Emotional witness effect: implications for memory accuracy, perceptions, misinformation, and credibility in forensic contexts Open Access

Eyewitness testimony refers to the account given by individuals who have witnessed an event, which is typically a crime or accident. This testimony can include details about what happened, descriptions of the scene and identification of any parties involved (Yarmey, 2010) and often serves as a key piece of evidence in both criminal and civil cases. This type of testimony can help establish timelines, identify suspects and provide context to events, making it a vital component of the narrative presented in court (Wells et al., 2020; Yarmey, 2010). However, the reliability of eyewitness accounts can vary, as factors such as stress, memory distortion and the emotional state of the witness can affect their recollection (Deffenbacher et al., 2004; Głomb, 2022). There are also variations in how credible a witness can appear.

The College of Policing (2020) recognises that a witness or victim of crime may be emotional when recalling an event and provides guidance for navigating interviews. However, there is limited research examining how the presence (or absence) of emotion might affect the credibility of that witness or how much information is remembered from the witness’ testimony.

Understanding the impact of an emotional witness is important for several reasons. For investigators, their emotions can impact how they process inconsistent information (Ask and Granhag, 2007). This is important because emotional (or distressed) witnesses may invoke sadness for the listener. Ask and Granhag (2007) induced negative emotions in police officers and then asked them to make judgements about a criminal case. They found that sad participants were better able to substantively process the consistency of evidence or lack thereof, whereas angry participants used heuristic processing. As such, recognising and managing the emotional states of witnesses can be crucial for ensuring accurate and fair evaluations of their testimonies.

The legal system places substantial weight on testimonies, considering them a primary source of evidence in most criminal cases (Davis et al., 2014; Kebbell and Milne, 1998). Witness testimony is frequently regarded as a reliable form of evidence (Wells et al., 2006). The accounts provided by witnesses can substantially influence jurors’ perceptions and decisions, especially in the absence of physical evidence (for an overview, see Leippe and Eisenstadt, 2009).

This is problematic as eyewitness testimonies can contribute to wrongful convictions, as they often involve misidentification or inaccuracies (see Loftus and Rakoff, 2018). There have been several high-profile cases where victims (i.e. Ronald Cotton case) and eyewitnesses (i.e. Kirk Bloodsworth case) have mistakenly identified the wrong person, resulting in incarceration (Brooks and Moshayedi, 2020; Loftus and Rakoff, 2018). This is further exacerbated by the incorrect and persistent belief in the confidence-accuracy relationship, where fact finders, jurors, and judges believe that a confident eyewitness is a reliable eyewitness (Leippe and Eisenstadt, 2009). It is therefore important that we understand the cognitive processes underlying the perception of eyewitness accounts and the factors that influence decision-making (Semmler and Brewer, 2002a; 2002b). Specifically, human evaluations (or judgements) are notoriously influenced by cognitive biases (Gilovich, 2002). Even when we try to make objective judgements, we cannot help but be affected by our own expectations, beliefs, emotions, or motivations.

According to expectancy violations theory (EVT) when we interact with others, we judge them partly based on whether they violate our expectations regarding normative behaviour (Burgoon, 1978). For instance, if individuals discuss a sensitive topic in a cold and callous manner and that manner of presentation violates our expectations as to how others would discuss this topic, we may judge these people negatively (Bederian‐Gardner and Goldfarb, 2014). Researchers have applied EVT to the relationship between jurors and testifying witnesses, including victims who testify about sexual assaults (Ask and Landström, 2010; Hackett et al., 2008). In both of the aforementioned studies, adults expected victims to appear emotional when testifying. As EVT would predict, victims who were more emotional when testifying, and thus met the participants’ expectations, were rated as more credible than non-emotional victims, who violated their expectations.

For the purpose of this paper, we have separated the research examining the testimony of the victims or witnesses, although we acknowledge that this is sometimes used interchangeably (Dahl et al., 2007).

A victim who exhibits signs of emotional agitation and discomfort is often believed more readily and considered a “real” victim compared to one who remains neutral, controlled or numb (Ask and Landström, 2010; Kaufmann et al., 2003). Previous studies have shown that emotional arousal during testimony can significantly impact credibility and juror perceptions (Ask and Landström, 2010; Dahl et al., 2007; Knight and Ponzio, 2013). This is partly because an emotional testimony enhances the perceived credibility of the individual (Bollingmo et al., 2008; Kaufmann et al., 2003).

This “emotional victim effect” suggests that emotionally expressive witnesses are deemed more believable (Ask and Landström, 2010), although there are boundary conditions which may affect this (van Doorn and Koster, 2019). The credibility of eyewitnesses can be influenced by factors such as gender (c.f. Wessel et al., 2012), where stereotypical expectations about emotional expression may affect juror judgements. To test this, research on emotional victims typically involves using a professional actress who provides testimony about being sexually assaulted (Kaufmann et al., 2003). Participants in these studies usually view one of three video-recorded versions of a victim’s statement, given in a free-recall manner, with the actress displaying one of three types of emotional expression: congruent, neutral, or incongruent.

A congruent emotional expression occurs when the witness’ emotional expression matches the content of their testimony. For example, a victim describes a traumatic event while displaying visible distress or sadness (Ask and Landström, 2010; Kaufmann et al., 2003). This congruence can enhance the perceived credibility of the testimony, as it aligns with decision makers’ expectations of how a victim should behave emotionally in such situations (Houston et al., 2013).

An incongruent emotional expression occurs when the witness’ emotional expression does not match the content of their testimony. For example, a victim may describe a traumatic event while appearing positive or even cheerful (Kaufmann et al., 2003). This incongruence can substantially undermine the perceived credibility of the testimony, as it contradicts the decision makers’ expectations and may lead them to question the authenticity of the witness’ account (Houston et al., 2013).

Finally, a neutral emotional expression typically involves the witness providing their testimony in a calm and emotionally neutral manner, regardless of the content (Ask and Landström, 2010; Kaufmann et al., 2003). This can sometimes be perceived as less credible, as it may not align with decision makers’ expectations of how a victim should emotionally respond to recounting a traumatic event.

In the criminal justice system, differences in emotional expression can strongly influence how jurors perceive and assess the credibility of eyewitness testimony. This influence may also extend to judges and other professionals in the system. However, the degree to which emotions affect credibility judgements can vary between laypeople and experts. For example, Wessel et al. (2006) found that court judges, unlike laypeople, were not influenced by a victim’s emotional display when evaluating credibility or deciding on a guilty verdict. In contrast, Bollingmo et al. (2008) reported that police officers rated a victim as most credible when they showed distress (e.g. crying and despair) and less credible when neutral or showing positive emotions. This discrepancy may stem from problematic beliefs some police officers hold about sexual assault victims (Sleath and Bull, 2017).

Although a witness’ emotional demeanour does not relate to their honesty or accuracy, some research indicates that sexual assault victims who appear distressed are perceived as more credible than those who appear composed (Ask and Landström, 2010; Bollingmo et al., 2008; Wessel et al., 2006). However, a recent meta-analysis of 20 studies (Nitschke et al., 2019) found that this perception remains consistent regardless of the perceiver’s background (whether they are criminal justice professionals or prospective jurors) or the format of the testimony (whether the perceivers read about or watched the complainant recount the alleged sexual assault).

To the authors’ knowledge, there is no published research examining the effects of an emotional witness. In fact, research in this field focuses on victims (usually actors playing the role of a sexual assault victim) providing their recollection (Deffenbacher et al., 2004; Nitschke et al., 2019; van Doorn and Koster, 2019). This is problematic for several reasons. Witnesses are often used by police officers in the initial stages of an investigation, and nervous witnesses are often perceived as being less credible (Bothwell and Jalil, 1992) because of the belief that nervousness is an indication of deception (Vrij, 2000, 2004). As witness testimony is critical for investigators, especially during the early stages of an investigation, we focus our research on those who have witnessed an event.

The current study uses an experimental approach to examine the interaction between the witness’ emotionality and the participants’ expectations in a lab-based setting. The emotionality of the (actor) witnesses, their gender and the emotionality-gender interaction are expected to influence participants’ judgements of credibility, memory and perceptions of the witnesses.

This study measured participants’ (i.e. decision makers’) perceptions of either an emotionally distressed or unemotional witness. Specifically, we were interested in how accurately participants recalled the statement they heard and their susceptibility to misinformation. We also explored the gender congruency effect to understand if participants’ expectations differ for male and female witnesses. To do this, we assessed participants’ perceptions of credibility, emotionality and empathy, as well as participants’ confidence levels, susceptibility to misinformation and memory accuracy.

We predicted that differences in perceptions and judgements would occur based on the witness’ emotionality, gender and the interaction of these factors. Given the limited research on this gender–emotionality interchange for witnesses, our approach was exploratory, focusing on effect estimation and quantifying differences and their uncertainty.

A total of N  = 168 participants (126 female, 40 male and 2 non-binary) took part in the study. The participants’ ages ranged from 18 to 85 (M = 27.98, SD  = 13.19). All participants provided informed consent and were debriefed at the end of the study. Participants were recruited via the university participant pool or through adverts placed on social media. Participants were required to be over 18 years of age.

A 2 (emotionality: emotional vs unemotional) × 2 (gender: male vs female) between-subjects design was used. The independent variables were the emotionality of the witness and their gender. The dependent variables were pre- and post-task confidence, levels of empathy towards the witness, the perceived credibility of the witness and their testimony, the memory accuracy in recalling the details of the incident, and the susceptibility to misinformation about the event.

The study was created using the Gorilla platform (www.gorilla.sc). All participants were randomly allocated to watch one video of a witness, with a final split of emotional female (n = 41), emotional male (n = 44), unemotional female (n = 21) and unemotional male (n = 62). Participants first responded to several demographic questions, followed by the pre-task confidence questions. They were informed that for the purposes of this study, they would be playing the role of a decision maker, where they would be responsible for making judgements about a witness statement.

After watching the condition-specific video, each participant rated the emotionality of the witness, followed by the adjusted state empathy scale. Next, they completed the memory recall task. This was followed by the four suspect detail questions and the witness credibility scale. Afterwards, participants completed the misinformation task. Finally, they answered the post-task confidence questions. Participants were debriefed at the end of the study. Ethical approval was obtained for data collection and stimuli creation from the university’s local ethics committee.

Video-recorded witness or victim testimony using actors is not uncommon (Bollingmo et al., 2008; Kaufmann et al., 2003). For this study, we created 4 video-recorded eyewitness testimonies. To do this, a male and a female actor were used. A script was designed for the actors to memorise and repeat. It consisted of a fictitious encounter of three adult males assaulting another two adult males late at night. The testimony was between 3 and 4 min long. The actors used the same script and varied their emotionality.

The videos described a sequence of events that begins with the witness recalling hearing a scream on the way home from work. This led to her (or him) investigating what was happening. At this point, the witness observes two males being attacked by three other males. From each actor, two testimony videos were created: one emotional and one unemotional. The former was delivered in an emotional (sad/distressed) tone and expression, while the latter was in a neutral manner.

An adapted version of the Shen (2010) state empathy scale was used, changed so that “the person’s emotions” was modified to say “the eyewitness’ emotions”. The scale consists of 12 questions, using a five-point Likert-type response scale (0 = not at all, 4 = completely). The questionnaire measures three sub-types of empathy: affective, cognitive and associative.

Memory recall accuracy was tested by using 15 multiple-choice questions based on the contents of the witness testimony. Each question asks about a particular detail from the video, for example, “How many attackers did the witness see?” The number of (correct) options per question varies between (1 and 3) 3 and 6.

Four questions were constructed to assess the alleged suspect’s appearance, focusing on their skin complexion, haircut, facial hair and facial piercings. Each question used a three-option forced-choice format. For example, “How did the eyewitness describe the suspect’s skin complexion?” with the options: a. Pale, b. Dark, c. Tanned.

To ensure that the emotionality manipulation was effective, two questions about the eyewitnesses’ emotional state while they testified were added. One regarded how the eyewitness appeared (“Which of the following statements best described how you believe the eyewitness appeared?”). The other concerned the eyewitness’ feelings (“Which of the following statements best described how you believe the eyewitness was feeling?”). Both questions used a seven-point Likert-type scale from 1 (very happy) to 7 (very sad). This operationalisation assesses perceived emotionality along a valence dimension (Barrett and Russell, 1999). Specifically, how emotionally positive or negative the eyewitness appeared. This approach captures the overall emotional tone perceived by decision makers, accommodating the variability in individual interpretations and emotional diversity.

The two questions capture theoretically different aspects of perception. Emotional displays can be genuine or posed, and decision makers must consider both what emotion is displayed and its authenticity (Zloteanu et al., 2018, 2020). Thus, perceiving someone as appearing emotional is not the same as believing they are genuinely emotional. Although related, these judgements can diverge in meaningful ways. For example, if an emotional witness is rated as “appearing sad” but not as “genuinely sad,” this discrepancy may influence subsequent credibility ratings.

To determine the perceived credibility of the testimony, two questions were used: “Please rate how credible you felt the testimony was?” and “Please rate how credible you felt the eyewitness was?”, with a six-point Likert-type scale from 0 (completely uncredible) to 5 (completely credible). Credibility was assessed separately for the testimony and the witness, as these judgements can meaningfully diverge. A witness may be viewed as credible, while their testimony is perceived as not credible (e.g. unclear, overly emotional or confused), and vice versa. Treating these as distinct measures allows for a more nuanced assessment of perceived credibility.

A six-question confidence scale was used to assess participants’ confidence in their abilities (five-point Likert-type scale). The questions related to their perceived ability to recall an event, recall details relating to a person, identify the suspect, recognise the emotional state of others and make judgements of information credibility. This task was mirrored at the end of the experiment, with the questions framed from a post-event perspective.

A multi-item task was used to reduce random responding and allow for more granular insights into judgement difficulty. This is important given the exploratory nature of the study and the potential for condition-specific effects. A pre-post design was adopted to account for baseline individual differences and better isolate condition effects.

To assess the susceptibility to misinformation, a 10-item questionnaire was developed where 5 items contained misleading information (e.g. “It was mentioned that the incident took place at 11 a.m.”). All questions used a 3-option forced-choice response (Yes, No, Did not mention).

All analyses were conducted within a Bayesian framework. Inferences are based on the mean of the posterior distribution, the 95% credible interval (CI; computed as the Shortest Probability Interval; Liu, Gelman, and Zheng, 2015), the probability of direction (pd) and the region of practical equivalence (ROPE). Together, these permit inferences about the direction (pd), size (mean), uncertainty (CI) and relevance (ROPE) of an effect. Decisions about negligible or no differences are made based on 100% of the posterior falling within the ROPE range (also known as ROPE(100%) rule)^[1] (Kruschke, 2018; Makowski et al., 2019a, 2019b).

Conditional distributional ordered probit models (Bürkner and Vuorre, 2019; Kurz, 2021) were used to estimate differences in ratings for ordinal dependent variables. These models ensure accurate and unbiased estimates (Liddell and Kruschke, 2018) and relax the assumptions around the homogeneity of variance. For the ratings, results are presented as model-estimated expected ratings and probit z-values; as the latter are on a standardised Gaussian scale, the values (i.e. effect sizes) can be interpreted akin to a Cohen’s d but on the latent space. Ordered beta models were used for proportion data (Kubinec, 2023) and logistic (Bernoulli) models were used for binary responses. Predicted probabilities per response level are presented for easier inference.

We consider the effect of statement emotionality on the perceived emotional behaviour of the witness for each gender. In the male condition, the emotional witness (M = −2.07, 95% CI = −2.36, −1.77) was rated as appearing substantially more distressed compared to the Unemotional witness (M = −0.78, 95% CI = −1.03, −0.56), z = 1.15, 95% CI [0.73, 1.59], pd = 100%, ROPE_[±0.11] = 0% (large and certain effect). In the female condition, a similar difference is observed, with the Emotional witness (M = −1.87, 95% CI = −2.09, −1.63) being rated as appearing more distressed than the unemotional (M = −1.03, 95% CI = −1.50, −0.59) witness, z = 0.62, 95% CI [0.16, 1.08], pd = 99.78%, ROPE_[±0.11] = 0.78% (moderate-to-large and certain effect).

We then consider differences in witness gender across emotionality conditions. In the emotional condition, the male witness’ behaviour was rated as more distressed compared to the female witness, but there is more uncertainty surrounding the difference, z = 0.34, 95% CI [0.00, 0.73], pd = 97.02%, ROPE_[±0.11] = 10.32% (small-to-moderate and uncertain). Inversely, in the unemotional conditions, the male witness was perceived as appearing on average less distressed than the female witness, but the evidence for the effect is poor, z = −0.19, 95% CI [−0.63, 0.23], pd = 82.58%, ROPE_[±0.11] = 26.15% (small, uncertain and unlikely). See Figure 1 (left panels).

Next, we consider the effects on the perceived emotional state of the witness. In the male condition, the emotional witness (M = −2.10, 95% CI = −2.38, −1.79) was rated as being substantially more distressed compared to the unemotional witness (M = −1.12, 95% CI = −1.35, −0.89), z = 0.92, 95% CI [0.53, 1.32], pd = 100%, ROPE_[±0.10] = 0% (large and certain effect). In the female condition, a similar difference is observed, with the emotional witness (M = −1.87, 95% CI = −2.12, −1.61) being rated as being more distressed than the unemotional witness (M = −1.24, 95% CI = −1.67, −0.81), z = 0.47, 95% CI [0.05, 0.92], pd = 98.65%, ROPE_[±0.10] = 3.57% (moderate and fairly probable, but imprecise).

Considering gender, in the emotional condition, the male witness’ behaviour was rated as more distressed compared to the female witness, but the effect is uncertain in size and direction, z = 0.35, 95% CI [−0.05, 0.74], pd = 95.80%, ROPE_[±0.10] = 9.53% (small-to-moderate and uncertain). Inversely, in the unemotional conditions, the male witness was perceived as appearing on average less distressed than the female witness, but the evidence is weak, z = −0.10, 95% CI [−0.50, 0.30], pd = 69.12%, ROPE_[±0.10] = 36.45% (negligible, uncertain and unlikely).

These results serve as a manipulation check for the stimuli, demonstrating perceptions that align with the goal of the experiment. Across all conditions, eyewitnesses were rated as being visibly distressed, with average ratings being negative, and participants not using the positive scale points almost at all (Figure 1, bottom panels). These results echo those of the perceived behaviour of the witness, finding that the manipulation was successful and that gender differences in perception are tentatively found only in the emotional condition (Figure 1, right panels).

Because of the sum-score nature of the data, the results (i.e. means and differences) are expressed in terms of a percent of maximum possible (POMP) scores (Cohen et al., 1999).

In the emotionality condition, participants showed little difference in the amount of affective empathy they expressed towards the male witness (57.5%, 95% CI [50.6, 64.2]) compared to the female witness (56.6%, 95% CI [49.6, 63.4]), POMP-difference = 0.95pp., 95% CI [−8.29, 10.00], pd = 57.86%, ROPE_[±5%] = 70.55% (uncertain effect). In the unemotional condition, participants expressed less affective empathy towards the female witness (43.1%, 95% CI [32.60, 53.30]) than the male witness (51.3%, 95% CI [45.0 57.3], POMP-difference = −8.20pp., 95% CI [−19.98, 3.32], pd = 91.83%, ROPE_[±5%] = 28.40% (uncertain effect). In the male witness condition, participants expressed less affective empathy in the unemotional condition than in the emotional condition, POMP-difference = −6.24pp., 95% CI [−15.42, 2.73], pd = 91.44%, ROPE_[±5%] = 38.97%. (uncertain effect). Finally, in the female witness condition, similarly, participants expressed less affective empathy in the unemotional condition compared to the emotional condition, POMP-difference = −13.49pp., 95% CI [−25.34, 1.42], pd = 98.65%, ROPE_[±5%] = 7.92%. See Figure 2 for details.

In the emotionality condition, participants showed little difference in the amount of cognitive empathy they expressed towards the male witness (79.7%, 95% CI [73.7, 85.5]) and the female witness (78.3%, 95% CI [71.8, 84.6]), POMP-difference = 1.39pp., 95% CI [−6.83, 9.70], pd = 62.64%, ROPE_[±5%] = 73.60% (uncertain effect). In the unemotional condition, similarly, participants expressed equivalent levels of cognitive empathy towards the male witness (73.9%, 95% CI [68.2, 79.5] as the female witness (73.2%, 95% CI [63.8, 81.9]), POMP-difference = 0.71pp., 95% CI [−9.72, 11.02], pd = 54.74%, ROPE_[±5%] = 65.58% (uncertain effect).

In the male witness condition, participants expressed less cognitive empathy in the unemotional condition than in the emotional condition, POMP-difference = −5.82pp., 95% CI [−13.86, 2.05], pd = 92.19%, ROPE_[±5%] = 41.85% (imprecise effect). In the female witness condition, similarly, participants expressed less cognitive empathy in the unemotional condition compared to the emotional condition; however, this effect is less certain, POMP-difference = −5.14pp., 95% CI [−15.48, 5.50], pd = 83.08%, ROPE_[±5%] = 46.66%.

In the emotionality condition, participants expressed less associative empathy towards the male witness (49.30%, 95% CI [42.0, 56.3]) than the female witness (54.2%, 95% CI [47.2, 61.8]), POMP-difference = −4.89pp., 95% CI [−14.67, 4.72], pd = 83.76%, ROPE_[±5%] = 49.50% (uncertain effect). In the unemotional condition, conversely, participants expressed higher levels of associative empathy towards the male witness (56.7%, 95% CI [50.7, 63.1] as the female witness (51.6%, 95% CI [41.6, 61.3]), POMP-difference = +5.17pp., 95% CI [−6.21, 16.42], pd = 82.09%, ROPE_[±5%] = 45.40% (uncertain effect).

For the male witness, participants expressed more associative empathy in the unemotional condition than in the emotional condition, POMP-difference = 7.43pp., 95% CI [−1.52, 17.01], pd = 94.63%, ROPE_[±5%] = 30.68%. For the female witness, conversely, participants expressed less associative empathy in the unemotional condition compared to the emotional condition; however, this effect is small and uncertain, POMP-difference = −2.60pp., 95% CI [−14.30, 8.95], pd = 67.06%, ROPE_[±5%] = 57.26%.

Accuracy was computed as a function of the number of correct answers provided to the 15 questions. If participants provided the complete correct answer, they were scored as 1; if they provided a fully incorrect answer, they were scored as 0, while if they provided a partial correct answer, they were scored based on the proportion [e.g. Question 1 (Q1) had 1/3 real details, and thus scores were 0 and 1; Q10 had 2/3 real details, and thus scores were 0, 0.5, and 1; Q11 had 3/3 real details, and thus scores were 0, 0.33, 0.66 and 1].

Accuracy was generally moderately high (64%); considering the varying number of options (3–6) and correct answers per question (1–4), and assuming all answers are provided per question, guessing for the task is an average probability of 34.44%. The task did not include a “do not recall” option to ensure participants engaged with the process more thoughtfully, in line with the principles of satisficing (Krosnick, 1991).

We first consider the effect of witness emotionality on memory recall. In the male condition, accuracy in the emotional witness condition (M = 64%, 95% CI = 59, 69) was similar to that in the unemotional witness (M = 62%, 95% CI = 58, 67), M_diff. = +2pp., 95% CI [−5, 8], pd = 69.47%, ROPE_[±5%] = 86.89% (no effect). In the female condition, accuracy was higher in the enemotional condition (M = 70%, 95% CI = 62, 76) than in the emotional condition (M = 64%, 95% CI = 59, 69), but the effect is quite uncertain, M_diff. = +6pp., 95% CI [−3, 14], pd = 89.83%, ROPE_[±5%] = 42.71%. Considering gender, in the emotional witness condition, accuracy in the male and female witness conditions was similar, M_diff. = 0.08pp., 95% CI [−7, 8], pd = 50.82%, ROPE_[±5%] = 87.26%. In the unemotional conditions, accuracy is higher for the female than the male witness, M_diff. = +7pp., 95% CI [−1, 16], pd = 95.73%, ROPE_[±5%] = 29.18%. See Figure 3 (left panel).

Participants’ susceptibility to misinformation is quantified as their accuracy in correctly recognising the statements that were true and not being swayed (i.e. agreeing) with statements that were false and aimed to mislead. Considering the 5 truthful statements, the effect of gender shows that in the emotionality conditions, the male condition (M = 65%, 95% CI = 59, 71) accuracy was similar to the female condition (M = 63%, 95% CI = 58, 69), M_diff. = +2pp., 95% CI [−6, 10], pd = 67.90%, ROPE_[±5%] = 78.13% (no effect).

In the unemotional condition, similarly, the male (M = 64%, 95% CI = 59, 69) and female (M = 63%, 95% CI = 56, 71) conditions showed similar scores, M_diff. = +0.9pp., 95% CI [−8, 10], pd = 57.45%, ROPE_[±5%] = 75.42% (no effect). Exploring the emotionality difference, we find no effect for male, M_diff. = +0.7pp., 95% CI [−6, 9], pd = 57.15%, ROPE_[±5%] = 83.84% (no effect), nor female, M_diff. = −0.2pp., 95% CI [−10, 9], pd = 51.85%, ROPE_[±5%] = 73.21% (no effect). See Figure 3 (middle panel).

Next, we consider the five misleading statements. Here, the percentage represents the amount by which participants were misled (i.e. 1 – accuracy). The effect of gender shows that in the emotional conditions, male witness (M = 46%, 95% CI = 40, 51) participants were less misled than female witness (M = 50%, 95% CI = 44, 55) participants, M_diff. = −4pp., 95% CI [−12, 3], pd = 85.75%, ROPE_[±5%] = 61% (small and uncertain effect). In the unemotional conditions, male witness (M = 46%, 95% CI = 42, 51) participants were more misled than female witness (M = 39%, 95% CI = 32, 47) participants, M_diff. = +7pp., 95% CI [−2, 15], pd = 94.17%, ROPE_[±5%] = 31.89% (moderate but imprecise effect).

Exploring emotionality differences, participants in the male witness condition showed similar levels of susceptibility to misleading information in both the emotional and unemotional conditions, M_diff. = −0.8pp., 95% CI [−8, 6], pd = 59.03%, ROPE_[±5%] = 88.74% (no effect). However, participants were misled less in the female unemotional witness condition than in the emotional female condition, M_diff. = −10pp., 95%CI [−19, 1], pd = 98.65%, ROPE_[±5%] = 10.58% (moderate but uncertain effect). See Figure 3 (right panel).

Perceptions of credibility were considered both from the perspective of the witness themselves and of the statement they provided.

We consider the effect of emotionality on the perceived credibility of the testimony provided. The testimony of the emotional male witness (M = 4.47, 95% CI = 4.19, 4.76) was rated as equally credible to the unemotional male witness (M = 4.50, 95% CI = 4.28, 4.71), z = 0.01, 95% CI [−0.38, 0.42], pd = 52.92%, ROPE_[±0.09] = 33.23% (no effect). The testimony of the emotional female witness (M = 4.43, 95% CI = 4.15, 4.72) was also rated similarly to the unemotional female witness (M = 4.34, 95% CI = 3.87, 4.73), z = 0.09, 95% CI [−0.49, 0.64], pd = 63.08%, ROPE_[±0.09] = 24.70% (no effect).

Considering gender differences, the testimony of the emotional female witness was rated similarly to the emotional male witness, z = −0.05, 95% CI [−0.51, 0.39], pd = 59.62%, ROPE_[±0.09] = 30.20% (no effect). Conversely, there is tentative evidence for the testimony of the unemotional female witness to be rated slightly lower than the unemotional male witness, z = −0.15, 95% CI [−0.68, 0.38], pd = 71.92%, ROPE_[±0.09] = 22.73% (small, uncertain and unlikely). See Figure 4 (left panels).

Next, we consider how credible the witness was perceived. The results largely follow the same pattern but with clearer trends. The testimony of the emotional male witness (M = 4.43, 95% CI = 4.20, 4.68) was rated slightly less credible than the unemotional male witness (M = 4.54, 95% CI = 4.30, 4.76), z = 0.17, 95% CI [−0.28, 0.61], pd = 77.10%, ROPE_[±0.09] = 22.35% (small effect, uncertain and unlikely). Conversely, the testimony of the emotional female witness (M = 4.53, 95% CI = 4.25, 4.80) was rated much higher than the Unemotional Female witness (M = 4.22, 95% CI = 3.85, 4.57), z = 0.44, 95% CI [−0.19, 1.05], pd = 92.75%, ROPE_[±0.09] = 7.25% (moderate but uncertain effect).

Considering gender differences, the testimony of the emotional female witness was rated higher than the emotional male witness, but there is substantial uncertainty in the result, z = 0.16, 95% CI [−0.35, 0.64], pd = 73.65%, ROPE_[±0.09] = 22.82% (small, uncertain and unlikely). Similarly, there is (clearer) evidence for the testimony of the unemotional female witness to be rated lower than the unemotional male witness, z = −0.46, 95% CI [−1.03, 0.16], pd = 94.12%, ROPE_[±0.09] = 7.10% (moderate but uncertain effect). See Figure 4 (right panels).

The accuracy for recalling the four details of the suspect was assessed across conditions. These relate to skin tone, general build, facial hair and facial piercings.

Accuracy was generally high across conditions (89%). First, considering the effect of witness emotionality, the emotional male witness condition (M = 84%, 95% CI = 74, 95) resulted in high accuracy, which was equal to that in the unemotional male witness condition (M = 84%, 95% CI = 73, 92), M_diff. = +3pp., 95% CI [−12, 17], pd = 65.58%, ROPE_[±5%] = 49.47% (no effect). The accuracy was lower in the emotional female condition (M = 83%, 95% CI = 70, 92) than in the unemotional female condition (M = 100%, 95% CI = 97, 100), M_diff. = −17pp., 95% CI [−29, −6], pd = 99.85%, ROPE_[±5%] = 0% (moderate and certain effect).

Considering gender, accuracy in the Emotional Male and Emotional Female witness conditions was similar, M_diff. = +4pp., 95% CI [−12, 19], pd = 68.92%, ROPE_[±5%] = 45.92% (no effect). Conversely, accuracy is higher for the Unemotional Female than the Unemotional Male condition, M_diff. = +16pp., 95% CI [−7, 26], pd = 99.92%, ROPE_[±5%] = 0% (moderate and certain effect) (Figure 5, first panel).

Accuracy was generally high across conditions (86%). Considering the effect of witness emotionality, the emotional male witness condition (M = 80%, 95% CI = 67, 90) resulted in high accuracy, but lower than in the unemotional male witness condition (M = 90%, 95% CI = 82, 96), M_diff. = −10pp., 95% CI [−24, 3], pd = 94.62%, ROPE_[±5%] = 19.84% (small but uncertain effect). The accuracy was similar in both the emotional female condition (M = 88%, 95% CI = 76, 96) and the unemotional female condition (M = 87%, 95% CI = 69, 97), M_diff. = +1pp., 95% CI [−15, 20], pd = 56.43%, ROPE_[±5%] = 46.24% (no effect).

Considering gender, accuracy in the emotional male was slightly lower than in the emotional female, M_diff. = −8pp., 95% CI [−23, 8], pd = 85.38%, ROPE_[±5%] = 30.79% (small but uncertain and unlikely effect). Conversely, accuracy was similar for the unemotional female and unemotional male conditions, M_diff. = +4pp., 95% CI [−11, 22], pd = 67.75%, ROPE_[±5%] = 46.47% (no effect) (Figure 5, second panel).

Accuracy was below chance across conditions, being the least well recalled detail (42%). The emotional male (M = 59%, 95% CI = 45, 73) resulted in higher accuracy than the unemotional male condition (M = 49%, 95% CI = 36, 61), M_diff. = +10pp., 95% CI [−9, 29], pd = 84.97%, ROPE_[±5%] = 24.89% (small but uncertain effect). Similar accuracy was observed in both the emotional female (M = 32%, 95% CI = 19, 47) and the unemotional female conditions (M = 28%, 95% CI = 12, 49), M_diff. = +4pp., 95% CI [−21, 25], pd = 61.02%, ROPE_[±5%] = 31.18% (no effect).

Next, accuracy in the emotional male was much higher than in the emotional female, M_diff. = +27pp., 95% CI [7, 48], pd = 99.33%, ROPE_[±5%] = 0% (moderate effect). Likewise, accuracy for the unemotional male was higher than the unemotional female conditions, M_diff. = +20pp., 95% CI [−2, 42], pd = 95.23%, ROPE_[±5%] = 8.39% (moderate but uncertain effect) (Figure 5, third panel).

Accuracy was high overall (86%). The emotional male (M = 87%, 95% CI = 75, 94) resulted in slightly lower accuracy than the unemotional male witness condition (M = 91%, 95% CI = 82, 96), M_diff. = −4pp., 95% CI [−17, 8], pd = 73.12%, ROPE_[±5%] = 52.95% (small, unlikely and uncertain effect). Comparably, accuracy was lower in the emotional female (M = 78%, 95%CI = 64, 89) than in the unemotional female condition (M = 86%, 95% CI = 68,97), M_diff. = −8pp., 95% CI [−26, 13], pd = 78.72%, ROPE_[±5%] = 29.66% (small, unlikely and uncertain effect).

Next, accuracy in the emotional male was slightly higher than in the emotional female, M_diff. = +8pp., 95% CI [−9, 23], pd = 85.25%, ROPE_[±5%] = 30.42% (small, unlikely and uncertain). Likewise, accuracy for the unemotional male was slightly higher than the unemotional female conditions but mostly negligible, M_diff. = +4pp., 95% CI [−11, 23], pd = 69.75%, ROPE_[±5%] = 46.34% (no effect) (Figure 5, fourth panel).

Bayesian Kendall tau correlations, using the correlation package (Makowski et al., 2022), were computed for all pre- and post-confidence questions, split across the four possible condition combinations. Because of the high number of comparisons, we focus on patterns within the heatmap matrices (Figure 6).

Heatmaps provide a visual representation of data patterns, where colour intensity reflects the magnitude of a given correlation. In our figure, darker colours indicate stronger correlations, while lighter colours indicate weaker correlations; red shades denote positive associations, and purple shades denote negative associations. This format allows for quick comparison across conditions and items, highlighting areas of relative increase or decrease. Axes labels denote the items being compared, and the colour scale provides a reference for interpreting the associations depicted. The full output can be found in the OSF folder (https://osf.io/cb5sk/?view_only=4e5983cce9d847abb1e68134714bac73).

In the emotional female condition, the mean rating across all six pre-task confidence questions was above the midpoint, M = 3.66. There is an average weak positive correlation between the six pre-task confidence questions, r_τ-b = 0.15, range: −0.03 to 0.43. The mean rating across the seven ^[2] post-task confidence questions was higher than at the start of the task, M = 3.78. The post-task associations across the seven questions show a moderate positive correlation, r_τ-b = 0.33, range: 0.09–0.71. The heatmap suggests negative pre–post associations, but the pattern is weak and unclear.

In the emotional male condition, the mean rating for the pre-task question was M = 3.52. Overall, there was a moderate-to-strong correlation, r_τ-b = 0.40, range: 0.22–0.70. The mean rating across the seven post-task questions was higher than at the start of the task, M = 3.91. The post-task associations show a moderate positive correlation, r_τ-b = 0.35, range: 0.16–0.72. The heatmap does not show a clear relationship between pre-post ratings.

In the unemotional female condition, the mean pre-task rating was M = 3.48. Overall, there was a moderate positive correlation, r_τ-b = 0.30, range: −0.02 to 0.67. The mean post-task rating was slightly higher than at pre-task, M = 3.88. The post-task associations show a moderate positive correlation, r_τ-b = 0.40, range: 0.07–0.80. The heatmap suggests a moderate positive pre–post association in ratings.

In the unemotional male condition, the mean pre-task rating was M = 3.52. Overall, there was a moderate positive correlation, r_τ-b = 0.35, range: 0.18–0.57. The mean post-task rating was slightly higher than at pre-task, M = 3.67. The post-task associations show a moderate-to-larger positive correlation, r_τ-b = 0.51, range: 0.31–0.84. The heatmap does not show any clear trends.

Understanding how an emotional witness influences perceptions and judgements is important for several reasons – including its effects on perceived credibility, vulnerability to misinformation and memory performance. In this study, we examined these factors in the context of individual judgements.

In the early stages of criminal investigations, practitioner judgements may be shaped by both the witness’ gender and emotionality. While much of the existing research on emotional victims has used group-based designs, rather than individuals (e.g. Ask and Landström, 2010; Kaufmann et al., 2003), this can introduce group dynamics that may alter decision-making (Hinsz et al., 1997). For instance, participants in groups may influence each other’s perceptions through discussion – an approach more suited to mock jury studies (Dahl et al., 2007). In contrast, our study examined responses at the individual level to better isolate how personal judgements are formed.

As expected, our emotionality manipulation was effective. Emotional witnesses were rated as being both physically and emotionally more distressed than their unemotional counterparts (see Figure 1). However, we were also interested in examining the interplay with gender. Typically, people expect males to show emotional restraint and females to display more emotionality.

We found that the emotional male witness was perceived as more distressed than the emotional female witness. One explanation for why this occurred is that participants viewed male emotional displays as indicative of extreme distress, overriding those typical expectations of male emotional restraint. By contrast, when appearing unemotional, the female witness was perceived as more distraught than the male witness. This reflects research showing that even when females maintain a neutral demeanour, they are still perceived as more emotional (Adams et al., 2012).

The empathic responses of participants towards the witnesses were assessed using the three empathy subscales. For affective empathy, which pertains to the ability to feel what another person feels, there were notable differences in the participants’ empathy levels towards the various witnesses. Overall, empathy levels were moderate; however, a substantial decrease was observed for the unemotional female witness, who saw below 50% on the POMP scale. The unemotional male witness received the second lowest empathy score, followed by the two emotional witnesses, who were rated similarly. These findings align with theoretical expectations, indicating that participants are more capable of experiencing empathy towards individuals displaying emotional behaviour. Additionally, consistent with gender expectation effects, the unemotional female witness, who did not exhibit emotions congruent with the situation, elicited the least empathy from the participants.

For cognitive empathy, which is the ability to understand what a person is feeling, a comparable pattern was observed. However, this pattern had less certainty (i.e. there was some overlap between effects). One explanation for this is a ceiling effect, as scores were high overall (70%–80% POMP range). Consequently, across all experimental conditions, the participants showed a strong understanding of the emotional state of the witness.

Finally, regarding associative empathy, which is the ability to relate to another’s experience by projecting oneself into it, the pattern of results was quite different. Participants associated most with the emotional female witness, meaning they could more easily imagine themselves behaving similarly in that situation. This aligns with gendered expectations that women are more likely to react emotionally to witnessing a crime. Participants also showed a similar level of associative empathy towards the unemotional male witness, supporting the expectation that a male witness of a crime would exhibit emotional restraint. Thus, the ratings for each gender align with stereotypical gendered norms (i.e. gender–emotion congruence).

In contrast, participants associated less with the emotional male and the unemotional female witnesses, potentially because of a behavioural norm incongruence. Emotional displays by males and unemotional displays by females do not align with stereotypical gender expectations, making it harder for participants to relate. This pattern indicates that participants more readily associate with a “stoic” male and an emotional female, consistent with gender–emotion expectations. When these expectations are not met, it can be harder for decision makers to relate to a witness.

We were interested in understanding how the gender–emotionality interaction impacts participants’ abilities to accurately recall information.

The relationship between memory accuracy and gender–emotionality is complex. We found that participants recalled accurate information 64% of the time. Interestingly, participants in the Unemotional Female condition more accurately remembered information compared to those in all other conditions (+7pp). This means that the proportion of correctly recalled details was higher if participants listened to a statement made by a female witness displaying a neutral demeanour. This highlights that emotional cues from a communicator (i.e. the female witness in our case) can divert a decision maker’s focus away from the testimony’s content and towards these emotional signals (see also Öhman and Mineka, 2001; Palermo and Rhodes, 2007).

In the context of memory accuracy and misinformation, Grice’s (1975) principles suggest that when emotional cues are absent, listeners (in this case, the participants) may focus more on the content of the statement. The lack of emotional distraction allows them to better adhere to the maxim of relevance, concentrating on the factual details being communicated. In our case, the unemotional female may have been viewed as incongruent with gendered expectations, resulting in more attention being given to the statement content without the participants being distracted by the emotions (or lack thereof) displayed. In other words, when a female witness is emotional (i.e. congruent with emotion–gender expectations), decision makers focus on (or are distracted by) the emotional cues. In contrast, when a female witness is unemotional (i.e. incongruent with emotion–gender expectations), decision makers focus more on the content, as they are not distracted by emotionality. Additionally, this lack of emotionality is unexpected, making them more alert. This explains why we only found a difference for the unemotional female and not the unemotional male witness.

Our data indicate that emotionality and gender impact attention, which may explain why some participants were distracted by emotions and subsequently more susceptible to misinformation. Specifically, participants were more prone to misinformation in the emotional female witness condition, while the unemotional female condition had the lowest susceptibility. Dovetailing with the memory results, these findings align with gender expectations and attention effects, suggesting that an emotional female witness may cause participants to focus more on emotional cues rather than factual details, thereby increasing the likelihood of misinformation during recall.

The witnesses in our research discussed the suspect and his physical appearance. Across the four details, no reliable pattern emerged in terms of witness gender or emotionality. Participants’ ability to remember suspect details was overall high (averaging 75% accuracy). Interestingly, across all conditions, participants struggled to remember what description was provided about facial hair (below chance accuracy). This could imply that certain aspects of physical appearance, such as facial hair, are harder to remember in witness testimonies. Tentatively, a trend towards the male witness conditions producing higher suspect detail recall was observed; however, the variability of the effects prohibits further speculation.

Participants generally rated the credibility of the witness highly, with one exception. The unemotional female witness received notably lower credibility ratings, particularly regarding her personal credibility. This suggests that when a female witness does not display strong emotions, participants may perceive her as less credible, potentially reflecting underlying gender expectations.

Participants were highly confident in their ability to recall accurate information about the witness testimony. Interestingly, after completing the task, this confidence rating improved. One possibility is that confidence and credibility judgements were impacted by the individual differences within our sample. Future research should examine this.

From a forensic perspective, capturing a range of confidence judgements in such research is clearly useful, given the range of associations observed. For example, participants may perceive certain conditions – such as those matching stereotypical expectations – as easier to evaluate, potentially inflating confidence without increasing accuracy. Identifying such patterns of overconfidence is important, as they may contribute to real-world misjudgements.

The study found that the emotionality and gender of a witness influenced participants’ perceptions. The findings align with the EVT (Burgoon and Hale, 1988; Burgoon, 1993). In situations where the witness violated gendered behaviour norms, participants’ perceptions were more negative. Deviations from the expectation of a male witness being more emotionally restrained or a female witness being more emotional resulted in marked patterns.

Such incongruencies affected the perceived credibility of the witness, with the unemotional female witness receiving the lowest ratings. This aligns with norms that females should act more emotionally in such situations. Likewise, perceived witness emotionality conformed to gendered expectations for males expressing emotions, which were taken as feeling stronger distress than their female counterparts, while females’ restrained behaviour was still interpreted as exhibiting stronger emotionality than their male counterparts.

Participants’ empathy towards the witness was similarly impacted, as ratings were generally lower for unemotional witnesses, especially for the female unemotional witness. In line with the above, associative empathy strongly indicated gendered expectations, with higher resonance of participants to the unemotional male and the emotional female. These patterns suggest that individuals have strong expectations of how male and female witnesses of a crime should behave.

While people may expect such emotionality, our data also suggests that emotional displays can distract decision makers and increase susceptibility to misinformation, especially when female witnesses are used. These findings imply that the emotionality and gender of a witness impact their credibility, the accuracy with which their statement is recalled, and the participants’ empathic response to them. Here, the unemotional female witness resulted in the most accurate memory recall and lowest rate of misinformation endorsed.

Future research should further explore these dynamics to allow us to better understand their implications in judicial settings. Finally, regardless of gender or emotionality, participants’ confidence was generally high.

These findings have practical relevance for police officers and legal professionals who routinely assess witness credibility. In law enforcement, where resources are limited (Porter and Gavin, 2024), officers are more likely to follow leads from witnesses they perceive as credible (Ask and Granhag, 2005; Kassin et al., 2003). However, credibility judgements are often influenced by nonverbal cues such as emotional expression, which may lead to accurate but emotionally expressive witnesses being unfairly dismissed (Vrij, 2008; Wells, 1993).

Legal practitioners should be mindful that overly emotional witnesses may appear more credible but may not aid juror recall, whereas less emotional witnesses may be perceived as less believable in spite of improving information retention. Practitioners may instruct clients/juries to focus on the content of statements rather than emotional displays to reduce bias and improve information processing and decision-making accuracy.

We recommend that practitioners – especially police officers, legal professionals and judges – be aware that emotional expressions from witnesses can distract from the actual content of their statements and should avoid overreliance on emotional cues when making credibility assessments. In our research, the perceived credibility of a witness was influenced by how well their emotional expression aligned with gender norms (e.g. emotional females and stoic males were rated more favourably). Likewise, it may be important to consider how court procedures aimed at reducing the emotional strain on a witness (e.g. live links) may produce a paradoxical effect of making them appear less sincere as they are not “emotional enough”. Our research can help us to understand these interplays, but additional research is needed to help us disentangle the effects.

Less emotional displays may actually be beneficial for female witnesses (and possibly victims). We found that the unemotional female witnesses elicited the most accurate memory recall in decision makers, and these were less susceptible to misinformation. Encouraging environments that allow witnesses to speak without pressure to display emotionality may support better fact-finding. As such, practitioners should be cautious not to misjudge credibility based on emotional congruence, as this can lead to biased evaluations.

Understanding gender norms in courtroom settings is important. Witnesses who violate gendered emotional norms (e.g. an unemotional female or emotional male) may be unfairly perceived as less credible or less relatable. We found that the unemotional witnesses both elicited the lowest empathic responses from our participants. These biases can shape empathic responses and decision-making, which may ultimately affect the fairness of investigations or trials.

This is the first study that examines emotion–gender expectations in the context of forensic decision-making outcomes. Our findings highlight the need for tailored guidelines for evaluating witness testimony that account for the influence of emotionality, gender and their interplay. Our recommendations aim to improve the accuracy and fairness of witness evaluations in forensic contexts by addressing the biases and challenges associated with emotional displays and gender expectations.

There are several limitations which must be considered when interpreting the results. Principally, the sample size is insufficient for precise effect estimates, with many results showing large uncertainty. Given the exploratory nature of the research and lack of empirical work in this area, a priori considerations were difficult. However, our modelling strategy focuses on accurate estimates of the most probable effects and should permit researchers to devise hypotheses and sample plans for future work.

The stimuli size should also be increased to ensure that current patterns are not because of actor-specific effects. Replications and extensions of the current work should use our materials to generate a corpus of witness videos with multiple exemplars per condition, avoiding the pseudoreplication issue. Correspondingly, the conditions and stimuli could be extended beyond two emotionality options, such as including a third incongruent emotion condition (e.g. positive valence). This would permit inferences about effect directionality and additional sources of judgement bias.

The gender of our participants may have influenced how much of the witness account was remembered, possibly because of cognitive differences between males and females (Astur et al., 1998; Lippa, 2005). For instance, females generally excel in face recognition (Rehnman and Herlitz, 2007) but tend to make more mistakes than males (Shapiro and Penrod, 1986). This is particularly evident in own-gender recognition, indicating an own-sex bias (Lewin and Herlitz, 2002; Wright and Sladden, 2003). Additionally, females often outperform males in describing people, especially victims Areh, 2011, while males are typically better at describing events (Areh, 2011). Although our research did not test for sex-based differences among decision makers, we acknowledge this may have had an impact. Future research should explore this further.

Eyewitness testimony is an important part of criminal investigations, yet our understanding of how a witness’ behaviour impacts decision-making judgements remains limited. This study examined the interplay between a witness’ gender and emotionality on the perceived credibility of their statement. Our study provides a methodological blueprint for future research, offering a pattern of results that can guide investigations into the emotional witness effect.

Potential extensions include using real witnesses or actors who witnessed a staged event to increase ecological validity, examining differences in emotional authenticity in testimonies and determining if decision makers can identify (in)sincere testimonies (Zloteanu and Krumhuber, 2021; Zloteanu et al., 2018). Additionally, researchers could investigate how emotion and gender stereotypes interact across different crime types and explore whether emotional effects are linear or exist only at extremes, determining if there is an “optimal” level of emotional display that influences judicial perception. Overall, the findings highlight the need for further exploration into these dynamics to better understand their implications in judicial settings.