The present experiment newly examined the effect of post-encoding elaboration on the production of inference-based memory errors, particularly for negative vs. neutral events. The main finding was that post-encoding elaboration increased the propensity to incur causal errors for negatively charged (but not neutral) events. Furthermore, the elaboration increased the propensity to incur gap-filling errors across the board. The results generally confirmed the expectations, although with some exceptions.
With regard to causal errors, the opposite patterns that emerged in the two groups were predicted on the basis of previous literature; indeed, the fact that participants who only performed the recognition task (without further re-elaboration) were less likely to incur negative than neutral false memories is consistent with the idea that emotional material encourage a more detailed and careful encoding, thus reducing distortions and false memories (Kensinger & Schacter, 2006b; Pesta et al., 2001). On the contrary, participants who re-elaborated the viewed episodes showed the opposite pattern. A two-step process was hypothesized to underlie this finding. First, it is known that emotionally charged events are more likely to be the focus of subsequent elaboration (Christianson & Engelberg, 1999; Walker et al., 2009), and indeed participants in the present study were more likely to mention the episodes having a negative outcome than those having a neutral outcome. Second, it is known that rehearsal of encoded material may lead to increased memory intrusions (e.g. Roediger &
McDermott, 1995; Zaragoza et al., 2011), likely due to boosted reconstructive processes, in an attempt to bring coherence to the complexity of an event (Bergman & Roediger, 1999).
Reconstructive processes may have been particularly relevant in the case of inference-based memory errors, which I examined in the present dissertation.
With regard to gap-filling errors, an across-the-board increase in the occurrence of false memories was observed. No interaction between group and valence emerged in this case, but it was expected; indeed, valence was manipulated through the outcomes of the episodes, which are temporally continuous and logically tied to the photographs representing the causal antecedents of the outcomes (i.e. to the causal errors), while they are less strictly related to the rest of the script. Nonetheless, it is interesting to note that post-encoding elaboration increased the tendency to produce false memories concerning aspects that fill the gaps between the scenes, thus confirming the notion that reconstructing an event leads in fact to manipulate and possibly distort the original memory trace (as first suggested by Bartlett, 1932). Consistently with this view, additional analysis showed that gap-filling errors (as well as hits) were more likely for episodes that were mentioned (and thus, to a certain extent, re-elaborated) than for those that were not mentioned in the written reports.
Some aspects still remain unclear, and raise new questions. First, it could be expected to observe a main effect of group not only for gap-filling errors, but also for causal errors; indeed, while the group-valence interaction in the latter case is of great interest, it is not clear why post-encoding elaboration did not increase, albeit to a smaller extent, also the probability of incurring neutral causal errors. Furthermore, although emotional episodes were more likely to be mentioned, being mentioned did not in turn directly increase causal errors probability. One hypothesis is that, since causal errors stem from an inferential elaboration concerning a very specific episodic aspect, an episode just being mentioned in its entirety is not a proper index of the fact that the specific causal episode has been elaborated. This should be investigate by more in-depth research. In general, however, a limitation of the present study is that the amount of evidence in favour of the effects, both in the case of causal and gap-filling errors, was not large. Furthermore, also the size of the effects (as it can be seen from the estimated probability in the figures, as well as from the odds ratios in Table 2.1) is limited; the estimated probability
of gap-filling errors goes from around .20 in the control group to around .30 in the post-encoding elaboration group, while the probability of causal errors in the negative episodes goes from just below .30 in the control group to just above .40 in the post-encoding elaboration group. This suggest that a large portion of these probabilities depend on other aspects, which will need to be investigated. Finally, in the present experiment valence and arousal were inherently confounded; showing negatively arousing events was decided in order to resemble ecological situations that could be forensic-relevant (i.e. memory of potential crimes);
however, future research may try to extend the results presenting positive events or low-arousal negative events.
Despite some limitations, the present study offers important suggestions, and practical implications can be derived. Indeed, while rehearsal and memory testing may help to improve long-term retention (e.g. Roediger & Karpicke, 2006), the present study indicates that free, unguided elaboration of witnessed events may have undesired effects, such as increasing inferential false memories. From a forensic point of view, this may indicate that repeated interviews should be avoided, especially with regard to negatively charged events; however, it may be interesting for future research to investigate how post-encoding elaboration (assuming that in certain cases it is inescapable) could be guided to minimize the occurrence of inference-based memory errors.
3. INDIVIDUAL DIFFERENCES IN THE OCCURRENCE OF INFERENTIAL