The motivation behind the instrument lies in a long-standing measurement problem. Most existing validated questionnaires assess problematic or compulsive patterns of pornography viewing, not the nature of the content consumed. Yet meta-analytic evidence, including work summarized by Wright and colleagues, suggests that associations between pornography and sexual aggression are stronger for violent than for nonviolent material. Without a validated way to distinguish content types, researchers have effectively treated pornography as a homogeneous category, making it difficult to test whether the violent portion of what people watch matters independently of how much they watch overall.

The PVES consists of ten thematic blocks covering specific content categories: nudity without sexual activity, consensual sexual relations, domination or submission without physical pain, body fluid play, sex with intoxicated or unconscious persons, verbal threats or insults, non-consensual relations with force, physical aggression such as hair pulling or slapping, asphyxiation, and BDSM with pain-inflicting implements. Each block contains sub-items differentiated by the sex composition and role of the depicted participants, yielding 38 sub-items in total. Respondents rated how often they encountered each content type when going online to watch pornography over the past 12 months, on a five-point frequency scale from never to always.

The ten blocks were classified into three graded levels of depicted violence. Level 1 covers non-violent practices, including nudity, consensual relations, domination without pain, and body fluid play. Level 2 groups coercive depictions without explicit on-screen physical aggression, such as sex with incapacitated persons, verbal threats, and non-consensual relations with force, categories that would constitute sexual violence under most legal frameworks. Level 3 comprises material with explicit physical aggression or pain infliction, including physical aggression, asphyxiation, and BDSM with implements. The authors stress that these designations describe the acts depicted, not their legal or ethical status, and that the classification targets the on-screen act rather than the consent framework portrayed around it.

Statistically, the scale’s structure was best captured by a bifactor model, in which a single general factor saturates all ten content blocks while three orthogonal specific factors capture variance unique to each violence level. The optimized bifactor model showed good fit, with a comparative fit index of 0.978, a Tucker-Lewis index of 0.957, and a root mean square error of approximation of 0.076. Critically, the general factor was dominant: it explained 81 percent of the common variance, with an omega-hierarchical reliability of 0.84, while the specific factors showed low reliabilities ranging from 0.11 to 0.34. This means most of the reliable variance in responses reflects a single dimension of graded violent-content exposure rather than separable content-type dimensions.

Importantly, the general factor appears to measure violent content specifically rather than overall consumption volume. It correlated 0.96 with Level 3 block sums and 0.91 with Level 2 sums, but only 0.42 with Level 1 content and just 0.32 with consensual sexual relations alone. Individuals with high exposure to non-violent material but low exposure to coercive or violent content consistently scored below the mean on the general factor. The scale also correlated only moderately with self-reported viewing frequency, with a Spearman coefficient of 0.34, further supporting the interpretation that elevated scores reflect the coercive and violent portion of what is viewed rather than viewing as such.

The validation battery produced a striking contrast between convergent and nomological associations. Correlations with sexual violence outcomes, measured with the Sexual Aggression and Victimization Scale and with sextortion threat items, averaged 0.24 in absolute magnitude, with victimization correlations ranging from 0.22 to 0.25 across partner, friend, and stranger contexts and perpetration correlations from 0.24 to 0.26. By contrast, correlations with sexual arousal, sexual enjoyment, orgasm frequency, partner evaluation, and self-evaluation averaged just 0.03, with most confidence intervals including zero. The ratio of mean convergent to mean nomological correlations was 8.2 to 1, which the authors present as strong quantitative evidence of construct specificity.

Incremental validity analyses reinforced this picture. In hierarchical logistic regressions controlling first for age and social desirability responding, adding PVES scores increased the Nagelkerke R-squared for sexual victimization from 0.014 to 0.102, an increment of 0.088, and for sexual perpetration from 0.004 to 0.205, an increment of 0.200. The corresponding odds ratios were 2.00 for victimization and 2.68 for perpetration per unit of the factor score. When participant sex was added in a third step, the perpetration effect was essentially unchanged, while the victimization effect strengthened to an odds ratio of 2.17, a suppressor pattern in which women reported more than twice the odds of victimization despite lower violent-content exposure. Receiver operating characteristic analysis showed acceptable discrimination for perpetration, with an area under the curve of 0.753, and more modest discrimination for victimization, at 0.660.

The study also established strict measurement invariance across sex and across age groups, meeting conventional thresholds at every step from configural to strict invariance. This means the scale measures the same construct equivalently in men and women and in younger adults aged 18 to 29 versus older adults aged 30 to 35, so group differences in scores would reflect genuine differences in exposure rather than measurement bias. A 50/50 cross-validation split stratified by sex and age confirmed the stability of the bifactor structure, with all general factor loadings differing by less than 0.10 between halves. Sensitivity analyses using sub-scores for violence directed at female versus male depicted victims showed that both correlated very highly with the general factor, at 0.96 and 0.94 respectively, and each independently predicted both outcomes.

The authors are careful to frame what the findings do and do not show. The cross-sectional design precludes causal inference about whether violent-content exposure influences sexual violence or vice versa, and the literature itself remains contested, with some researchers arguing that observed associations may reflect pre-existing individual characteristics rather than causal effects. The sample was restricted to young Spanish adults who watch pornography, recruited through a volunteer online panel despite probability-based sampling procedures, and relied on retrospective self-report. The scale aggregates across actor-gender configurations within each block, and the authors note it should be used for population-level research on associations, not to characterize, screen, or label individuals. Within those boundaries, they argue the PVES opens the door to more precise research on whether specific content types, rather than pornography viewing in general, are associated with sexual violence outcomes, and to prevention and education initiatives informed by that finer-grained evidence.

Subject of Research: Development and validation of the Pornography Violence Exposure Scale, a bifactor measure of graded exposure to violent pornographic content

Article Title: Pornography Violence Exposure Scale: a graded bifactor measure of exposure to violent pornographic content

Article References: Estévez-García, J. F., Pérez-Martínez, V., Cerdán-Torregrosa, A., Vives-Cases, C., Otero-García, L., & Sanz-Barbero, B. (2026). Pornography Violence Exposure Scale: a graded bifactor measure of exposure to violent pornographic content. Frontiers in Psychology, 17, 1885321. https://doi.org/10.3389/fpsyg.2026.1885321

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1885321

Keywords: pornography, sexual violence, psychometrics, bifactor model, measurement invariance, young adults, Spain, scale validation, sexual aggression, victimization, ROC analysis, cross-sectional study