback to the use of these sensitivity indexes when variance is unknown. Finally, their relative statistical performances suggest thatAz is preferable toA’.

Interpreting the effects of response bias on remember-know judgments using signal detection and threshold models

In recognition memory experiments, the tendency to identify a test item as “old” or “new” can be increased or decreased by instructions given at test. The effect of such response bias on remember-know judgments is to change “remember” as well as “old” responses. Existing models of the remember-know paradigm (based on dual-process and signal detection theories) interpret this effect as a shift in response criteria, but differ on the nature of the dimension along which the changes take place. We extended the models to account simultaneously for remember-know and confidence rating data and tested them using old-new (Experiment 1) and remember-know (Experiment 2) rating designs. Quantitative fits show that the signal detection models provide the best overall description of the data.

Type I error rates and power analyses for single-point sensitivity measures

Experiments often produce a hit rate and a false alarm rate in each of two conditions. These response rates are summarized into a single-point sensitivity measure such as d’, and t tests are conducted to test for experimental effects. Using large-scale Monte Carlo simulations, we evaluate the Type I error rates and power that result from four commonly used single-point measures: d’, A’, percent correct, and γ. We also test a newly proposed measure called γinC. For all measures, we consider several ways of handling cases in which false alarm rate = 0 or hit rate = 1. The results of our simulations indicate that power is similar for these measures but that the Type I error rates are often unacceptably high. Type I errors are minimized when the selected sensitivity measure is theoretically appropriate for the data.