How bad is bad? Perceptual differences in the communication of severity in intimate partner violence

The study addresses whether individuals who experience intimate partner violence (IPV) and those who evaluate narratives about IPV agree on the severity of the violence. Discrepancies can influence legal judgments, custody decisions, and relationship outcomes. IPV encompasses physical, psychological, and sexual harm, with substantial prevalence and varied consequences. Communicating severity is particularly challenging for psychological violence due to context dependence and second-order conditioning. Cognitive biases and cultural norms (e.g., gender stereotypes, media framing) may shape perceptions. The authors introduce a paradigm to quantify perceptual differences between narrators (victims, offenders, bystanders) and recipients (third-party readers) across violence types, defining four perceptual differences: calibration (level differences between narrator and recipient ratings), accuracy (correlation between narrator and recipient ratings), gender (effects of victim and rater gender), and role (victim/offender/bystander differences). They hypothesize: H1, recipients will generally rate higher than narrators, especially for physical/sexual violence and for victim/bystander narratives; H2, lower accuracy (weaker correlation) for psychological than for physical/sexual narratives; H3, role effects (victims > offenders; bystanders > victims); H4, gender effects (higher severity for female victims; female recipients rate higher than male recipients).

Forms and prevalence: IPV includes physical (e.g., hitting, pushing), sexual (forced/attempted sexual contact), and psychological (threats, humiliation, control) violence. Prevalence is high; psychological violence is common but harder to document legally. Definitions and measurement inconsistencies, especially for intimate partner sexual violence, complicate research. Psychological abuse can have severe, lasting effects and is less visible and context-dependent. Perceptual differences and communication: Cognitive biases and heuristics influence evaluations, shaped by culture and media. Prior work shows difficulties communicating IPV severity, especially psychological IPV; recipients often base severity on type rather than description and differ from experiencers; ratings vary by narrator role (victim/offender/witness). Offenders may minimize, justify, or shift blame; bystander perspectives can affect legal outcomes. Gender and IPV: Stereotypes influence perceptions; scenarios with female victims and male perpetrators are typically rated as more severe. Male recipients often rate IPV as less severe, attribute more blame to victims, and accept rape myths more than females in some literature. Media framing, alcohol presence, and language aggressiveness also affect judgments. These patterns motivate testing gender-related perceptual differences in victim and rater genders within IPV narratives.

Design: Mixed design with between-group factor of self-experienced (narrator) vs communicated (recipient) violence. Within-group variables: type of violence (sexual, physical, psychological), narrator role (offender, victim, bystander), participant gender (male/female), and victim gender (male/female). Dependent variable: severity rating (0–10). Participants: Recruited via Prolific (UK, age 18–45, heterosexual, English-speaking, minimum 6-month relationship). Phase 1: 300 recruited; 287 included (M age = 31.65, SD = 7.11; 72 men, 215 women). Phase 2: 608 recorded; 489 included after applying the same inclusion criteria (sexual: 307 female, 173 male; psychological: 315 female, 173 male; physical: 313 female, 176 male). Procedure:

• Phase 1 (Narrator generation): Each participant wrote 9 first-person narratives (≥50 words) about IPV in a heterosexual romantic relationship of ≥6 months: 3 as victim (sexual, physical, psychological), 3 as offender (sexual, physical, psychological), 3 as bystander (sexual, physical, psychological). Definitions of violence were provided and confirmed. If no direct experience, participants described the most similar situation they had experienced. After each narrative, participants rated severity (0–10). Exclusions (narrative-level) were applied if: the requested violence type was denied, violence was not IPV or not heterosexual, the wrong type/role was described, second-hand accounts, or illegible writing. Result: 1,596 narratives included (SO 123; SV 171; SB 83; PsO 187; PsV 222; PsB 227; PhO 179; PhV 196; PhB 208); 987 excluded.
• Phase 2 (Recipient ratings): New participants each read 9 narratives (one per scenario: 3 violence types × 3 roles), randomized, and rated severity (0–10) and provided five keywords per narrative. Demographics were collected. Narratives lacking ratings (n=109) and duplicate server responses were excluded. Final scenario Ns (ratings): SO 490; SV 490; SB 481; PsO 489; PsV 501; PsB 494; PhO 492; PhV 492; PhB 489. Analysis: Where possible, Phase 1 and Phase 2 ratings were paired at the narrative level (fixed effects). Pairing as fixed effects: by narratives for Phase comparisons; by participant for role comparisons; by narratives for gender comparisons. Tests included paired-sample t-tests (narrator vs recipient), mixed ANOVAs (narrator vs recipient × role within each violence type), univariate ANOVA on discrepancy by violence type, correlations (Pearson r) between narrator and recipient ratings (accuracy), and t-tests/ANOVAs for gender effects. Demographics showed no significant Phase 1 vs 2 differences and were not included as covariates.

General calibration (H1): Recipients rated severity higher than narrators overall (recipients M=5.54, SD=2.97; narrators M=4.83, SD=3.07), t(4417) = -14.37, p < 0.001. Across nine scenarios, recipients typically rated higher (Table 2), except:

• Physical offender (PhO): narrators rated higher than recipients (narrators M=4.81, SD=3.10; recipients M=4.23, SD=2.83), t(491)=3.55, p<0.001.
• Psychological bystander (PsB): no significant difference (narrators M=6.32, SD=2.22; recipients M=6.24, SD=2.25), t(493)=0.61, p=0.543. By scenario examples (Table 2):
• Sexual victim (SV): narrators M=5.07 (SD=3.22) vs recipients M=6.42 (SD=3.02), t(489)=-10.03, p<0.001.
• Physical victim (PhV): narrators M=5.11 (SD=3.34) vs recipients M=6.00 (SD=3.10), t(491)=-5.39, p<0.001.
• Sexual offender (SO): narrators M=2.63 (SD=2.37) vs recipients M=3.48 (SD=2.85), t(489)=-6.07, p<0.001. Mixed ANOVAs:
• Physical violence: main effect recipients > narrators, F(1,1470)=91.07, p<0.001, partial η²=0.06; interaction with role F(2,1470)=82.52, p<0.001, partial η²=0.10.
• Sexual violence: main effect recipients > narrators, F(1,1458)=136.50, p<0.001, partial η²=0.09; role interaction F(2,1458)=11.63, p<0.001, partial η²=0.02; largest discrepancy for victim narratives.
• Psychological violence: main effect recipients > narrators, F=19.21, p<0.001, partial η²=0.01; role interaction F(1,1481)=8.68, p=0.01, partial η²=0.01; largest discrepancy for victim narratives; no difference for bystander narratives. Violence type comparison of discrepancies: differences significant, F(2,4415)=14.18, p<0.001. Calibration discrepancy smaller for psychological (M=0.34, SD=2.97) than sexual (M=0.89, SD=2.93) and physical (M=0.89, SD=3.77); no difference between sexual and physical (Tukey p=0.999). Accuracy (H2): Pearson correlations between narrator and recipient ratings were significant (p<0.001) but low-to-moderate (Table 3). Lowest for psychological offenders (r=0.25) and psychological bystanders (r=0.21), moderate for psychological victims (r=0.35). Sexual violence showed higher correlations (SV r=0.54; SB r=0.526; SO r=0.305). Physical offenders were low (r=0.265); physical victims r=0.358; physical bystanders r=0.301. H2 partially supported (accuracy notably poorer for psychological violence and for offender narratives). Role differences (H3): Offenders rated lower than victims, t(1011)=6.7, p<0.001 (H3a supported). Bystanders rated higher than victims, t(1020)=4.2, p<0.001 (H3b supported). Calibration interacted with role: larger discrepancies for victim/bystander narratives than offender narratives. Gender differences (H4):
• Victim gender (H4a): Larger narrator–recipient discrepancy when the victim was female (M=0.94, SD=3.27) vs male (M=0.25, SD=3.30), t(2952)=5.75, p<0.001. In rater×victim gender ANOVA, F(3,2950)=19.85, p<0.001; Tukey: female raters rated female-victim narratives higher than female raters rating male-victim narratives (difference significant, p<0.001). Male raters showed no significant difference between male vs female victim narratives (p=0.07).
• Rater gender (H4b): Contrary to hypothesis, discrepancies were larger for male raters (M=1.02, SD=3.11) than female raters (M=0.37, SD=3.38), t(2952)=-5.16, p<0.001; H4b not supported.

Findings demonstrate robust perceptual differences in communication of IPV severity. Generally, recipients rated higher than narrators (calibration difference), especially for physical and sexual violence and for victim and bystander narratives. Exceptions include physical offender narratives (narrators higher than recipients), possibly reflecting offender guilt or stigma leading to heightened self-ratings and downplaying in narrative content, and psychological bystander narratives (no difference), consistent with the difficulty of understanding context-dependent psychological abuse without intimate knowledge of the relationship. Accuracy differences indicate that narrative content only moderately predicts recipient ratings; psychological violence is particularly hard to communicate accurately due to its learned, second-order nature. Offender narratives also showed lower accuracy, consistent with self-interest in minimizing or vagueness that reduces predictability of recipient judgments. Role effects replicate established patterns: offenders downplay severity relative to victims; bystanders rate higher than victims, supporting validity of the paradigm. Gender-related discrepancies suggest that violence against female victims is perceived as more severe and that cross-gender rater–victim pairings amplify discrepancies, complicating assumptions that male raters are consistently less sympathetic. These systematic perceptual differences have direct implications for legal and clinical assessments: overestimation risks for physical/sexual violence and under-communication risks for psychological violence warrant careful consideration by professionals interpreting language-based accounts.

The study shows a general perceptual difference in calibration: recipients typically rate IPV severity higher than narrators, more strongly for physical and sexual violence than for psychological violence. Accuracy between narrator and recipient ratings is low-to-moderate across scenarios and is particularly poor for psychological violence and offender narratives, indicating challenges in communicating and interpreting such events. Gender-related perceptual differences were evident: narratives with female victims were rated more severely, and male raters exhibited larger discrepancies than female raters, especially in cross-gender contexts. These findings highlight the need for awareness and potential corrective strategies in legal and professional settings to mitigate miscommunication and bias when evaluating IPV narratives.

• Causal mechanisms (e.g., desensitization vs cognitive dissonance) cannot be determined from the design.
• Potential selection artifact: Phase 1 involved narrative-writing exclusions, while Phase 2 had no analogous exclusion criteria; if excluded Phase 1 participants differ systematically (e.g., socioeconomic status), results may be biased.
• Generalizability limits: restricted to UK-based, heterosexual participants aged 18–45 and English speakers; LGBTQ+ populations and broader demographics not included.
• Accuracy was never high; psychological violence particularly difficult to communicate, limiting interpretability of severity from narratives.
• Online sample (Prolific) and self-report narratives may introduce biases (memory, social desirability, narrative quality).
• The unexpected reversal for physical offender narratives requires replication to confirm robustness.
• Data availability is restricted due to ethical concerns, limiting external verification and reanalysis.