Negative Emotion (dys)regulation Predicts Distorted Time Perception: Preliminary Experimental Evidence and Implications for Psychopathology

ABSTRACT

Objective

Accurate time perception is crucial to daily life but vulnerable to interference, particularly through negative affect, which dilates individuals' sense of time passing. Regulation strategies like rumination, and disorders like borderline personality disorder (BPD), are linked to time distortion, yet their interrelationships remain untested. We investigated whether rumination and BPD symptoms increase time dilation in negative affective states to understand the clinical implications of time distortion.

Methods

In an online pilot study, we tested whether negative affect (NA) predicts subjective time perception and explored how rumination, BPD symptoms, and their interaction predicted time perception using a between‐subjects online experimental mood induction. Adult participants (N _combined = 760) were recruited from Prolific Academic and a large, Midwestern U.S. university.

Results

State NA and increased BPD features predicted increased time dilation. The role of trait‐level rumination was nuanced, with individuals low in BPD symptoms and elevated trait rumination exhibiting reduced time dilation in response to NA. Conversely, those with elevated rumination and BPD symptoms reported increased time dilation in the neutral condition.

Conclusion

Findings offer foundational evidence of NA and rumination's roles in time dilation for individuals across levels of BPD symptom endorsement. Subsequent replication and extension could flesh out these relationships and inform psychotherapeutic treatment targets.

“I wasted time, and now doth time waste me.” – William Shakespeare, Richard II (1595)

People tend to believe, with conviction, that their perceptual experiences reflect objective reality (Ross and Ward 1996). Yet this widely documented belief is faulty, such that individuals consistently demonstrate that their interpretations of the world are colored by internal, top‐down processes, e.g., attitudes, intentions, emotions, and values (Pronin 2007). The sensation of time passing, or time perception, is one domain where the potential impact of perceptual distortions may seem innocuous but can have broad reaching impacts. Theorists argue that humans fundamentally need time in order to successfully maintain their physical and mental health and to understand the coordination of natural and social relationships (Burns 1977). Societies are built on mutual understanding of a common metric for time in which to organize their entire way of life. However, time perception in humans is highly subject to interference (Angrilli et al. 1997; Mella, Conty, and Pouthas 2011; Tian, Liu, and Huang 2018). Episodically, this could have implications for occupational or social role obligations, regulation and synchrony in interpersonal interactions, and failure to maintain health behaviors. When chronic, these may then contribute to pathology. In the current work, we explore how alterations in time perception as a result of emotion processing may contribute to impairment in psychological functioning.

Researchers can alter participants' sense of time perception by manipulating emotional state (Angrilli et al. 1997; Mella, Conty, and Pouthas 2011), emotion regulation (Tian, Liu, and Huang 2018), or even how participants focus attention (Mella, Conty, and Pouthas 2011). The reverse is also possible; by changing time cues given to participants, we can manipulate their reports of processes we think of as under our own volition, such as pain tolerance (Pomares et al. 2011; Rey et al. 2017) and the ability to hold our breath (Vigran et al. 2019), and those more involuntary, such as blood sugar levels (Park et al. 2016) and physiological arousal (Mella, Conty, and Pouthas 2011). These studies demonstrate that time perception is a fluid, malleable process, despite its appearance of relative stationarity. Timing and time perception are vital to everyday life, from safely crossing a bustling street to mapping associations between cause and effect (Eagleman et al. 2005; Faro, McGill, and Hastie 2013; Meck 2005). Just as individuals express conviction about their experience of time, they defend the veracity of their emotional experiences, both without the express realization that time perception is heavily influenced by affect, and affective cues by time perception (Angrilli et al. 1997; Droit‐Volet and Wearden 2016; Lake, LaBar, and Meck 2016). Thus, while no less real, both are subject to influence. When trying to understand time perception and emotion each, it is critical that we also understand how one affects the other, particularly when they can be leveraged to promote physical and mental functioning.

1. Time Perception

Time perception refers to the way an individual experiences and senses time, rather than the duration of chronological, or “clock,” time itself (Lake, LaBar, and Meck 2016). It is expressed in colloquial phrases that underscore chronological time's involvement with emotion, such as, “time flies when you're having fun,” in reference to a particularly positive experience, or “those were the longest five minutes of my life,” in reference to a particularly boring or painful experience. While a review of specific models of time perception is beyond the scope of this investigation, theoretical models and existing literature in time perception indicate that one's internal (e.g., volitional attention, cognitive load, awareness of time distortion) and external (e.g., emotionally evocative stimuli) environment shapes time perception (Lake, LaBar, and Meck 2016; Buonomano and Karmarkar 2002; Gibbon, Church, and Meck 1984; Lamotte et al. 2014; Treisman 1963). In particular, a growing body of evidence over the past twenty years in cognitive psychology suggests that emotion itself is a key modifier of time perception (e.g., Lake, LaBar, and Meck 2016; Bar‐Haim et al. 2010; Buetti and Lleras 2012; Droit‐Volet and Meck 2007; Gil and Droit‐Volet 2009; Vohs and Schmeichel 2003) and that emotion distorts the relative accuracy with which we perceive clock time.

2. Emotion and Time Distortion

Time distortion refers to the phenomenon in which the amount of time an individual perceives to have passed differs from the actual amount of clock time that has passed. When time feels like it is moving by more slowly than clock time and is thus extended, this is referred to as time dilation; conversely, the sensation that time is moving by more quickly than actual clock time and its subjective estimate is shortened, is referred to as time constriction. Core to this phenomenon is the specific emotional state in which one finds oneself. Broadly, negative affect (NA) has been consistently linked with time dilation (Droit‐Volet 2013) while positive affect has been linked to time constriction (Droit‐Volet and Wearden 2016). A significant body of literature is devoted to the effects of more specific emotional cues on time distortion. These include studies in which experimenters presented stimuli to induce mood, using emotional faces (Angrilli et al. 1997; Bar‐Haim et al. 2010; Gan et al. 2009; Tipples 2008), emotional body postures (Droit‐Volet and Gil 2015), music (Droit‐Volet et al. 2010; Sackett et al. 2010), emotional sounds (Noulhiane et al. 2007), film clips (Droit‐Volet, Fayolle, and Gil 2011), and phobic stimuli (Buetti and Lleras 2012) in order to produce time distortion. Across these studies, negative affectivity has reliably predicted time dilation. ¹ However, as noted above, what individuals focus their attention on and how they regulate their emotions alters the influence of emotion on time distortion (e.g., Tian, Liu, and Huang 2018). Moreover, the dearth of experimental research on time perception and emotion regulation as a function of emotional stimuli is a critical missing piece in understanding their interplay. To understand this association, we must also consider the flexibility of time perception more generally.

3. Emotion Regulation and Time Distortion

Time dilation or constriction themselves may not be harmful per se, since some would argue that time perception is expertly flexible and this adaptability is a strength (e.g., Gil and Droit‐Volet 2009). Rather, it could be the case that it is the relative rigidity in one mode that is ineffective for broader functioning across life's domains. One example in which temporal rigidity can be ineffective for broad functioning is when there exists the chronic presence of negative internal affective states, because we would expect to see persistent time dilation in this case. Indeed, extant literature demonstrates that negative emotion itself dilates the experience of time (Lake, LaBar, and Meck 2016) and focusing on one's emotional state in response to negative events increases healthy individuals' time dilation (Mella, Conty, and Pouthas 2011; Tian, Liu, and Huang 2018). Recent preliminary evidence also suggests that increased belief that negative emotion will persist (i.e., “last forever”) is linked to increased emotion dysregulation, namely rumination (Veilleux et al. 2023).

Perhaps unsurprisingly, chronic dilated time perception has been preliminarily linked to depression (Droit‐Volet 2013), anxiety (Bar‐Haim et al. 2010; Buetti and Lleras 2012), and borderline personality disorder (BPD; Mioni et al. 2020). To varying degrees, these disorders also feature persistent rigidity in the form of rumination, or “moody pondering,” which tends to predict chronic, protracted negative mood states (Gross and Muñoz 1995; Nolen‐Hoeksema, Wisco, and Lyubomirsky 2008; Selby et al. 2009). While typically employed to understand the causes and consequences of negative emotion, rumination paradoxically intensifies negative emotions by maintaining negative information in conscious awareness (Nolen‐Hoeksema, Wisco, and Lyubomirsky 2008; Lyubomirsky and Nolen‐Hoeksema 1995). Ample literature supports that rumination increases NA in clinical (Selby et al. 2009; Baer and Sauer 2011; Tuna and Bozo 2013) and non‐clinical populations (Moberly and Watkins 2008; Selby, Anestis, and Joiner 2008; Schwartz and Koenig 1996), and evidence suggests that increased trait‐like rumination is a risk factor for developing or maintaining mental health disorders like depression, anxiety, and BPD (Selby, Anestis, and Joiner 2008; Aldao, Nolen‐Hoeksema, and Schweizer 2010; Selby and Joiner 2009). Moreover, at least one empirical study found that ruminating predicts increased time dilation (Mella, Conty, and Pouthas 2011).

Rumination plays a central role in the maintenance of psychopathology and chronic negative mood states. Therefore, if increased trait rumination does dilate time perception, then individuals prone to experiencing protracted NA, those who ruminate, and those who experience both, may be at an increased risk of experiencing chronic temporal distortion, or temporal dysregulation. In other words, individuals whose struggles are characterized primarily by rumination and/or NA may be stuck in a cycle where their negative mood states are experienced as longer than they actually are relative to clock time. In an attempt to think their way out of their emotional state and its context, they have prolonged their own time spent suffering and now feel as though time itself has begun to consume them—a parallel to Shakespeare's soliloquy of Richard II, who reflected on wasting time and money not attending to his country's immediate well‐being and eventually ending up in prison brooding while paying for it.

4. Clinical Applications

Though yet untested, if supported, the association described above could be of considerable clinical significance for multiple disorders with core features relating to emotion dysregulation. One prime candidate disorder may be BPD, as a core feature of BPD is difficulties with emotion regulation, particularly regulating NA (Baer and Sauer 2011). As one potential mechanism, time dilation fits well with one conceptualization of NA and rumination that spiral into dysregulation, as described by Selby and colleagues in the Emotional Cascade Model (ECM; 37, 42). In the ECM, when individuals with BPD experience strong negative emotion, they typically report relying on unskillful emotion regulation strategies that maintain negative information in conscious awareness. Principal among these is rumination. In turn, ruminative thinking compounds their NA, which fuels further rumination. This model proposes that behavioral dysregulation, such as non‐suicidal self‐injury, is a result of the individual attempting to short circuit the negative affect‐rumination positive feedback loop (Selby et al. 2009; Selby, Anestis, and Joiner 2008; Selby and Joiner 2009). Therefore, if there is a way to verify that time dilation is associated with this positive feedback loop, then that could inform how clinicians might best address and undercut time dilation and subsequent symptom intensification. Critically, the associations between rumination, time distortion, and NA remains unexplored in the literature, and no research has yet examined these associations in a clinical context.

5. Present Study

While considerable cognitive psychology research has been devoted to affect and time perception in the past 20 years, most studies have investigated objective time judgments (e.g., temporal bisection task) in response to emotional stimuli, rather than exploring the effects of emotion on subjective time perception. The present pilot study addressed this gap in knowledge through a foundational experiment in which we manipulated mood to examine the effects of emotion reactivity on subjective time distortion in an online paradigm. In exploratory analyses, we also investigated whether these associations bear out in the presence of elevated trait rumination and BPD symptoms to test for applicability to these populations and establish preliminary evidence for this pathway. By studying objective ratings' counterpart, subjective time distortion, we aimed to (1) strengthen the literature on emotion‐based time perception by joining cognitive and clinical psychological approaches, (2) assess how conscious individuals are of their experience of temporal distortion retrospectively, and (3) test for the first time whether trait rumination and BPD traits modify the well‐noted association between NA and time dilation.

6. Hypotheses

6.1. Primary

Negative mood, achieved via induction, will predict increased subjective time dilation.

6.2. Exploratory

Elevated trait rumination will amplify the effects of negative mood induction on subjective time dilation, given preliminary literature linking increased rumination and time dilation (Mella, Conty, and Pouthas 2011).

Elevated trait rumination and increased BPD symptoms should interact to predict subjective time dilation, given that preliminary evidence links both rumination and BPD symptomology to increased time dilation (Mella, Conty, and Pouthas 2011; Mioni et al. 2020). Specifically, this relationship should emerge as a three‐way interaction with mood induction, such that negative mood induction × elevated rumination × elevated BPD traits predict the strongest time dilation response. Of note, we label the hypotheses above as exploratory to indicate that there are individual components of the hypotheses that have not been previously explicitly linked in the literature, to our knowledge. As this is a pilot study, it has yet to be assessed if our chosen combination of operationalizations/methods from different fields will be compatible.

7. Transparency and Openness

The research questions presented here are part of a larger study undertaken to understand time perception and affect's bidirectional effects. Deidentified data and annotated syntax for analyses are openly available through OSF at https://osf.io/236sm/?view_only=366b1291bd9343bc9245f875b2e538ad. We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the present study in the following section. All procedures were approved by the host institution's Institutional Review Board. Participants were compensated for participation.

1. Methods

1.1. Participants

This study was reviewed and approved by the host university's Institutional Review Board (IRB‐118‐2021). Participants were recruited from two sources (see Table 1 for combined and by‐source sample descriptive statistics). First, U.S.‐based participants were recruited using the research site Prolific Academic (https://www.prolific.co/), a UK‐based online platform designed for surveys and experiments. Seed funding of $3000 allowed recruitment of 322 participants from Prolific for the study at $7.50 remuneration per person for the 40‐min protocol. Participants from this sample all lived in the U.S., spoke English as their primary language, and were aged between 18 and 60 (N = 322, M _age = 30.40, SD = 8.85, N _men = 155, N _women = 156, N _nonbinary = 9, N _{not listed} = 2, 76.7% White, 13.0% Asian or Asian American, 10.9% Black or African American, 4.0% Latinx, and 1.9% Alaskan Native or Native American). A second sample, whose final number was limited to data collection in Spring 2022, was comprised of 569 undergraduates from the host university's Introduction to Psychology research pool, all spoke English as their primary language, and were currently residing in or were citizens of the U.S. (M _age = 19.41, SD = 1.19, N _men = 338, N _women  = 227, N _nonbinary = 4, 65.5% White, 22.7% Asian or Asian American, 5.4% Black or African American, 3.3% Latinx, and 1.0% Alaskan Native or Native American). We opted to combine our samples a priori because the combined sample affords us a more robust test, with greater power, to test our core hypothesis and exploratory hypotheses. This larger, more heterogeneous sample is also less susceptible to random sampling variability broadly. For interested readers, we additionally provide information in the descriptive tables outlining the ways in which the samples differed (see Table 2). Finally, we felt comfortable combining a partly undergraduate sample with our national sample as there is good evidence to suggest that non‐clinical, undergraduate samples can and do evidence clinically meaningful levels of psychopathology, such as BPD features (Trull 1995; Trull et al. 1997).

TABLE 1.

Sample demographic information of final sample.

Variable	Levels	Combined		Prolific		Sona
		N	%	N	%	N	%
Sex at Birth	Female	323	42.60	150	51.0	173	37.20
	Male	435	57.30	144	49.0	291	62.60
	Intersex	0	0.00	0	0.00	0	0.00
Gender	Woman	313	41.20	141	48.00	172	37.00
	Man	433	57.00	144	49.00	289	62.20
	Non‐Binary	11	1.40	8	2.70	3	0.60
	Not Listed	1	0.10	1	0.30	0	0.00
Sexual Orientation	Straight or Heterosexual	643	84.70	229	77.90	414	89.00
	Gay or Lesbian	23	3.00	14	4.80	9	1.90
	Bisexual	74	9.70	38	12.90	36	7.70
	Not Listed	18	2.40	13	4.40	5	1.10
Relationship Status	Married	77	10.10	76	25.90	1	0.20
	Living w/Partner	44	5.80	33	11.20	11	2.40
	Widowed	2	0.30	1	0.30	1	0.20
	Separated	4	0.50	2	0.70	2	0.40
	Divorced	6	0.80	6	2.00	0	0.00
	Never Married	625	82.30	176	59.90	448	96.60
Native English	Yes	705	92.90	294	100	411	88.40
	No	53	7.00	0	0.00	53	11.40
English Fluent	Yes	734	96.70	292	99.30	442	95.10
	Mostly	23	3.00	2	0.70	21	4.50
	Minimally	1	0.10	0	0.00	1	0.20
Race/Ethnicity	White	555	73.10	225	76.50	330	71.00
	Asian	151	19.90	40	13.60	111	23.90
	Black or African American	55	7.20	30	10.20	25	5.40
	Latinx	26	3.40	12	4.10	14	3.00
	Native American	11	1.40	5	1.70	6	1.30
	Hispanic	42	5.5	15	5.10	27	5.80
	Non‐Hispanic	717	94.50	279	94.90	438	94.20
Education	< High School	4	0.50	3	1.00	1	0.20
	High School or GED	154	20.30	43	14.60	111	23.90
	Some College	420	55.30	77	26.20	343	73.80
	2‐year Degree	27	3.60	27	9.20	0	0.00
	4‐year Degree	100	13.20	93	31.60	7	1.50
	Some Graduate or Professional School	15	2.00	14	4.80	1	0.20
	Masters, Doctorate, or JD	38	5.00	37	12.60	1	0.20

Note: Participants could select multiple applicable race/ethnicity categories allowing the percentages to sum to great than 100%. Combined sample bolded for ease of quick viewing.

TABLE 2.

Independent samples t‐tests of recruitment group differences on key study variables.

	Prolific		Sona		t‐test	df †
	M	SD	M	SD
Baseline NA	16.15	7.10	20.31	8.49	−7.007***	755
Post NA	17.11	8.65	19.14	8.43	−3.181**	611.948
Baseline PA	28.14	9.39	29.50	8.49	−2.019*	577.680
Post PA	27.41	9.76	28.45	9.09	−1.465	591.129
Age	30.29	8.79	19.47	1.16	26.158***	756
PAI‐BOR	27.14	13.20	30.27	11.38	−3.469***	754
Affective Instability	7.01	4.26	7.07	3.56	−0.197	754
Identity Disturbances	8.22	4.33	9.92	4.08	−5.458***	756
Negative Relationships	8.03	4.04	7.97	3.62	0.243	756
Self‐Harm	3.88	3.59	5.31	3.43	−5.437***	602.39
RRS Brood	11.52	3.72	11.32	3.33	0.754	756

Note: Baseline NA = Negative affect subscore from PANAS pre‐mood induction; Post NA = Negative affect subscore from PANAS post‐mood induction; Baseline PA = Positive affect subscore from PANAS pre‐mood induction; Post PA = Positive affect subscore from PANAS post‐mood induction; PAI‐BOR = Score from Borderline features subscale of the PAI; Affective Instability = affective instability subscore of the PAI‐BOR; Identity Disturbances = identity disturbance subscore of the PAI‐BOR; Negative Relationships = negative relationships subscore of the PAI‐BOR; Self‐Harm = self‐harm subscale of the PAI‐BOR; RRS brood = Score from brooding subscale of the RRS.

^† Fractional degrees of freedom in the case that variance was not pooled for t‐test.

^* p ≤ 0.05.

^** p ≤ 0.01.

^*** p ≤ 0.001.

1.2. Procedure

After providing informed consent, participants completed a number of self‐report measures and computer tasks as part of a larger protocol. Included within this manuscript are descriptions of the measures that are relevant to this research question. Self‐report measures were trait‐level measures of rumination and BPD symptoms (Ruminative Response Scale‐RRS, 48; Personality Assessment Inventory‐Borderline Subscale, 49; see Measures for more information) and demographic information. Next, participants rated their current positive and NA using an abbreviated Positive and Negative Affect Schedule (PANAS; 50) to provide an affective baseline. Participants then answered diversion questions about themselves (e.g., questions about their height or their favorite color) to take their focus off of their emotion ratings and conceal the purpose of the experiment.

1.2.1. Mood Induction

Participants were oriented to study procedures with the following instructions displayed onscreen, “Make sure to complete this entire protocol somewhere quiet, where you can focus. You will be performing a number of tasks today, starting with watching a video. Please watch the following video closely. Notice how you feel.” Participants then underwent between‐person negative, positive, or neutral (control) mood manipulations in the form of 5‐min video clips selected for their effectiveness in inducing target mood in an online paradigm (for negative and positive film clips; 51). The neutral clip was selected for its low arousal and neutral valence ratings but was not previously tested online (Hewig et al. 2005). The negative film clip depicted police officers negatively interacting with people skateboarding and the positive clip depicted a compilation of cats in amusing situations (Gilman et al. 2017), while the neutral film clip depicted two older men speaking in a courtroom (from the film Crimes and Misdemeanors; 52). Participants were asked an accuracy question about the clip they just watched to identify whether they paid attention to the clip and they were also asked two questions to indicate if there was an issue with the video displaying properly (i.e., “What animals are depicted in this film?”, “Did the video play for you?”, “Did you encounter any technical difficulties while watching the video?” (Gilman et al. 2017)). Next, participants immediately reported their affect on a second PANAS to measure emotional reactivity to the film clip. Individuals who were not able to view the clip due to technical difficulties (N = 131) were excluded from final analysis (N _Total = 760). Finally, participants answered a question that indicated the speed at which they noticed time was passing by (i.e., subjective time perception; see Measures). This last question served as the outcome variable in the present study.

1.3. Measures

1.3.1. Borderline Personality Disorder Symptoms

The Personality Assessment Inventory‐Borderline Subscale (PAI‐BOR; (Morey 1991)) is a 24‐item scale measuring BPD symptoms divided into four subscales (Affective Instability, Identity Problems, Negative Relationships, and Self‐Harm). Respondents answer questions such as, “My mood shifts quite suddenly” (Affective Instability), and, “When I am upset, I typically do something to hurt myself” (Self‐harm), on a four‐point Likert scale false (0), slightly true (1), mostly true (2), very true (3). Item responses are summed to create the subscales and can also be combined (i.e., summed) to create a composite score on the PAI‐BOR, where higher scores indicate greater presence of BPD symptoms. Internal consistency was strong or better for each sample (α_Prolific = 0.90; α_{Undergraduate} = 0.87). Endorsement of each item, by sample, is provided in supplement (see Table S3), alongside subscale scores for the PAI‐BOR (see Figure 1).

FIGURE 1.

Scores for each of the PAI Borderline subscales by recruitment group. Subscale scores range from 0 to 18; Prolific = participants recruited from Prolific website; Sona = participants recruited from undergraduate pool enrolled in introductory psychology course at host institution.

1.3.2. Ruminative Response

To measure rumination, we employed the Ruminative Response Scale‐RRS (RRS; 48). The RRS is a 22‐item scale consisting of three subscales (Depression‐related, Reflection, and Brooding) assessing the amount that the respondent tries to regulate dysphoric emotions by dwelling on their cause. Respondents answered to what extent they typically engage in certain activities when they feel, “sad, down, or depressed.” Questions include items such as, “Think about how alone you feel” (depression), “Write down what you were thinking and analyze it” (reflection), and, “Think about a recent situation, wishing it had gone better” (brooding), on a four‐point Likert scale ranging from almost never (1), sometimes (2), often (3), to almost always (4). The five‐item Brooding subscale of this measure was used in hypothesis testing because this subscale is reflective of rumination as defined in the present study and it has been linked to intensified emotion dysregulation (Treynor, Gonzalez, and Nolen‐Hoeksema 2003), BPD (Selby et al. 2015), and time dilation (Mella, Conty, and Pouthas 2011). The Brooding subscale was calculated by summing the scores of each item, with lower scores indicating less rumination and higher scores indicating more rumination. Internal consistency was acceptable across samples for the Brooding subscale (α_Prolific = 0.83, α_{Undergradaute} = 0.76).

1.3.3. Positive and Negative Affect

State affect was measured with the Positive and Negative Affect Scale (PANAS‐X; 50). The PANAS‐X negative and positive affect subscales contain 10 items each. Participants indicated to what extent they felt “this way, right now” at that moment in the protocol, using various descriptive emotion words such as “inspired” (positive affect) and “guilty” (NA) on a five‐point Likert scale ranging from very slightly or not at all (1) to extremely (5). Scoring for the positive and NA subscales is calculated by summing the scores for the 10 positive affect items and summing the scores for the 10 NA items, respectively. Higher scores on either subscale indicate greater levels of that affective state. Reliability for positive and NA subscales across Prolific (α_{NA_bl} = 0.92, α_{NA_post} = 0.93) and undergraduate samples (α_{NA_bl} = 0.91, α_{NA_post} = 0.92) was high. The reliability of change for both positive and negative affect scales was RC = 0.78–0.86. Person‐level reliabilities were RKF > 0.99 (Schrout and Lane 2012).

1.3.4. Demographics

Participants reported various demographic information including age, sex, gender identity, race, ethnicity, nation of origin, primary language, and English language proficiency.

1.3.5. Subjective Time Distortion

In order to assess whether they experienced subjective time distortion, participants answered one item, which asked, “did time feel like it was passing by, [1] faster than usual, [2] slower than usual, or [3] about the same as normal?” as they moved through the experiment. This item was displayed as a multiple‐choice question and was based on current best approaches to measuring subjective time passage (e.g., Berry et al. 2015; Thönes and Wittmann 2016).

1.4. Analytic Approach

Prior to examining the effect of manipulated emotion on time perceptions, we first tested if the emotion manipulations were successful. We estimated ANCOVA models on post‐induction negative and positive affect separately, with covariates included for baseline negative and positive affect, respectively, as well as participant gender, age, sample platform, and study duration. Next, bivariate associations among study variables were examined, with two binary codings for each of the three time perception categories: each category compared to another individually and combining across the other two. A chi‐square test examined the omnibus effect of whether time perception proportions were different as a function of emotion manipulation condition. Finally, a series of cumulative multinomial logistic regression models were fit, with time perception categories (ordered faster, normal, slower) predicted by the same covariates as in the ANCOVA and sequentially adding PAI‐BOR (Model 1a) and RRS Brooding (Model 1b) separately and then together (Model 1c), then baseline negative and positive affect (Model 2), and finally emotion manipulation condition (negative emotion as the reference; Model 3; H1). As an exploratory analysis, in light of theory and suggestive evidence that the effects of emotional states may be magnified by general emotion dysregulation (Selby, Anestis, and Joiner 2008; Selby and Joiner 2009; Treynor, Gonzalez, and Nolen‐Hoeksema 2003), we fit an additional cumulative multinomial logistic regression model including the two‐way interactions between the PAI‐BOR and RRS Brooding scales and the emotion manipulation condition (H2), as well as the three‐way interaction between the three variables (H3). Analyses were conducted using the aov(), chisq.test(), and polr() functions in R (R Core Team 2021) and the effects package for visualization (Fox 2003).

We additionally conducted a sensitivity analysis to assess our power to detect effects within the sample. As our primary comparison was with respect to a binary outcome of time judged as slower than normal time versus not, we simplified our sensitivity analysis model into a logistic regression. Through simulation, using a canonical effect size transformation that is appropriate for binomial base rates between ~5%–95% (Chinn 2000), an interval within which our rates fell, our final sample of 760 individuals was powered at ≥ 80% to detect odds ratio (OR) effect sizes as small as OR = 1.23, which approximates to d = 0.114. However, if the effect size were as small as OR = 1.20 (d = 0.10), the power for all commensurate effects in a logistic model would be ~70% (assuming non‐independence on the order of r = 0.30). Acknowledging that this is only one layer to sensitivity, that is, the power of a differential effect given a fixed sample size, we examined the power of a differential sample size, given a specified effect, and the manifold in between. We focus on smaller effects that result in > 80% power. Specifically, sensitivity analyses suggest that an OR = 1.15 (d = 0.077) is an inflection point of 50% power where N = 760. This changes markedly by OR = 1.20 (d = 0.10), when the power would be 70%, but the required sample size to have 80% power at such an effect size is N = 960.

We note that with the final sample of 760 individuals, both our primary hypothesis and emotion manipulation check achieved 100% power based on previously reported effect sizes (Angrilli et al. 1997; Gilman et al. 2017; Hewig et al. 2005). Additionally, our obtained effect sizes suggest that we were adequately powered (≥ 80%) for all effects of interest (i.e., primary and exploratory hypotheses), such that any effect exceeding ~ OR ≥ 1.23 (Cohen's d = 0.114) is sufficiently powered.

2. Results

2.1. Emotion Manipulation Check

The ANCOVA predicting post‐induction NA by emotion condition (negative, neutral, positive), adjusting for covariates, revealed a significant condition effect, F(2,796) = 122.32, p < 0.001, partial‐η ² = 0.24. Specifically, the negative emotion manipulation (M = 2.40, SE = 0.07) was associated with greater NA than both the positive (M = 1.68, SE = 0.07, d = 0.39, p < 0.001) and neutral (M = 1.79, SE = 0.07, d = 0.33, p < 0.001) manipulation conditions. Additionally, the neutral manipulation condition was associated with greater NA than the positive condition (d = 0.06, p = 0.022). Similarly, the ANCOVA predicting post‐induction positive affect by emotion condition revealed a significant condition effect, F(2,796) = 30.06, p < 0.001, partial‐η ² = 0.07. The positive emotion manipulation (M = 2.89, SE = 0.06) was associated with greater positive affect than both the negative (M = 2.69, SE = 0.06, d = 0.12, p < 0.001) and neutral (M = 2.61, SE = 0.06, d = 0.16, p < 0.001) manipulation conditions. The neutral and negative conditions did not differ with respect to positive affect (d = 0.04, p = 0.096).

2.2. Descriptive Associations

Table S1 presents the bivariate correlations among study variables. Of note, BPD and brooding are highly correlated with one another (r = 0.67, p < 0.001) and both are moderately highly correlated with baseline NA (rs > 0.51, ps < 0.001). These associations are to be expected, however, given the presence of distress captured by measures of NA, BPD features, and brooding rumination. Despite these associations, we proceeded with analyses as an empirical test of whether BPD features, brooding, and NA would perform differently in our hierarchical models, given that there is ample theoretical research to suggest that these constructs are related but not identical (e.g., Baer and Sauer 2011; Kirkegaard 2006; Oliva et al. 2023) All three had small to moderate negative correlations with baseline positive affect (rs < −0.21, ps < 0.001).

The negative emotion manipulation predicted greater likelihood of slow time perception ratings (rs > 0.08, ps < 0.030), and overall a lower likelihood of fast ratings (r = −0.07, p = 0.034) and lower likelihood of fast versus slow comparisons specifically (r = 0.12, p = 0.010). The positive emotion condition was consistent with fewer slow time ratings (rs < −0.07, ps < 0.048) and the neutral condition was surprisingly associated with more fast time ratings (r = 0.08, p = 0.016). Next, NA (r > 0.14, ps < 0.001) and BPD features (rs > 0.10, ps < 0.006), but not brooding (rs = 0.06, ps > 0.080), were also associated a greater likelihood of slow time perception ratings.

2.3. Effect of Trait, State, and Induced Affect on Time Perception

Table 3 presents the counts of time perception ratings as a function of emotion condition. A chi‐square test of the difference in proportions was significant, χ ²(4) = 13.58, p = 0.009. In particular, there was a higher proportion of slow time ratings in the negative emotion condition (21%), compared to the other two (17%). In the cumulative multinomial logistic models (Table 4), BPD symptoms positively predicted increased likelihood that individuals would report that time passed by more slowly than usual (OR = 1.17, p = 0.038; Model 1a), though brooding rumination did not (OR = 1.14, p = 0.085; Model 1b). When included together (Model 1c), neither was associated with time perception ratings. This persisted in subsequent models. In Model 2, baseline NA was the sole positive predictor of slower time perceptions (OR = 1.39, p = 0.004). When the emotion manipulation factor is included (Model 3), baseline NA remained predictive of slower ratings (OR = 1.41, p = 0.003), and the emotion manipulation factor itself was statistically significant (χ ²(2) = 8.68, p = 0.013) such that the negative emotion condition predicts a higher likelihood of endorsing slow time ratings compared to both the positive (OR = 0.64, p = 0.013) and neutral (OR = 0.62, p = 0.009) conditions.

TABLE 3.

Frequencies of time perception ratings as a function of emotion condition.

		Time perception			Total
		Fast	Normal	Slow
Emotion Condition	Negative	14 (2%)	83 (11%)	158 (21%)	255 (34%)
	Neutral	30 (4%)	87 (11%)	131 (17%)	248 (32%)
	Positive	21 (3%)	108 (14%)	128 (17%)	257 (34%)
	Total	65 (9%)	278 (36%)	417 (55%)	760 (100%)

TABLE 4.

Cumulative multinomial logistic regression results for stepwise models.

Effect	Model 1a		Model 1b		Model 1c		Model 2		Model 3
	Est.	OR	Est.	OR	Est.	OR	Est.	OR	Est.	OR
Fast|Normal Intercept	−2.39***	0.09	−2.37***	0.09	−2.38***	0.09	−2.45***	0.09	−2.74***	0.06
Normal|Slow Intercept	−0.15	0.86	−0.13	0.87	−0.14	0.87	−0.20	0.82	−0.46*	0.63
Sex at birth	0.14	1.16	0.11	1.12	0.15	1.16	0.13	1.14	0.13	1.14
Age	−0.03*	0.97	−0.03*	0.97	−0.03*	0.97	−0.03*	0.97	−0.03 †	0.97
Duration	0.00	1.00	0.00	1.00	0.00	1.00	0.00	1.00	0.00	1.00
Platform	0.20	1.23	0.23	1.26	0.22	1.24	0.12	1.13	0.16	1.17
PAI‐BOR	0.15*	1.17			0.13	1.13	0.02	1.02	0.02	1.02
RRS Brood			0.13 †	1.14	0.04	1.04	−0.02	0.98	−0.04	0.96
Negative Affect							0.33**	1.39	0.34**	1.41
Positive Affect							−0.04	0.96	−0.05	0.96
Neutral Induction a									−0.48**	0.62
Positive Induction a									−0.44*	0.64

Note: Fast|Normal Intercept is the average likelihood threshold at which individuals are more likely to report normal than fast ratings. Normal|Slow Intercept is the average likelihood threshold at which individuals are more likely to report slow than normal time rating. Sex at birth is coded: female = −0.5, male = 0.5. Duration = time to complete protocol (in minutes). Platform is coded: Prolific = −0.5, Sona = 0.5. PAI‐BOR = total PAI‐BOR score. RRS Brood = total RRS Brooding subscale score. Negative Affect = total PANAS negative affect subscale score. Positive Affect = total PANAS positive affect subscale score. Neutral Induction = neutral mood condition. Positive Induction = positive mood condition.

Abbreviation: OR, Odds ratio.

^a Negative mood induction is the reference category.

^† p < 0.10.

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

2.4. Exploratory Analysis

In the cumulative multinomial logistic model that included the two‐way interactions between BPD and brooding rumination with emotion condition and the three‐way interaction across all three variables, NA (OR = 1.45, p = 0.002) and emotion condition (χ ²(2) = 8.70, p = 0.013; OR_Pos = 0.64, p = 0.040; OR_Neu = 0.46, p < 0.001) continued to exhibit independent effects, such that NA and the negative emotion condition were associated with a higher likelihood of slow time ratings. Additionally, both a three‐way interaction (H3; χ ²(2) = 8.90, p = 0.012) and a two‐way interaction between brooding and emotion condition (H2; χ ²(2) = 6.55, p = 0.038) were observed. Exploring the 3‐way interaction, the results were limited to the negative compared to neutral conditions (OR = 1.42, p = 0.036), while the three‐way interaction comparing negative to positive emotion conditions was not statistically significant (OR = 0.87, p = 0.424). In contrast, the two‐way interaction between brooding and emotion condition was limited to the negative compared to positive conditions (OR = 1.93, p = 0.008). Full results of the model can be found in Table S2. For simplicity, we provide Figure 2 to illustrate these effects.

FIGURE 2.

Predicted probabilities for time perception judgments as a function of mood condition, trait borderline symptomology, and trait brooding. PAI = PAI‐BOR total score (centered, graphed in standard deviations); RRS = RRS Brooding subscale score (centered, graphed in standard deviations); Time = Subjective time judgment (i.e., faster, slower, or normal); Mood = mood induction condition (negative, positive, or neutral). Panel numbers are displayed in upper center of each panel.

To interpret the interactions, we will refer to the individual cell numbers in the grid of predicted trajectories in Figure 2. The two‐way interaction between brooding and emotion condition is depicted in panels 4–6. Slow time ratings are highest in the negative emotion condition at low brooding levels but increasing brooding scores are associated with less slow and more normal perceptions of time speed (Figure 2, panel 4). There is an opposite effect in the positive condition, such that those with high brooding are more likely to report that time felt slower and less likely to say that time past normally (Figure 2, panel 6). There was no effect of brooding in the neutral condition where slow and normal ratings were equally and more likely than fast ratings.

The three‐way interaction is visualized by comparing the NEG and NEU columns of Figure 2. Whereas the negative emotion manipulation column (panels 1, 4, 7) shows little effect of BPD and the consistent effect of higher brooding on reducing slow time perceptions, the neutral induction column (panels 2, 5, 8) indicates that the impact of brooding on the slow/normal slopes changes as a function of BPD symptom levels. Namely, at high levels of both BPD and brooding, the slow time perception slope reverses direction and becomes positive and significant, such that individuals high on both BPD symptoms and brooding are more likely to perceive time passing more slowly under no emotion manipulation, or analogously in neutral situations. ²

3. Discussion

A wealth of empirical evidence suggests that emotion alters our sense of time perception, creating time distortion. In the present pilot study, we used an online paradigm to manipulate mood and examine the effects of emotion reactivity on subjective time distortion. In exploratory analyses, we investigated whether these associations would change in the presence of increased trait rumination and elevated BPD symptoms to test for applicability to clinical populations. We hypothesized primarily that negative mood would predict increased subjective time dilation (i.e., slowing; H1) and that this effect would be amplified by elevated trait rumination, as rumination itself is linked to increased NA (H2). Further, we endeavored to explore how one disorder in which rumination and NA are prominent, BPD, relates to time distortion (H3). Although results are preliminary, this study helps lay the foundation for future examination of distorted time perception's associations with emotion dysregulation and psychopathology.

First, we established that our emotion manipulation was successful for all three conditions, which is expected given the use of standardized stimuli, most of which were standardized using online mood induction protocols (Gilman et al. 2017). The neutral condition was associated with increased NA compared to the positive condition, which could reflect minor boredom or the moderate tendency for positive and NA to be negatively correlated (Watson and Clark 1999). Others have exhibited similar reactions to mood inductions in an experimental setting (Bieleke, Barton, and Wolff 2021). Importantly, the condition of greatest interest, the negative mood condition, was significantly associated with greater NA and much larger effect sizes when comparing to the neutral and positive mood conditions.

In the cumulative multinomial models, BPD symptoms, baseline NA, and negative mood induction, but not rumination, predicted increased subjective time dilation. The results pertaining to rumination are contrary to expectation as we predicted increased trait rumination would increase time dilation. It is important to note however, that as we measured trait, and not state, rumination, these results cannot be generalized to people who endorse increased in vivo rumination, but rather those who endorse an increased tendency to ruminate. Although people high in trait rumination may ruminate more often than people low in trait rumination, the trait measure does not capture whether participants were ruminating in vivo during the protocol. It is still possible that state rumination may increase time dilation.

Furthermore, the lack of a statistically significant effect, though in the correct direction, may be best understood in the context of the final model, which supported that trait rumination and the negative mood condition interacted to predict subjective time distortion. However, the effect of trait rumination on time dilation reflected an opposite pattern in Figure 2, panels 1, 4, and 7 (i.e., no effect of PAI‐BOR) specifically, from what we tentatively predicted in H2. Trait rumination did not magnify the effect of a negative mood induction on time dilation but rather reduced it. Further, for individuals higher in rumination, especially those at average or lower BPD trait levels, attempts to positively impact their mood resulted in greater dilation and perceptions that time was moving more slowly. This same pattern was observed with respect to the three‐way interaction between mood condition, rumination, and BPD symptoms (H3) but extended further into neutral situations. Namely, in ostensibly neutral, control situations, only when BPD symptoms were elevated did increased trait rumination predict a higher likelihood of dilation and reports that time was taking longer than normal to pass.

The association between BPD symptoms score and time dilation is consistent with preliminary literature (Mioni et al. 2020), and findings for baseline and induced negative mood support H1, as well as considerable prior research (e.g., see Lake, LaBar, and Meck 2016).

The evidence connecting BPD symptoms and time dilation is preliminary, and indirect evidence suggests that this connection may be a result of emotion regulation difficulties within BPD that lead to high levels of NA (Lake, LaBar, and Meck 2016; Mioni et al. 2020). One additional lens through which to view BPD and distorted time perception is through the prevalence of dissociation symptoms within BPD, as dissociation might serve as an upper bound on time perception tracking itself. This association may be important to consider in discussions of time perception within BPD as dissociation entails a disturbance in general consciousness, identity, perception, and behavior, and some measures of disassociation severity include items that refer to a sense of timelessness (American Psychological Association 2013; Dalenberg et al. 2012). However, as impulsivity is linked with time perception (Moreira et al. 2016) and features strongly in BPD, it could also be that impulsivity partly tied BPD symptoms to time dilation before the more proximal consideration of state affect during the protocol. Unfortunately, as the measure of BPD symptomology used in the current study (PAI‐BOR; (Morey 1991)) does not capture dissociation symptoms explicitly, we could not factor dissociation into our models, nor could we explicitly model the separate effect of impulsivity. Future studies examining subjective time perception within BPD should consider the components which drive behaviors in BPD to more fully understand whether BPD is uniquely tied to time distortion and which mechanisms drive such an association.

In light of the general negative connotation associated with time passing more slowly, these results suggest that individuals higher in rumination, BPD symptoms, or both, find neutral/positive situations counter‐normative to expectations and consequently, these situations may provoke ambivalence and subsequent worsened mood (e.g., Baer et al. 2012; Coifman et al. 2012; Domes, Schulze, and Herpertz 2009; Dyck et al. 2008).

Though unexpected, this pattern of findings is critical to the future of time perception research in psychopathology. One way to interpret the results is that it is possible that individuals who endorse elevated rumination live their day‐to‐day lives in a more negative, perhaps chronically dilated state. According to this rationale, it would follow that when these individuals encounter negative stimuli, the result is split for whether they encode that time as passing slowly or not, perhaps because time already feels slower for them generally. We have some evidence for this explanation, given that the main effect of our rumination measure exhibited a non‐significant pattern, such that it may predict greater likelihood of slowed time perception, though it was not significant (OR = 1.14, p = 0.085). It is equally possible that rumination, as measured, does not function as predicted (i.e., trait‐level tendencies further amplify NA's effect on time distortion) and future work is needed to understand if trait rumination exerts effects on time distortion, as well as which direction in which rumination influences time distortion. Unfortunately, there exists very minimal literature with which to compare and contextualize this result.

While tentative, we interpret the finding that those high on rumination report much higher probability of time dilation after positive mood induction as an attention‐meets‐valence issue, such that the more attention is focused on internal, mood‐congruent stimuli, the slower time feels. High rumination individuals experience increased focus on negative, largely internalized stimuli by definition, so when they encounter mood‐congruent (i.e., negative) stimuli, time perception may feel about the same or slightly faster, i.e., more of what they're used to. This is supported by literature that suggests that individuals who experience protracted negative mood report greater time dilation as their normative state (Droit‐Volet 2013; Mioni et al. 2020). Conversely, when we tried to draw participants' attention outward to non‐mood‐congruent stimuli (i.e., positive stimuli), only participants high on rumination reported that time dilated, perhaps because for these participants, the presence of positive stimuli failed to effectively compete with negative, inward focus. Indeed, we see a similar effect in people with impaired attentional processes in depression, such that there is greater emotion capture for negative (mood‐congruent) stimuli and less emotion capture for positive stimuli (Duque and Vázquez 2015; Peckham, McHugh, and Otto 2010). Essentially, what we may be seeing is a habituation effect to negative stimuli, while less familiar or less relevant stimuli, such as a positive mood induction, backfire (i.e., slowed time perception). Importantly, however, as this finding was unexpected, our conclusions and interpretation are speculative. Our future research, as well as that of others, will benefit greatly by replicating and expanding on this empirical question.

3.1. Strengths and Limitations

This study featured several strengths. First, our study design took place entirely online and tested a novel research question using a well‐validated online emotion manipulation. Designs such as these increase access to research participation (Gosling et al. 2004; Meyerson and Tryon 2003; Palan and Schitter 2018) and capture greater heterogeneity of participants' lived experience compared to convenience‐only samples. Next, this study featured a large sample, which enabled us to test interaction effects with confidence. We also were able to capture subjective time distortion in this experiment, without relying on long‐term or “trait” recall, and thus enrich the current literature, which focuses primarily on objective measures of time distortion. Finally, this study contributes to a growing body of literature that suggests that for relatively average individuals (i.e., low on rumination and BPD traits), negative emotion does dilate time perception, whereas positive mood constricts time perception.

There were also key limitations. First, our sample was well‐educated, and the majority identified as cisgender and white. This limits the generalizability of our findings to primarily cisgender, white individuals and leaves an open question regarding the role of time perception among and within more diverse racial and ethnic groups (both inside and outside of the U.S.). Next, we aimed to examine clinical phenomena in a relatively healthy sample, though approximately 25% of the sample reported clinically meaningful levels of BPD features. Given, though, that more extreme differences tend to emerge on the extreme ends of clinical spectra, a more straightforward test could include analog or clinical samples in order to maximize variability among participants from non‐clinical, representative, and clinical strata. Further, we did not include a rumination induction in this study. Instead, we relied on trait rumination to inform how time dilation would function at the state level. Future work should include both state‐ and trait‐level measures of both rumination and time perception if we wish to more fully understand these associations. Next, we did not disambiguate arousal from valence when testing our hypotheses. Given the role of arousal in driving attention‐capture, future work would do well to further explicate these roles within negative or positive emotion. Finally, our measurement of subjective time perception was limited to one item. However, other research that relies on in vivo measures of subjective time perception have used similar methods and highly similar items to assess this novel outcome (e.g., Berry et al. 2015; Thönes and Wittmann 2016) and this represents the current gold standard in the field (Thönes and Stocker 2019), though it is evolving. Moreover, other work suggests that duration judgments are not correlated with time passage judgments, as we aimed to capture (Droit‐Volet and Wearden 2016; Wearden 2015) and so a duration estimate would not have strengthened our outcome. Finally, the present study's item has high face validity and little room for interpretation relative to other, more ambiguous measures of state subjective time perception. Therefore, we maintain that our one‐item measure was an appropriate choice given the novelty of the research question and what contemporary research suggests for use.

4. Conclusion

With the present pilot study, we sought to help merge research from cognitive and clinical psychological literature by examining how affect influences subjective time distortion, but with the novel addition of clinically‐relevant traits in which time distortion and emotion are of particular relevance. Although only our first hypothesis was supported unilaterally, with mixed evidence for our exploratory hypotheses, this study provides rich insight into the previously unexplored relationship between NA, rumination, BPD symptomology, and time dilation. Our data suggest that those with elevated BPD features experience dilated time compared to the general population, however our data only partly suggest that this is a result of rumination. This study supports the current literature on time perception and affect, highlighting that, before clinical trait consideration, negative mood does promote time dilation. Results for clinically‐relevant traits provide clear pathways to extend this research into the clinical domain and examine clinical implications. For example, our findings regarding rumination should be further probed to verify whether individuals prone to rumination endorse greater trait time dilation compared to individuals who do not tend to ruminate, as understanding this question will help determine whether time distortion influences the positive feedback loop between negative affect and rumination, as articulated in the ECM (Selby et al. 2009). If subjective time perception modifies the experience of affect, as affect modulates time perception, then we may be able to short circuit these feedback loops by intervening on time distortion. One method of intervention that has preliminary evidence is the awareness of time distortion itself. In a 2015 (Droit‐Volet, Lamotte, and Izaute 2015) study conducted by Droit‐Volet and colleagues, the researchers found that declarative knowledge of the effects of emotion on time perception attenuated the typical dilation effect of emotional stimuli on a temporal bisection task (Droit‐Volet, Lamotte, and Izaute 2015). Similar to being informed that an image could be either a duck or a rabbit, simple psychoeducation on time distortion could attenuate time distortion in the moment and realign subjective perception to clock time. It is our hope that ultimately, this research assists in uncovering common causes to maintenance factors of psychopathology and in turn, helps inform treatment for individuals prone to rumination and emotion dysregulation.

5. Author Contributions

S.C.N. developed the concepts and research questions and designed the methods for the study. S.P.L. assisted with protocol design. Formal analyses were conducted by S.C.N., S.P.L., and I.K.P. S.C.N. wrote the main manuscript text, with assistance from S.P.L. and I.K.P. All authors contributed to editing drafts of the manuscript. S.P.L. is responsible for results visualization and funding acquisition. S.C.N. was responsible for all aspects of project administration (i.e., building protocol and associated stimuli, participant recruitment, data management). All authors reviewed the final manuscript and approved of its submission for publication in its present form.

Ethics Statement

This study was reviewed and approved by the host university's Institutional Review Board (IRB‐118‐2021).

Consent

All participants gave consent to participate in the research study.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Tables S1‐S3.

Data Availability Statement

The dataset supporting the conclusions of this article is available in the OSF repository, [https://osf.io/236sm/?view_only=366b1291bd9343bc9245f875b2e538ad].