Academic journal article Memory & Cognition

Remembering the News: Modeling Retention Data from a Study with 14,000 Participants

Academic journal article Memory & Cognition

Remembering the News: Modeling Retention Data from a Study with 14,000 Participants

Article excerpt

A retention study is presented in which participants answered questions about news events, with a retention interval that varied within participants between 1 day and 2 years. The study involved more than 14,000 participants and around 500,000 data points. The data were analyzed separately for participants who answered questions in Dutch or in English, providing an opportunity for replication. We fitted models of varying complexity to the data in order to test several hypotheses concerning retention. Evidence for an asymptote in retention was found in only one data set, and participants with greater media exposure displayed a higher degree of learning but no difference in forgetting. Thus, forgetting was independent of initial learning. Older adults were found to have forgetting curves similar to those of younger adults.

Starting with Ebbinghaus (1885), memory researchers have attempted to find mathematical functions that might describe the shape of the retention curve. Some of the proposed functions were purely descriptive (e.g., the exponential, power, or logarithmic curves), whereas others were based on more or less detailed models of memory (Chessa & Murre, 2002; Wickelgren, 1974; Wickens, 1999). Both types of functions have been successfully fitted to large numbers of retention curves.

Nevertheless, many questions surrounding the retention curve are still unanswered. For example, it is still unclear whether the rate of forgetting is or is not independent of initial learning (Bogartz, 1990; Loftus, 1985; Slamecka & McElree, 1983) or whether older adults forget faster than do younger adults (Brainerd, Reyna, Howe, & Kingma, 1990; Cohen, Stanhope, & Conway, 1992; Wheeler, 2000). One reason for the long life of these controversies is disagreement about what would constitute a proper answer to the questions; in particular, how to measure and compare rates of forgetting has been hotly debated. Some researchers have suggested that a rate of forgetting is only meaningfully measured within a model of retention (Bogartz, 1990; Rubin & Wenzel, 1996), and in that case, whether or not different conditions exhibit the same level of forgetting becomes a question of whether a decline parameter has the same value when the model is fitted to those conditions.

Unfortunately, whether two conditions yield the same decline parameter value is not independent of the forgetting function used: Conclusions about parameter values are often bound by the model in which the parameters in question play a role (Rubin & Wenzel, 1996). Therefore, an ideal study of (e.g.) the dependence of forgetting on initial learning would fit several models to the data, so that dependence or independence could be corroborated according to different models.

In this article, we will present a study in which retention for news events was tested for around 14,000 participants and 500,000 data points. The participants were Internet volunteers who could log into a Web site after giving relevant personal details, and take a test in which they answered questions about news events, such as What was the name of the American country singer who died on September 12, 2003? (Question 1,430). Our primary goal while developing the Web site was to create a new retrograde amnesia test by submitting news questions to Web controls to test the appropriateness of the questions for inclusion in the test itself (Meeter, Murre, & Janssen, 2005). However, the control data are interesting in their own right and can be used to study retention and forgetting. For each participant, 30 or 40 questions were sampled concerning news events that had occurred at different moments in time. In this way, retention was measured at intervals ranging from a single day to up to 2 years.

The sheer size of the data set in the study allowed models to be fitted to the data and rejected with a high degree of precision. Several retention functions were fitted to the data, and two of them are of particular interest: the memory chain model (MCM) and what we will here refer to as the extended Weibull model. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.