More than meets the eye: emotional stimuli enhance boundary extension effects for both depressed and never-depressed individuals

ABSTRACT

Boundary extension is a memory phenomenon in which an individual reports seeing more of a scene than they actually did. We provide the first examination of boundary extension in individuals diagnosed with depression, hypothesising that an overemphasis on pre-existing schema may enhance boundary extension effects on emotional photographs. The relationship between boundary extension and overgeneralisation in autobiographical memory was also explored. Individuals with (n = 42) and without (n = 41) Major Depressive Disorder completed a camera paradigm task utilising positive, negative, and neutral stimuli. Across all participants, positive (d = 0.37) and negative (d = 0.66) stimuli were extended more than neutral stimuli. This effect did not differ between depressed and never-depressed participants. Across all participants, images containing objects were extended more than images containing faces. An association was also evident between extension effects in memory for perceptual space and extensions of autobiographical memory across time.

KEYWORDS:

• Boundary extension
• depression
• self-schema
• autobiographical memory
• scene perception

The way we encode perceptual experience critically influences how we interact with the world around us, and our understanding of our physical and social position in that world. However, memory is prone to distortion. Boundary extension is a memory error in which an individual reports seeing more of a scene than they actually did (Intraub & Richardson, 1989). For example, if shown a photograph of a toy on a stair, participants will report seeing more stairs than were shown. Theoretical models propose that a mental representation of a scene is derived by combining a number of sources, such that incoming sensory information is integrated with egocentric schema (that is, pre-existing models of the self and world), notably of space and the relation of incoming information to the self (e.g. Intraub, 2010). This combination of sensory perception with pre-existing schema is thought to underlie boundary extension effects (Bainbridge & Baker, 2020; Intraub, 2010), as an over-reliance on schema (relative to sensory input) extrapolates the view of a presented scene, leading the individual to mistakenly believe that they saw more of the scene. The need to reconcile sensory input with pre-existing schema is a defining feature of predictive processing frameworks of perception and cognition (Clark, 2013), which have been applied to depression (Kube et al., 2020). Here, we were interested in whether the experience of recurrent major depressive disorder, which is characterised by strongly consolidated, pre-existing models of the self and world as highly negatively valenced, may accentuate boundary extension effects when incoming stimuli are negative in nature.

There are a number of reasons why this may be the case. The presence of strong (and easily activated) negative schema in depression (Beck et al., 1979) may increase the likelihood that schema are over-emphasised in mental representation of scenes containing negative stimuli (for detailed discussion see (Kube et al., 2020)). Within the predictive processing models developed to account for the integration of incoming sensory input with existing models of the self and world, this would be consistent with a relative increase in the precision of pre-existing models relative to information from sensory input (e.g. active inference; (Constant et al., 2022; Friston et al., 2017)). In essence, depressed individuals make negative perceptual predictions about the sensory world that are over-extended (Clark, Watson, & Friston, 2018), and they may be subsequently more likely to encode incoming information in a manner that supports those negative, overgeneralised predictions. Indeed, in another mnemonic domain, autobiographical experiences that are negative in nature are frequently recalled in an overgeneralised manner during depressive states (Williams & Dritschel, 1988). Given the overlapping neural underpinnings of memory for perceptual and autobiographical experiences (Burgess et al., 2002), and the role that memory for episodes plays in mental scene construction (Madore et al., 2019), we sought to provide the first evaluation of whether the overgeneralisation of negative memory content during depression may also be evident in memory for scenes. To begin to explore the mechanism of action for any such effect, we also evaluated whether the strength of egocentric schemas, indexed here as positive and negative self-schemas defined by Beck’s cognitive model of psychopathology, related to boundary extension effects on positive and negative stimuli.

This pre-registered experiment administered the camera distance paradigm (Intraub et al., 1992) to participants with recurrent major depression and those with no prior experience of depression or other mental health difficulties. Our primary hypothesis predicted a significant group by valence interaction. We predicted never-depressed participants would demonstrate a larger boundary extension effect for positive relative to negative and neutral stimuli, consistent with positive overgeneralisation biases in good mental health (Mezulis, Abramson, Hyde & Hankin, 2004) and prior reports of boundary extension on positive visual stimuli in the mentally well, relative to negative and neutral stimuli (Ménétrier et al., 2013). However, there is a separate body of literature which suggests that mentally-well individuals may demonstrate no valence bias (e.g. Emery & Hess, 2008) or a negativity bias (e.g. Charles et al., 2003) in recognition and free recall of images (although this is influenced by age of the viewer; e.g. Charles et al., 2003; Emery & Hess, 2008). Indeed, there is some prior literature to suggest that boundary extension may occur for images of any emotional valence, with one caveat being that neutral control conditions have not always been utilised (e.g. Mathews & Mackintosh, 2004; Beighley et al., 2019). There is therefore an apparent need to ascertain whether a valence bias is evident in boundary extension in mentally-well individuals.

In depressed participants, if boundary extension is influenced by the strength of egocentric schema, a larger boundary extension effect would be expected for negative relative to positive and neutral stimuli. Alternatively, depressed individuals may demonstrate increased focus on a negative central object and subsequently boundary restriction effects for negative relative to positive and neutral stimuli, as in those with subclinical levels of anxiety and posttraumatic stress (Takarangi et al., 2016). We also predicted a positive correlation between the self-reported strength of positive and negative self-schema, and the degree of boundary extension on positive and negative stimuli, respectively. We anticipated this relationship would be strongest between negative self-schema and extension of negative stimuli in depressed participants.

Our secondary hypothesis aimed to provide an initial evaluation of whether boundary extension is associated with overgenerality in autobiographical memory. We hypothesised that the magnitude of boundary extension would be correlated with overgeneral memory on the Autobiographical Memory Task (Williams & Broadbent, 1986), such that a larger degree of boundary extension would be associated with more overgeneral autobiographical memory.

Method

Participants

From August 2019 to March 2020, Never-depressed individuals (n = 52) and those (n = 42) experiencing recurrent depressive disorder with a current Major Depressive Episode were recruited from the MRC Cognition and Brain Sciences Unit’s volunteer panel. The volunteer panel consists of individuals who responded to online and print adverts to participant in psychological research. Inclusion criteria were aged 18–70 years, fluency in English, and no known (self-reported) cognitive or neurological impairment, or active psychosis (determined by clinical interview). Power analysis estimating the effect size (d = 0.86, two tailed, α = 0.05) for emotional valence on boundary extension (Ménétrier et al., 2013) indicated 20 participants per group would provide 95% power to examine within-subject valence effects. As we had no feasible estimate of the between-group effect size between never-depressed and depressed participants, we aimed to increase upon this estimate by 50%, with a pre-registered minimum of 30 participants per group.

Depressive status was determined by the Structured Clinical Interview for DSM disorders – Fifth edition (SCID; First, 2015). Never-depressed participants self-reported no experience of any mental health difficulties, and prior history of depression was confirmed using the SCID. Family history of mental health issues was not taken. Any never-depressed participants scoring above 13 (indicating mild depression) on the Beck Depression Inventory-II (BDI-II) (Beck et al., 1996) were excluded (n = 9). An additional participant was excluded due to misunderstanding task instructions, yielding a final sample of 41 never-depressed and 42 depressed participants.

Camera distance paradigm

The experimental task was coded using jsPsych, a free library for JavaScript (de Leeuw, 2015). Participants were instructed that they would see a number of images, which they would be asked about later. During the Encoding block, the experimental stimuli were presented in a random order, each for 5 s. Each stimulus was preceded by a 500 ms fixation cross. During the Test block, stimuli were presented in the same order as the Encoding block. Participants were told that some of the images may appear closer or further away than originally viewed. Participants were asked to indicate on a Visual Analogue Scale (VAS) the extent to which the picture was closer/further away than the original. On the VAS, –100 was anchored as “Much closer than original” and +100 was anchored as “Much further than original”. Camera distance was explained in terms of the amount of background the participant could see (as in Takarangi et al., 2016). Participants also used a VAS to rate confidence in their judgement for each image (“Not at all confident” to “Very confident”), along with how much attention they were paying to the image when it was initially presented (“None at all” or “Very much”). All VAS stepped in increments of 0.1 and were recorded from –100 to +100. Participants used a mouse to move a marker (which began at a random point for each trial) on a sliding scale to indicate their response.

Stimuli

Stimuli were 24 IAPS photographs (Lang et al., 1997), to match prior studies evaluating boundary extension in the context of mental health (e.g. Beighley et al., 2019; Takarangi et al., 2016). Images were chosen and categorised based on valence, such that eight stimuli were neutral (e.g. scene containing a person sitting at a desk), eight stimuli were positively-valenced (e.g. scene containing a sports team celebrating a win), and eight stimuli were negatively-valenced (e.g. scene containing a person crying). Negative items were not specifically “depression-related” and were not of a traumatic nature (i.e. did not contain violence and were not rated as highly arousing) to avoid confounding effects. Within each valence category, half of the stimuli contained human faces that were clear to see, and the other half were comprised of objects. All images were object oriented, and taken from a participant perspective to replicate natural viewing conditions. Valence ratings for each image were obtained from the IAPS database. Neutral stimuli were less negative than negative stimuli (p < .001, d = 3.86) and less positive than positive stimuli (p < .001, d = 7.11). The positive and negative stimuli did not significantly differ in arousal ratings (p = .99, d = 0.004). The neutral stimuli were significantly less arousing than both the positive (p < .001, d = 4.89) and negative stimuli (p < .001, d = 5.56). There was no significant difference between images containing faces and objects in either valence, p = .64, or arousal, p = .70.

Measures

Autobiographical Memory Test – Alternating Instructions (AMT-AI; Dritschel et al., 2014)

In response to neutral, positive, or negative cue words, participants were asked to retrieve specific, single-incident memories which lasted for less than one day (i.e. a specific memory) to a block of six cue words and general memories (which summarise repeated events) for a block of six cues. For a block of twelve cues, the individual alternated between specific and general memories. Overgeneral memory was indexed as proportion of specific memories provided during the specific block (whereby a lower proportion reflects greater overgenerality). We used proportion of specific memories, rather than general memories, as in meta-analysis, number of specific memories has been shown to predict the course of depressive symptoms (Hallford et al., 2021). Proportion of correctly-retrieved specific memories also represents the ability to isolate an autobiographical episode in time, just as the camera paradigm task assesses the ability to isolate an image in space. We also indexed the number of extended memories produced across the entire task (i.e. errors in which a memory is produced which refers to an event lasting longer than a day), as these memories represent extension over time.

General cognitive measures

The Verbal Fluency Test (VFT, (Lezak, 1995)) was administered to index category and semantic fluency. The forward and backward versions of the Digit Span test from the Weschler Adult Intelligence Scale (Weschler, 2010) measured working memory. The National Adult Reading Test (NART, (Nelson, 1982)) provided a measure of verbal comprehension.

Mental health measures

The Beck Depression Inventory-II (BDI-II, (Beck et al., 1996)) assessed depressive symptomatology and had strong internal consistency (current α = .97). The Ruminative Response Scale (RRS, (Nolen-Hoeksema & Morrow, 1991)) was administered (current α = .95) as rumination may influence boundary extension (Beighley et al., 2019), and is elevated in depressed samples.

Positive and negative schemas

Questionnaires indexed positive and negative self-beliefs, or schemas, defined within Beck’s cognitive model of depression (Beck, 1967). The Dysfunctional Attitudes Scale (DAS, (Weissman & Beck, 1978)) measured negative schematic beliefs (e.g. It is difficult to be happy, unless one is good looking, intelligent, rich and creative) and had strong internal consistency (current α = .91). Beck Hopelessness Scale (BHS, (Beck, 1998)) indexed hopelessness about the future (e.g. things just don’t work out the way I want them to) (current α = .95). General Self Efficacy Scale (GSE, (Schwarzer & Jerusalem, 1995)) measured positive self-beliefs in the context of daily hassles (e.g. I can usually handle whatever comes my way) (current α = .93). Rosenberg Self Esteem Scale (RSES; Petersen, 1965) measured global self-worth (e.g. On the whole I am satisfied with myself), and had strong internal consistency (current α = .93).

Procedure

Hypotheses, method, and statistical analyses were preregistered, https://osf.io/kqybr. Cambridge Psychology Research Ethics Committee granted approval (PRE.2019.011). Participants provided written informed consent prior to completing the boundary extension task, an event segmentation task involving narratives (reported elsewhere, see pre-registration for details), the AMT-AI, Digit Span task, VFT, NART, and questionnaires. Sessions were in-person, lasted 90-120 minutes and participants were reimbursed £6.00/hour.

Results

Calculation of primary outcome

A participant’s mean boundary extension scores for each task condition were calculated by averaging the VAS ratings for positive, negative, and neutral trials, respectively. Using this method, scores below zero indicate a boundary extension effect and scores above zero indicate a boundary restriction effect. A score of zero would indicate accurate report of boundaries.

Sample characteristics

Sample characteristics are displayed in Table 1. All depressed participants were experiencing recurrent Major Depressive Disorder according to the SCID for DSM-5, with a minimum of two prior depressive episodes. The mean number of prior depressive episodes was “too many to count”. Twenty nine percent of depressed participants were currently receiving psychological therapy, which included counselling and support groups and cognitive behavioural therapies. Medication use included antidepressants, benzodiazepines and lithium, with 50% of those receiving medication taking more than one drug type. Demographic and general cognitive functioning measures did not significantly differ between groups, suggesting that results are not attributable to general cognitive impairment due to depression. Depressed participants demonstrated significantly higher BDI-II, t(48.15) = 14.64, p < .001, d = 3.20, and rumination, t(81) =  10.46, p < .001, d = 2.30, than never-depressed participants. The depressed group demonstrated stronger dysfunctional attitudes, t(81) =  5.53, p < .001, d = 1.22, and hopelessness, t(48.11) = 11.57, p < .001, d = 2.52, along with lower strength of positive schema regarding self-efficacy, t(67.78) =  −7.09, p < .001, d = 1.55, and self-esteem, t(81) =  −12.76, p < .001, d = 2.30. The groups reported paying a similar degree of attention to the images during the encoding phase (never-depressed M = 64.89, SD = 29.69; depressed M = 64.03, SD = 30.97), t(81) = .13, p = .90, d = 0.03.

Table 1. Mean (SD) sample characteristics by group.

Variable	Never Depressed (n = 41)	Depressed (n = 42)
Age in years	37.79 (15.47)	38.64 (14.24)
Number male:female:non-binary	19:22:0	16:25:1
Number employed/studying full-time	32	30
Level of Education	2:11:19:8:1	5:7:14:13:3
Ethnicity	37:3:1	37:4:1
Psychotropic medication use	0	23*
Beck Depression Inventory-II	4.05 (3.42)	31.55 (11.67)*
Ruminative Response Scale	36.41 (11.74)	62.31 (10.81)*
Beck Hopelessness Scale	1.95 (1.61)	12.21 (5.51)*
Dysfunctional Attitudes Scale	31.17 (6.97)	40.71 (8.65)*
Rosenberg Self Esteem Scale	33.29 (3.60)	20.88 (5.11)*
General Self Efficacy Scale	33.32 (3.74)	25.38 (6.18)*
Digit Span	20.59 (4.48)	17.86 (4.71)
Verbal Fluency Test	38.61 (10.43)	36.02 (9.94)
NART number of errors	12.78 (7.96)	12.52 (7.26)

Note. * p < .05. For education, number of participants that have completed high school: 6th form or college: undergraduate degree: postgraduate degree: other (including apprenticeship or diplomas). For ethnicity, number of participants that self-identified as of White: Asian: Mixed race heritage.

Primary hypothesis

Figure 1(a) displays mean boundary extension scores for each condition. As all means indicated extension (i.e. mean scores were below zero), we present the unsigned mean to improve readability, whereby higher score indicates greater boundary extension. Our primary hypothesis predicted a group by valence interaction on boundary extension scores. A mixed ANOVA predicting boundary extension demonstrated no significant interaction, F(2,81) =  0.12, p = .89, d = 0.08 [−0.36, 0.52]. Our hypothesis was therefore not supported. We did observe a significant main effect of valence, F(2,162) = 8.46, p < .001, such that both positive, F(1, 81) = 5.55, p = .021, d = 0.37 [0.06, 0.69], and negative stimuli, F(1, 81) = 18.17, p < .001, d = 0.66 [0.34, 0.98], were extended more than neutral stimuli. The degree of extension for positive versus negative stimuli was not significantly different, F(1,81) = 2.77, p = .10, d = 0.19 [−0.26, 0.64]. No significant main effect of group was observed, F(1, 81) = 0.16, p = .69, d = 0.09 [−0.35, 0.53].

Figure 1. a. Estimated marginal mean (standard error) boundary extension scores across all stimuli and for each stimuli valence, by group. b. Estimated marginal mean (standard error) boundary extension scores by content type and valence, for each group. Note: Boundary extension scores were measured on an index of 0 to 100, such that higher scores indicate greater boundary extension. Error bars represent standard error.

We also examined, in line with our preregistration, whether the content of images may impact extension (see Bainbridge & Baker, 2020) using a group × valence × content (face or object) mixed ANOVA predicting boundary extension (Figure 1(b)). No significant group × content, F(1,81) = 1.10, p = .30, d = 0.23 [−0.22, 0.68], emotion × content, F(2, 162) =  1.88, p = .16, d = 0.21 [−0.10, 0.52], or three-way interaction, F(2, 162) = 1.20, p = .30, d = 0.17 [−0.14, 0.48], were observed. A large main effect of content was evident, F(1, 81) = 27.74, p < .001, d = 1.16 [.68, 1.64], such that the boundaries were further extended on images which contained objects (M = 14.37, SD = 13.9) relative to faces (M = 7.86, SD = 10.83). In light of this, it is important to note that images containing faces and objects were matched on arousal (see Method).

Relationship with self-schema

As our hypothesis concerned the difference between groups, we explored correlations in each group separately. Stronger endorsement of negative beliefs was associated with boundary extension in depressed participants, with a small-moderate correlation between DAS and boundary extension for negative images, r = -.39, n = 42, p = .01.¹ This correlation was significantly different, z = −2.20, p = .01, from that in the never-depressed group, r = .09, n = 41, p = .58. A similar pattern was observed for neutral stimuli, depressed r = -.36, n = 42, p = .02; never-depressed r = .06, n = 41, p = .70, z = −1.92, p = .03, but not positive stimuli, depressed r = -.24, n = 42, p = .12; never-depressed r = -.22, n = 41, p = .17, z = −0.09, p = .46. No significant correlations were observed between boundary extension and positive schemas or hopelessness in either group (see Supplementary Table).

Secondary hypothesis

We predicted a significant association between boundary extension and overgeneral autobiographical memory. Overgeneral memory is commonly indexed by the proportion of specific memories correctly retrieved on the AMT. We observed a trivial and non-significant negative correlation between boundary extension scores across all images and proportion of specific memories, r = -.06, n = 83, p = .49. To provide a more precise examination of whether extension effects in memory for perceptual space are associated with extensions of memory across time, we also correlated boundary extension scores with the proportion of extended memories. A small, significant correlation was observed between total number of extended memories and boundary extension, r = -.23, n = 83, p = .03. No correlation was observed between boundary extension and the proportion of categoric memories, which summarise repeated categories of event, r = .10, n = 83, p = .35.

Discussion

This pre-registered study provided the first examination of boundary extension in individuals diagnosed with depression. The presence of either positive or negative stimuli was associated with greater boundary extension relative to neutral stimuli across our sample. We obtained adequate statistical power, and found no evidence of a significant difference, with a negligible effect size between depressed and never-depressed participants for boundary extension. This suggests that it is the emotional content of stimuli, rather than the individual’s experience of emotional disturbance, which is likely to influence the manner in which perceptual experiences are encoded and later reported.

There has been an increasing focus on the role of self-schema in perceptual experience of the world (Constant et al., 2022). However, our findings provide limited support for further exploration of the association between self-schema and boundary extension. We did observe a correlation between DAS scores and boundary extension for negative images in the depressed but not never-depressed participants, however boundary extension did not differ between depressed and never-depressed individuals despite a large effect size for the difference in schemas between these groups. Limitations of this study may have impacted these results. We employed self-report measures of self-beliefs due to their use in clinical practice, but other methodologies may offer more precise assessment of the relationship between schema and memory processing, such as computational approaches to objectively quantify the strength of prior beliefs, or experimental paradigms which objectively index the strength of existing self-relevant positive or negative memory bias (e.g. Kuiper & Derry, 1982). Such paradigms would reduce the impact of meta-cognitive awareness of one’s own beliefs, which may have influenced our results. Self-relevance of images should also be considered. Self-referential images would need to be selected individually for each participant, would be difficult to match between participants, and particularly difficult to match between positive and negative conditions for depressed participants. We elected to prioritise experimental control, but this will have impacted results.

Our findings offer implications for future examination of boundary extension. Our finding that both positive and negative stimuli were extended more than neutral is inconsistent with Ménétrier et al.’s (2013) report of extension on positive stimuli only. Ménétrier et al. (2013) made use of short film clips, and our findings align with prior observations that both positive and negative still images are extended (Beighley et al., 2019; Takarangi et al., 2016), suggesting that use of film clips vs still images may impact boundary extension. We also demonstrated, for the first time, that scenes containing faces as the central object are less likely to be extended than images containing a non-human central object, for both depressed and never-depressed participants, regardless of emotional valence. This could potentially be due to the inherent social and evolutionary value of faces, and thus different attentional capture and response mechanisms (Mavratzakis et al., 2016), or recruitment of different neural networks (Keightley et al., 2010), for emotional faces relative to emotional scenes. Future studies should thereby consider both the central focus of, and emotional valence of an image when exploring boundary extension effects.

Identification of domain-general processes which are associated with poor mental health, and influence multiple cognitive systems (e.g. memory for events, perception of space and time) could emphasise novel treatment targets. Our results suggest that the over-extension of memory over time could potentially be related to the extension of perceptual space in memory. Overlap in the neural networks and processes which support memory for autobiographical events and spatial–temporal features have been previously documented (for review see Burgess et al., 2002). This study has however been the first to demonstrate a small but significant correlation between extended autobiographical memories and boundary extension effects. These findings tentatively suggest that it may be useful to further evaluate whether the propensity to extend temporal–spatial features is a domain-general process that is evident across memory for perceptual scenes and autobiographical events. Cross-domain cognitive processing is garnering considerable interest within cognitive psychology and neuroscience (e.g. Lau et al., 2022), and the need to explore intersection between memory, emotion, and time has been emphasised (Petrucci & Palombo, 2021). Further evaluation of the underpinnings of the overgeneralised cognitive style which characterises mental ill health, and is a key focus of psychological therapy (Beck et al., 1979), may improve understanding of the cognitive mechanisms that drive depression. Our findings suggest that the over-emphasis and over-extension of negative experiences (relative to positive) during depression is unlikely to be driven by perceptual experience, and reiterate the need to target memory retrieval and interpretative biases (Hitchcock et al., 2017) in depressive treatments.

Author contributions

CH and TD conceived the study. CH, CE, and SP designed study materials. CE, MS and SP completed data collection. CH and SP analysed data. All authors contributed to writing the manuscript. Data and pre-registration record available at https://osf.io/kqybr.

Acknowledgments

SP and CH were funded by an UKRI Economic and Social Research Council award to CH, (ES/R010781/1). This work was also supported by the Medical Research Council, SUAG/043 G101400. Dr Hitchcock also received salary support from the Australian Research Council (Grant Number DE200100043).

Disclosure statement

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

Additional information

Funding

This work was supported by Economic and Social Research Council: [Grant Number ES/R010781/1]; Medical Research Council (MRC) : [Grant Number SUAG/043 G101400]. Dr Hitchcock also received salary support from the Australian Research Council (Grant Number DE200100043).

Notes

1 To ensure that multicollinearity was not impacting this result, given the moderate-high correlations between self-belief measures (reported in Supplementary Table 1), we ran a simultaneous regression in which both positive and negative self-beliefs predicted boundary extension for negative images. Consistent with the reported correlations, DAS was the only significant predictor, β = -.42, t = -2.15, p = .038.