The animacy (bias) effect in recognition: testing the influence of intentionality of learning and retrieval quality

ABSTRACT

The animacy effect, a memory advantage for animate/living over inanimate/non-living items, is well-documented in free recall, but unclear in recognition memory. This might relate to the encoding tasks that have been used and/or to an unequal influence of animacy on the processes underlying recognition (recollection or familiarity). This study reports a recognition memory experiment, coupled with a remember/know procedure. An intentional and two incidental learning conditions (one animacy-related and one animacy-unrelated) were used. No animacy effect was found in discriminability (A’) irrespectively of the encoding condition. Still, different mechanisms in incidental and intentional conditions conducted to said result. Overall, animates (vs. inanimates) elicited more hits and also more false alarms. Moreover, participants tended to assign more remember responses to animate (vs. inanimate) hits, denoting higher recollection for the former. These findings are suggestive of an animacy bias in recognition, which was stronger in the animacy-related encoding condition. Ultimate and proximate mechanisms underlying the animacy effect are examined.

KEYWORDS:

• Animacy
• Adaptive memory
• Recognition
• Intentional and incidental learning
• Remember/know

Acknowledgements

The authors thank Rui Lebre and Pedro Bem-Haja for their help in coding the online task and on Bayesian statistics, respectively. We also thank all the professors who provided us access to some of the sample. A special thanks to Prof. James S. Nairne for his invaluable review and comments on the drafts of this paper.

Disclosure statement

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

Data availability statement

The stimuli and instructions, as well as the data that support the findings of this study, are openly available in Open Science Framework: https://osf.io/utgpv/?view_only=None.

Notes

1 The first author coded the responses provided by participants regarding the recognition instructions and examples of remember and know responses (see Procedure section), as per the definition provided by Rajaram (1993). In case of doubts the second author was consulted. Examples of responses counted as correct and incorrect are available in Supplemental Material 2. Such exclusion criterion was applied in other studies using this procedure (e.g., Eldridge et al., 2002; see final Note of Gardiner & Java’s paper, 1990). This exclusion criterion was not implemented in previous animacy studies, but they all disclosed concerns regarding the confusion between remember and know judgements by the participants (Bonin et al., 2014; Komar, Mieth, Buchner, et al., 2023a; Rawlinson & Kelley, 2021).

2 Due to the random selection of the 50 words to be presented as targets (for each participant and the remaining 50 words as distractors) from the pool of 100 words, and the subsequent exclusion of some participants, some variability existed on each words’ presentation. Specifically, the percentage of times a given word was presented as a target varied between 39% and 63%, for the words “woman” and “revolver”, respectively. To explore the potential impact of such variability on our results, we conducted item-based analyses (i.e., by word; not by participant). This way, the dependent variables (e.g., hits, false alarms) were calculated as a function of the number of times each word was presented. Overall, the patterns of results resemble those reported in the manuscript, with some minor differences that do not change our main conclusions. Data by item are also available through OSF.

3 For comparison with prior studies (e.g., Leding, 2020), we report BF_Inclusion, which quantifies the strength of the evidence of a particular effect (Rouder et al., 2017); BF_Inclusion > 0 indicates evidence for the inclusion of a variable in the model, whereas a BF_Inclusion < 0 suggests its non-inclusion. BF₁₀ is reported along with t-tests and represents the evidence for the alternative hypothesis (H1) as compared to the null hypothesis (H0; Wagenmakers et al., 2011). For example, BF₁₀ ≥ 3 indicates that the results are at least three times more likely under H1 than under H0, while values equal or below 1/3 would provide evidence for the null hypothesis (Wagenmakers et al., 2018, 2011).

4 Hit and false alarm rates were calculated as the number of hits/false alarms obtained for animate and inanimate words, divided by 25 (the number of animate/inanimate targets/distractors; Macmillan & Creelman, 2005).

5 A’ was calculated as described by Stanislaw and Todorov (1999), who recommend reporting nonparametric measures of sensitivity. For comparability with previous studies, we also report the results for d’: the Animacy main effect, F(1, 145) = 0.39, p = .536, and the Animacy X Encoding Condition interaction, F(2, 145) = 0.76, p = .472, were non-significant. There was a significant Encoding Condition main effect, F(2, 145) = 51.39, p <.001, η_p² = .415. Independent t-tests revealed the same pattern as reported for A’.

6 Criterion C was calculated following Macmillan and Creelman (2005) and Stanislaw and Todorov (1999). Positive values indicate a tendency to respond "no". C was computed by adjusting the proportions of Hits = 1 to 1-1/2N and False Alarms = 0 to 1/2N, with N = 50 representing the number of targets or distractors, respectively.

7 For each participant, the proportion of remember and know responses was calculated as the number of remember and know responses given following a hit, divided by the number of targets. The sum of the proportion of remember and know responses equals the overall hit rate (Gardiner & Java, 1990; Mulligan et al., 2010). The same was done for the remember and know responses regarding the false alarms.

8 These analyses comprehended the corrected proportion of animate (and inanimate) remember false alarms (i.e., the number of animate [or inanimate] remember false alarms, divided by the total number of animate [or inanimate] false alarms provided by each participant). Some participants were not considered in these analyses as they committed no false alarms (which represented a division by 0). The same is valid for the follow-up analyses on the know false alarms.

9 We thank to Reviewer 1 for bringing this issue to discussion.

Additional information

Funding

This work was supported by the Portuguese Foundation for Science and Technology through a Doctoral Fellowship awarded to SBF (SFRH/BD/145097/2019, COVID/BD/153450/2024), a grant to JNSP (CEECIND/01914/2017), and multiannual funding to the William James Center for Research (UIDB/04810/2020).