Frequency Estimation

We report two experiments about how people estimate the frequency of event properties when explicitly presented (e.g. spinach-GREEN) and implicitly (e.g. spinach). In Experiment 1, verbal reports indicated that, for explicitly presented properties, participants used several retrieval- and impression-based strategies and were relatively accurate. Implicitly presented properties led to “off-target retrieval” which brought to mind more instances of non-target than target properties and degraded estimates. A third group estimated the frequency of taxonomic categories (e.g. furniture) much as the explicit property group did suggesting that people can use properties to organize remembered events. In a second experiment, estimation time patterns underscored the Experiment 1 results and eliminated reactive verbal reports as an explanation. Off-target retrieval was both ineffective and slow.

When people answer survey questions of the form “During the past month, how many times did you …?” their responses provide valuable data for researchers and policy makers. Yet the way respondents produce their answers to these “behavioral frequency questions” is not well understood. This article demonstrates that survey respondents can use an array of distinct estimation strategies, depending on what information is available in their memories. The kinds of event information that people use is related to factors such as the regularity of occurrence, similarity of one episode to the next, and frequency. In a study conducted as a telephone survey, respondents’ verbal reports and response time patterns indicate that they usually answer behavioral frequency questions by either retrieving and counting episodes, retrieving or estimating rates of occurrence, or converting a general impression of frequency into a numerical quantity. The third strategy should be of particular concern to survey researchers because respondents provide a quantitative estimate without any relevant numerical knowledge. The set of strategies and the factors that influence their use are integrated into a statistical model that could help survey practitioners improve data quality and memory researchers broaden their perspective.