Many studies sustain that false memories depend on semantic elaboration since false memories tend to increase as the semantic elaboration of to-be-remembered information increases (11, 12). Moreover, studies on text comprehension have amply demonstrated that when reading stories, people seek to identify factors that link characters and events [e.g., Ref. (13)] in order to achieve coherence. This search for coherence can lead to memory errors, and accordingly, false memories tend to increase as semantic processing, aimed at gaining coherence, increases just as they do when video content is involved (14, 15).

False memories also seem to interact with the affective content of presented material. Numerous studies have demonstrated that production of false memories is modulated by the emotional content of the stimuli [e.g., Ref. (12, 16, 17)] although effects are complex and contradictory. On the one hand, affective content reduces false memories since individuals tend to focus on affective content more than neutral content. This, in turn leads them to encode and remember affective information better. In particular, the decrease in false memories for affective information may be because false memories for emotional stimuli are encoded with more contextual details and thus facilitate the distinction between presented and imagined stimuli (18, 19). In fact, many studies have shown that affective content reduces false memories (18) and leads to monitoring accuracy (12, 20).

On the other, affective content leads to an increase in false memories due to heightened elaboration processes and is further complicated by the fact that positive and negative content also seem to produce different effects on memory (21–25). Indeed, studies on memory recognition have reported an increase in false alarms to emotional stimuli compared to neutral stimuli. These researchers affirmed that affective information induces an attention bias that leads to both enhanced memory accuracy and increased the likelihood of recognizing an emotional stimulus regardless of prior exposure to it (16, 26, 27). Others affirm that false memories for affective content increase since emotions can also serve as organizational factors and can lead to conceptual relatedness. In general, the more stimuli are similar, the more they can be confused in memory. It particular, affective stimuli may also lead to confusion in memory and increase false memories for non-studied affective information since they are conceptually related in that they include emotions (28).

More specifically, an increasing number of studies has begun investigating false memories in individuals diagnosed with schizophrenia [for a review, see Ref. (29)]. Results however, are mixed. In general, studies have shown how patients with schizophrenia are more prone to false memories than controls (30). Others, using recognition tasks following the presentation of lists of words phonetically or semantically associated with specific unpresented target words [the Deese–Roediger–McDermott (DRM) task (31)], found no differences in the production of false memories between patients and controls or, in some cases, controls actually made more errors than patients (32–34). Elvevåg et al. (32) found that patients with schizophrenia made more errors (i.e., remember responses) but fewer false recognition errors at recall and argued that the production of false memories in patients may reflect poor memory which makes these population not particularly prone to false recollections. More recently, studies found that patients with schizophrenia are particularly prone to imagination inflation effects. That is, when schizophrenia patients repeatedly imagine an action they often remember the action as carried out (35, 36). According to the Source Monitoring Framework (SMF) (20), these errors occur because imagined events, typically accompanied by more operational attributes, are enriched with so many perceptual details that patients subsequently remember them as having been actually carried out. The same may happen when schizophrenia patients make inferences during story elaboration and lead them to make causal errors. Therefore, it is clear that research on false memories portray a complex pattern of memory errors in schizophrenia, indicating that different factors may differently mediate memory distortions.

Finally, the possible interaction between the affective valence of to-be-remembered material on memory in general, and on false memories in particular, has not been extensively studied in patients with schizophrenia. In particular, studies that adopted recognition tasks have demonstrated that patients with schizophrenia, just as normal individuals, show better performance for negative than positive words (37), while others found that patients remember positive events better than negative ones (38, 39). More recently, using the Remember/Know procedure, a number of studies found that patients with schizophrenia consciously recognized (remember responses) more positive than negative stimuli (40, 41) and recalled more positive than negative autobiographical memories compared to normal controls (42). Less is known, instead, about the interplay between affective information and false memory in schizophrenia.

The present study aims to examine the production of false memories in a group of high-functioning patients with schizophrenia and in a group of healthy adults in order to clarify mechanisms underlying the elaboration of scripted everyday events and the implications of affective content. In fact, the effects of affective content on memory distortions may have critical clinical implications since affective false memories may give rise to psychotic symptoms including hallucinations and delusional thinking (43). Moreover, the elaboration of affective content may also offer insights into aspects of reality monitoring deficits, namely qualitative source encoding errors (e.g., difficulties in encoding source features that typically accompany external vs. internal events) typically shown by patients with schizophrenia.

To this end, we adopted a modified version of the paradigm originally proposed by Hannigan and Reinitz (44) and successively adapted by Toffalini et al. (17) and Mirandola et al. (45) to include affective (both negative and positive) as well as neutral information. This paradigm includes pictorial material representing daily routines designed to investigate two types of memory errors that can occur during script elaboration for coherence and comprehension, plausible, script consistent errors, and causal errors deriving from erroneous inferences.

Based on previous findings (30), we expected patients with schizophrenia to be more susceptible to false memories than healthy controls with both types of errors. With regards to the affective valence of the outcomes on memory, we expected healthy controls, in line with Mirandola et al. (45), to show a general decrease in false alarms when exposed to episodes with affective outcomes compared to those with neutral outcomes. Although we did not make any specific predictions regarding errors in the patient group, given that schizophrenia patients generally do poorly in memory tasks and therefore are less prone to false memories, we may expect fewer errors in general. However, if schizophrenia patients show better memory performance in tasks that include affective information, we may find that false memories increase and be influenced by the affective valence of the outcome.

Our study adopted a 2(group: healthy adults vs. schizophrenia patients) × 2(type of error: plausible script vs. causal) × 3(valence: positive, negative, and neutral) mixed design.

We used nine episodes in our study. Episodes included waking up (i.e., typical morning routine before going to school), going shopping (i.e., young boy going grocery shopping with his mother), dating/meeting a friend (boy and girl meeting in the park), bike trip (i.e., girl going on a bike trip in a downtown area), rock climbing (i.e., boy climbing a wall), track competition (i.e., young girls getting ready for and performing a competition), coming back from a long trip (i.e., girl coming back by train from a trip and entering her home), playing games (i.e., video games in a bar), and a party (i.e., a girl welcoming guests and blowing out candles). Episodes were specifically designed to investigate two types of memory errors that can occur during elaboration for coherence and comprehension, and different affective outcomes (positive, negative, and neutral). Plausible errors (participants say “yes” to an unseen picture that represents a plausible action) are an index that participants actually processed the episodes and causal errors (participants say “yes” to an unseen picture inferred as the cause of the outcome of the episode) are an index of memory distortions determined by the affective outcome of the episode.

In particular, each episode was composed of 15 colored photographs depicting young actors engaged in typical everyday activities. Twelve of the 15 pictures served to create the encoded episode. The remaining three served as plausible errors during the recognition phase. We created two versions of each episode so that targets and plausible script errors were counterbalanced across participants. We also created a single photograph depicting a causal scene that could be inferred as the causal antecedent of the affective outcome for each episode.

In each episode, the 11th picture communicated the affective outcome of the episode. All of the other photographs were identical across the three affective versions of the episode. For example, the girl and bike on the ground and the driver, out of the car, with her hands on her head (negative), the bike at the entrance with the car already out of the entrance, and the girl getting off the bike (neutral), the girl getting off the bike and hugging a friend (positive) in the “bike” episode. Valence was counterbalanced across participants so that each participant saw three episodes for each affective outcome (Figure 1).

Eighteen independent judges rated valence and arousal levels of the pictures representing the negative, positive, and neutral outcomes using the Self-Assessment Manikin (SAM) scale (50). Paired-samples t-tests showed that outcomes significantly differed: negative outcomes had lower valence (M = 2.22, SD = 0.94), t(16) = 12.62, p < 0.001 and positive outcomes had higher valence (M = 7.67, SD = 0.79), t(16) = −15.04, p < 0.001, than neutral outcomes (M = 5.04, SD = 0.45). Furthermore, arousal was significantly higher in both negative (M = 7.40, SD = 0.88), t(16) = −8.74, p < 0.001, and positive outcomes (M = 6.90, SD = 0.93), t(16) = −12.92, p < 0.001, than in neutral outcomes (M = 3.89, SD = 1.30). Finally, although arousal was slightly higher in the negative than in the positive outcomes, the difference did not reach significance, t(16) = 1.64, p = 0.12.

Lastly, five unassociated pictures were presented at the beginning and five at the end of the encoding session to avoid primacy and recency effects. Episodes were not introduced by titles and there were no intervals between them.

For recognition, we used a unique sequence of 90 photographs composed of both old and new photographs. In particular, participants saw eight photographs per episode, four old ones from the encoded episode, three new ones that were plausible with the episode, and one never seen causal photograph that could be inferred as a plausible causal antecedent for the outcome of the episode. They also saw 18 single photographs, 9 old and 9 new. Photographs were presented randomly, irrespective of episode, with the constraint that participants never saw two causal photographs in subsequent positions and these were never in the first or last position, or to the slide immediately before the photograph depicting the affective outcome. All actors gave written informed consent to the original authors of the modified paradigm (17, 45) and these authors gave us permission to use the pictorial material.

We tested all participants individually in two sessions. In the first session, we collected demographic information and administered psychological general screening tests. In the second session, participants completed the false memory experimental task. See Figure 1 for a graphical representation of stimuli presentation.

A preliminary analysis concerned accuracy. To this end, we calculated two separate accuracy scores, one for unassociated photographs and one for episode photographs. Accuracy scores, in both cases, were calculated as HITs–FA. For episode accuracy scores, false alarms were relative to the episodes regardless of the type of error. We found a significant main effect of group F(1,46) = 41.78, p < 0.001, η² = 0.48 because schizophrenia patients were less accurate than controls with unassociated photographs (0.54 vs. 0.85). With regards to accuracy for photographs, we found a significant main effect of group F(1,46) = 56.05, p < 0.001, η² = 0.55 because schizophrenia patients were, in general, less accurate than controls with photographs (−0.05 vs. 0.41). The main effect of valence was also significant F(2,92) = 5.79, p < 0.01, η² = 0.11. An analysis post hoc LSD confirmed that participants were more accurate when the outcome of the episode was positive (0.24) or negative (0.21) with respect to when the outcome was neutral (0.09). There were no differences in accuracy for the episodes with positive or negative outcomes (p = 0.44).

The principal analysis concerned memory errors, calculated as mean proportions [as done in previous studies, e.g., Ref. (12, 51)], in reporting new pictures as old. Table 2 shows the mean proportions of causal and plausible errors.

A 2(group: AD vs. controls) × 2(type of error: plausible vs. causal) × 3(valence of outcome: positive vs. negative vs. neutral) mixed analysis of variance (ANOVA), with the last two factors as within participant factors, on the proportions of false memory as the dependent measure, revealed a significant group effect, F(1,46) = 7.17, p < 0.05, η² = 0.13 showing that schizophrenia patients made more errors than healthy controls. The main effect of type of error was also significant revealing that plausible errors were more frequent than causal errors, F(1,46) = 19.82, p < 0.001, η² = 0.30. The main effect of valence of consequence was significant F(2,92) = 8.82, p < 0.001, η² = 16. Post hoc LSD tests revealed how, in general, participants made more errors when the episode contained a neutral outcome compared to a positive (p < 0.001) or negative outcome (p < 0.01). There were no differences between errors in the negative and positive outcome episodes. The effect was specified by a significant 3-way interaction, F(2,92) = 3.20, p < 0.05, η² = 0.07. LSD post hoc comparisons revealed that neutral plausible errors were more frequent than affective ones in schizophrenia patients (p < 0.01) whereas there were no differences between affective and neutral causal errors. Differently, in controls, LSD post hoc comparisons revealed that there were no differences between neutral and affective plausible errors while neutral (p < 0.001) and negative (p < 0.05) causal errors were more frequent than positive ones (Figure 2).

Last, we used Pearson’s correlations, as typically done in studies regarding schizophrenia patients and false memories, in order to exclude possible correlations between false memories, clinical symptoms, and antipsychotic medication (chlorpromazine equivalent) that would confound results. No significant correlations were observed between chlorpromazine equivalent and false memories (plausible and causal errors; all >0.05).