Date of Award
Doctor of Philosophy (PhD)
This thesis examines people's estimates of the number of times events have occurred. Specifically, the thesis investigates how frequency estimates for pairs of related events differ from estimates for pairs of unrelated events. Previous research on "illusory correlation" (Chapman, 1967) has led to the conclusion that people show a bias that causes them to systematically and grossly overestimate the frequency or correlation of related pairings relative to unrelated pairings. The introduction of this thesis presents some empirical and theoretical grounds to question this characterization of "illusory correlation". The contention is that the theories and the existing body of research about frequency estimation are at odds with the conclusion that there is an overall bias. The introduction develops a theoretical view which predicts that frequency estimates for related pairs will show lower sensitivity and worse discriminability than estimates for unrelated pairs. There are also good reasons to suppose that related and unrelated pairs will not differ in the overall average magnitude of the frequency estimates they each receive.
An important consequence of difference in sensitivity or frequency discrimination is that such differences can look like magnitude differences (a bias) if only a small range of actual frequencies is examined. It is possible that the characterization of illusory correlation as a response bias resulted from a failure to examine a wide enough range of actual frequencies.
The first experiment demonstrates that the important difference between frequency estimates for related and unrelated events is lower sensitivity for estimates of related pairings. The second experiment provides evidence that this sensitivity difference occurs because subjects treat related and unrelated pairs differently during study. Basically, this difference in encoding strategy is characterized as attention to the general, categorical or semantic features of related events, and attention to spatial, temporal, episodic characterisitics of unrelated events. Because general, semantic encoding is less useful as a basis for later frequency judgments than is specific, episodic encoding, frequency estimates for related pairs are less sensitive to actual frequency than are estimates for unrelated pairs.
The next two experiments demonstrate that the conclusions one draws about frequency estimates depend importantly on the relationship between the demands of the final test and the nature of the subjects' encoding strategy. If, for example, the test of frequency judgment is a task which allows subjects to make advantageous use of the association between the members of related pairs, results consistent with a response bias view are obtained.
The fifth experiments extends the findings to judgments of conjoint frequency. Again, the important result is that unrelated pairings show higher frequency discriminability and sensitivity than related pairings. If only low actual correlations are examined, however, results are obtained that look like the operation of a bias. The sixth experiment shows that the results reported in this thesis cannot be attributed to the semantic association between pair members, per se. Instead, the encoding strategy together with the demands of the frequency test are crucial. Finally, the seventh experiment extends the analysis to still other encoding strategies and frequency tests, and confirmation of the main theoretical account is obtained.
In the final section, theoretical issues are re-examined. A union among traditional theories of frequency estimation is proposed. In addition, the theoretical position advocated in this thesis is discussed in the context of more general approaches to human memory.
Harris, Grant Thomas, "Frequency Judgments for Related and Unrelated Events" (1981). Open Access Dissertations and Theses. Paper 574.
McMaster University Library