Jerome L. Myers

Regression analyses of repeated measures data in cognitive research.

Repeated measures designs involving nonorthogonal variables are being used with increasing frequency in cognitive psychology. Researchers usually analyze the data from such designs inappropriately, probably because the designs are not discussed in standard textbooks on regression. Two commonly used approaches to analyzing repeated measures designs are considered in this article. It is argued that both approaches use inappropriate error terms for testing the effects of independent variables. A more appropriate analysis is presented, and two alternative computational procedures for the analysis are illustrated.

Accessing the discourse representation during reading

The process of text comprehension requires the integration of the information in the sentence currently being read with information previously read. This, in turn, implies that information presented earlier in the text must be accessed. We present a view of that access process as one in which concepts and propositions in the discourse representation resonate in response to related elements in the current sentence, initiating a process that makes available a subset of the information in the representation. In support of our position, we summarize the effects of several variables that have been shown to affect the availability of information in the discourse representation, and we describe a simulation model of the hypothesized resonance process, together with the results of several applications of that model.

Degree of causal relatedness and memory

Abstract J. M. Keenan, S. D. Baillet, and P. Brown ((1984) Journal of Verbal Learning and Verbal Behavior , 23 , 115–126) varied the causal relation between two sentences in passages read by their subjects. Subsequent recall of one sentence cued by the other improved, and then deteriorated as the causal relatedness of the two sentences increased. The present experiments extended this work and replicated the basic finding of a quadratic relation between recall and causal relatedness. Several explanations are considered to account for these results. The long reading times together with relatively poor recall at low levels of causal relatedness argue against a pure processing effort model. Variations in the integration and elaboration of the representation of the sentence pairs would seem to better account for the relation of recall and causal relatedness. Several issues raised by this explanation are then briefly considered.

The role of causal connections in the retrieval of text

In two experiments, subjects read a series of passages, each containing two target concepts that appeared in widely separated positions within the passage. Following each passage, the time to retrieve each of these concepts was recorded. Several measures from both the Kintsch and van Dijk (1978) model and Trabasso and Sperry’s (1985) causal analysis were derived to predict retrieval time. Regression analyses showed that substantial proportions of variance were accounted for by measures derived from a causal analysis. Neither physical position nor measures based on the Kintsch and van Dijk model accounted for significant proportions of variance. Following Experiment 2, a reanalysis of O’Brien’s (1987) results revealed that the number of causal connections to an antecedent was a significant predictor of antecedent search time. Results are discussed in terms of a spread of activation through an integrated text representation in which causal connections play a major role.

MAINTAINING GLOBAL COHERENCE DURING READING

Previous research demonstrated a contradiction effect: although text currently being processed was locally coherent, readers were aware of inconsistencies between a protagonists actions and characteristics of the protagonists that were described earlier in the text and subsequently backgrounded. In the present experiments, not only the critical characteristics but also the protagonist were backgrounded. When the protagonist was reintroduced into the narrative in the context of carrying out an inconsistent action, the contradiction effect again occurred. In experiments 1-3, we examined the effects of several variables on the onset of the effect. In Experiment 4, we presented probes at various points in the passage with results tha supported he conclusion that the critical characteristics had been backgrounded before the sentence that contradicted them

Memory search without interference: The role of integration☆

Several researchers have shown that the time required to retrieve a sentence increases with the number of unrelated facts learned about concepts in that sentence. L. M. Reder and J. R. Anderson (1980, Cognitive Psychology, 12, 447–472) have argued that such fan effects also occur when the facts are integrated, provided that subjects must carry out a search of memory. In the present set of three experiments, we followed Reder and Andersons procedure but, in a highintegration condition, used facts that were causally linked. In the first experiment, recall and recognition accuracy were better when fan was six than when it was three, and this effect was more pronounced in a high- than in a low-integration condition. In the second experiment, the overall fan effect was negative for recognition time in the high-integration condition, whereas in the low-integration condition, we obtained the usual positive fan effect. In the third experiment, subjects learned the materials on their own to provide a better opportunity for them to integrate facts. All fan effects became smaller or more negative relative to those observed in the preceding study. We consider a class of models for the findings in which subjects use confirming and disconfirming evidence as a basis for early termination of search.

Anaphoric inference during reading.

Three experiments provide evidence that an anaphoric noun phrase reinstates its antecedent in the course of comprehension. Subjects read a series of texts each containing a target item. Immediately after the last line of each text, the item was probed using a recognition task in Experiment 1 and a naming task in Experiment 2. Subjects were faster to respond to the item when the last line contained an anaphoric reference to it than when the last line referred to a different item from the text. Additional control conditions ensured that the effect was not due to semantic priming and that the probed item was not in working memory when the last line was encountered. A third experiment suggested that previous evidence for reinstatement reflected interference from a change of topic in the last line rather than facilitation due to reinstatement of the probed item.

Connecting goals and actions during reading.

Three experiments showed that reading about a characters actions can reactivate a goal of the character stated earlier in the passage and backgrounded by intervening material. Subjects were slower to read a line describing an action that was inconsistent with a goal of the protagonist than they were to read about an action that was consistent with the goal, even though both lines were locally coherent. Goals were reactivated even when the intervening material did not describe attempts to achieve the goal (Experiment 2) and when the intervening material described another goal of the protagonist (Experiment 3). The results suggest that reading a sentence can reactivate relevant information from earlier in the text, even when the sentence is coherent with its immediate context and the reactivated information has been backgrounded by several lines of unrelated text.

Accessing distant text information during reading: Effects of contextual cues

In 5 experiments, we investigated conditions under which readers access distant information. Participants read texts containing 2 episodes; the first of which described a goal that was satisfied or unsatisfied. Following the 2nd episode, participants read a statement that reinstated some part of the goal context. Experiments 1 through 3 demonstrated that reactivation of unsatisfied goal information depended on the type of original goal context and whether it was fully reinstated. Experiments 4 and 5 demonstrated that, when the context was associated with both episodes, reinstatement of goal information from the distant target episode was successful only when the context was fully reinstated (Experiment 4). The results are consistent with models of comprehension that assume reactivation of relevant backgrounded information is achieved via a resonance process.

The effect of encoding task on memory for sentence pairs varying in causal relatedness.

Abstract Subjects read pairs of sentences varying in degree of causal relatedness and then recalled one sentence of each pair given the other as cue. When subjects read for comprehension, recall was a quadratic function of causal relatedness. When subjects wrote an elaboration sentence for each pair, recall did not vary with causal relatedness. When subjects studied each pair for 30 s, recall again varied with causal relatedness. When recall was delayed, performance dropped more steeply in the comprehension and study conditions than in the elaboration condition. The results support a model in which elaborations are the critical factor influencing the shape of the recall function. Other models based on strength or quality of links or on reconstructive strategies are also discussed.