The feature boost in false memory: the roles of monitoring and critical item identifiability

ABSTRACT

The feature boost refers to increased false memories for word lists that are both associatively and categorically (C + A) related to a non-presented critical item (CI) relative to lists that are only associatively (NC-A) related [Coane, J. H., McBride, D. M., Termonen, M.-L., & Cutting, J. C. (2016). Categorical and associative relations increase false memory relative to purely associative relations. Memory & Cognition, 44(1), 37–49. doi:10.3758/s13421-015-0543-1]. We explored the replicability of the feature boost and its dependance on monitoring processes by explicitly warning participants about the nature of the lists or by asking participants to guess the CI (implicit warning). Overall, the feature boost was replicated. Guessing performance was higher for C + A lists than for NC-A lists. Explicit warnings were equally effective for both list types in reducing false memory relative to recall and to a no-recall math condition. When the CI was not guessed or recalled, the feature boost emerged. However, when the CI was guessed or previously recalled, false alarms did not differ as a function of list type. The feature boost seems to be driven in part by differences in the identifiability of the CI, such that CIs related to C + A lists are harder to identify and thus reject. These results suggest that differences in monitoring processes that are sensitive to CI identifiability contribute to the effect.

KEYWORDS:

• False memory
• monitoring
• DRM paradigm
• feature boost

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The design used by Coane, McBride et al. (2016) and in the present study does not allow us to fully isolate the effects of similarity from those of association. We acknowledge this is the case; however, given the difficulty of developing lists that fully cross associative strength and feature similarity, we opted to hold BAS constant and examine the additive effects of shared features.

2 Some types of relations provide preferential or stronger activation of the CI relative to others (Cann et al., 2011). In particular, situation features (Wu & Barsalou, 2009), which include semantic properties such as an object’s function (e.g., clothes-wear), action (e.g., horse-gallop), or typical location (e.g., car-garage), taxonomic relations (i.e., category membership), and synonyms have emerged as primary predictors of both BAS and false recall (Cann et al., 2011). This suggests that the semantic content of the lists is predictive of CI activation and subsequent false recall and recognition. It is worth noting, however, that although Cann et al. classified these types of features as “semantic”, relations such as function and location are also strongly associative – thus, in our lists, garage is in the NC-A list, not the C + A list for car and wear would also be in the NC-A list for clothes. So, this distinction, although clearly important, is not as narrowly focused as ours is on “shared taxonomic features”.

3 Sample size was determined based on the samples in Coane, McBride et al.’s (2016) study, where approximately 80 participants were tested in each experiment. In addition, an analysis with G*Power (Erdfelder, Faul, & Buchner, 1996) showed that the sample sizes in the current experiments were more than sufficient to achieve power greater than 90% based on the effect size found in Coane et al.’s study. Thus, in the present studies, we aimed for a minimum of 80 participants in the experiment. In addition, participants were recruited for each between-subjects condition in similar numbers from the two samples.

4 The full set of stimuli with lexical characteristics for all lists is available on the website of the first author at http://web.colby.edu/memoryandlanguagelab/.

5 To ensure the pattern of performance as a function of prior guessing was not an artifact of the fact that only 45 participants’ data were included in this analysis, we examined the pairwise comparisons in the entire sample. Between 50 and 59 participants’ data were included in the comparisons. The effect of list type for correctly guessed CIs was not significant, t(58) = 1.72, p = .09, whereas the robust effect of list type for non-guessed CIs was still significant with the larger sample, t(49) = 3.73, p = .001. Thus, the effects of prior guessing on false recognition emerged in the whole sample for whom data were available.

Additional information

Funding

This work was supported by funding from a James S. McDonnell Foundation Understanding Human Cognition [grant number #220020426] awarded to JHC. The funding agency had no input on the study design, data analysis, or writing. Data will be available upon request from the first author.