Identifying Subtypes of Image-Based Sexual Abuse and their Distinct Incident Characteristics, Victimization Histories, and Personal Correlates

Abstract

Image-based sexual abuse is common among adolescents, yet most research treats it as a single, uniform construct, obscuring important variation in how image-based harms occur and who is most vulnerable. This lack of differentiation represents a central gap in the field, limiting theoretical development and hindering understanding of the behavioral and relational dynamics that shape youth image-based sexual abuse. The present study addresses this gap by examining incident-level heterogeneity in image-based sexual abuse experiences that occurred before age 18. Data were drawn from a national sample of 2,854 young adults aged 18–28 (Mean age = 22.16, SD = 3.1) who reported details on 4,205 unique image-based sexual abuse incidents. Almost three-quarters (73.6%) of participants were female at birth, 25.3% identified with a gender minority identity, and 62.2% with a sexual minority identity. Incident-level latent class analysis of types of image-based sexual abuse identified four distinct profiles: (1) Coerced Production and Distribution: peer-driven pressure to produce sexual content, marked by frequent, sustained incidents and low perceived intent to harm; (2) Non-Coerced Image Sharing with Older Individuals: age-discrepant exchanges with people 5 + years older that, while lacking overt coercion, raise concerns about grooming and exploitation; (3) Image Re-Distribution: peer-based non-consensual sharing and the highest rates of disclosure; and (4) Threat-Based Exploitation: blackmail, commercial exchange, and predatory adult perpetrators. Findings demonstrate that image-based sexual abuse is not a single type of harm but a set of behaviorally and relationally distinct patterns. By identifying the specific constellations of behaviors, this typology provides a clearer foundation for future theory-building and research on mechanisms, resilience, and risk in image-based sexual abuse among youth.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10964-026-02316-x.

Introduction

Image-based sexual abuse has emerged as a significant form of technology-facilitated sexual harm among adolescents, with growing evidence that such experiences carry substantial developmental, emotional, and relational consequences (Hellevik et al., 2025). Despite its prevalence and severity, much of the existing research treats image-based sexual abuse as a single, uniform phenomenon, overlooking the diverse ways in which image-based harm occurs, the different relationships and power dynamics involved, and the distinct vulnerabilities faced by various youth populations (Stänicke et al., 2024). This lack of differentiation limits theoretical understanding and obscures important behavioral patterns that may shape risk, severity, and developmental impact. This study addresses this gap by using an incident-level, behavior-based approach to identify distinct constellations of image-based sexual abuse experiences prior to age 18, laying the groundwork for a more refined understanding of how youth encounter and are affected by image-based harm.

Image-based sexual abuse refers to non-consensual taking, sharing or threats to share sexually explicit images or videos involving individuals under the age of 18 (18 USCS § 2256). While this form of abuse also occurs in adulthood (Umbach et al., 2025), the present study focuses exclusively on incidents that occurred prior to the age of 18 due to the distinct developmental, legal, and clinical implications of image-based harm experienced during childhood and adolescence (Finkelhor et al., 2023; Turner et al., 2025). In the United States, any visual depiction of sexually explicit conduct involving a minor is legally classified as child pornography, increasingly referred to as child sexual abuse material, and is prohibited under federal statutes such as 18 U.S.C. §§ 2251 to 2252 A and the definitional Sect. 18 U.S.C. § 2256. At the state level, many jurisdictions have enacted specific laws related to revenge pornography or sexual cyber-harassment, and others address image-based sexual abuse through existing statutes related to child pornography, harassment, or cyberstalking. Recent data underscores the growing prevalence and complexity of image-based sexual abuse. In 2023, the National Center for Missing & Exploited Children received 36.2 million reports of suspected online child sexual exploitation, with 35.9 million related to image-based sexual abuse (NCMEC, 2023). In a nationally representative sample from the United States, 15.6% of young adults reported online child sexual abuse, and 11.0% reported image-based sexual abuse during childhood, 88% of which involved youth-produced content, typically among adolescents aged 13 to 17 (Finkelhor et al., 2022). One of the defining features of image-based sexual abuse is the indefinite digital circulation of explicit content, which can resurface unpredictably and remain accessible long after the initial incident. This ongoing exposure often results in chronic psychological distress, internalized shame, and a persistent sense of vulnerability (Gewirtz-Meydan et al., 2018a, b). Image-based sexual abuse has been associated with a range of mental health difficulties, including anxiety, depression, post-traumatic stress symptoms, and social withdrawal (Chauviré-Geib & Fegert, 2024; Paradiso et al., 2024; Schmidt et al., 2023).

Despite the harm caused by image-based sexual abuse, much of the existing literature treats image-based sexual abuse as a single kind of abuse experience, which obscures important differences in nature, severity, and context of the victimization, including variations in the identity of the person responsible, relational context, coercive tactics, repetition, duration, and the presence of threats or commercial elements (Stänicke et al., 2024). This is particularly concerning given the different pathways through which image-based sexual abuse unfold, frequently existing along a continuum of agency, coercion, and exploitation. In response, broader and more contemporary conceptualizations have emerged, aiming to capture the multifaceted nature of image-based sexual abuse, including adult made images, images non-consensually made by other youth, and self-produced images (Finkelhor et al., 2023).

Youth may experience image-based sexual abuse through a range of unauthorized actions, such as having sexual images taken, created, or distributed without their consent. Others may be subjected to coercive dynamics, including threats or psychological pressure to provide sexual images. More overtly exploitative scenarios involve leveraging images for further gains, such as threatening to release content unless sex, money, or additional images are provided, or using broader forms of manipulation and control (Finkelhor et al., 2023; Gewirtz-Meydan et al., 2025). Additional vulnerabilities include adolescents who share sexual images with much older individuals while under the age of 18 (Quayle et al., 2018), or those who engage in image exchange for monetary or material compensation (Walsh & Tener, 2022). These diverse experiences reflect distinct relational contexts and varying degrees of power imbalance and consent. Yet, they are often collapsed into a simplistic understanding of image-based sexual abuse, limiting the understanding of victimization profiles and the differential risks they entail.

A growing body of theoretical work also suggests that image-based sexual abuse is shaped by broader developmental, gendered, and sociocultural forces. Gendered power models and sexual script theory highlight that girls and young women often navigate relational expectations and coercive pressures that differ markedly from those experienced by boys, while boys may be exposed to more anonymous, financially driven, or exploitative interactions (Reed et al., 2017). Sexual minority stress models further emphasize that sexual and gender minority youth face unique vulnerabilities stemming from stigma, limited access to affirming environments, and greater reliance on digital spaces for exploration and connection (Meechan-Rogers et al., 2022). These intersecting frameworks underscore that image-based sexual abuse is not a uniform experience but varies substantially across gender, sexual identity, and social context. This heterogeneity reinforces the importance of identifying empirically derived profiles rather than treating image-based sexual abuse as a single category, as different groups of adolescents encounter distinct pathways, dynamics, and risks that require differentiated analysis and response (Flynn et al., 2023).

Prior research indicates that image-based sexual abuse in adolescence typically unfolds across several relational contexts. Within romantic or dating relationships, consensual “sexting” can become abusive when images are subsequently redistributed or weaponized without consent, blurring the line between normative sexual exploration and victimization (Dolev-Cohen & Shaul, 2025). In peer networks, image sharing is often intertwined with bullying, humiliation, or attempts to control social status. A third context involves interactions with much older individuals or strangers, where youth-produced images are elicited through grooming, deception, or extortion and leveraged for further sexual or commercial gain (Paradiso et al., 2024). These contexts are not evenly distributed: girls and young women more often report image-based sexual abuse linked to intimate partner violence and reputational harm, while sexual and gender minority youth face elevated risks for technology-facilitated abuse, sometimes within age-discrepant or clandestine relationships that fill gaps in support and affirmation (Paradiso et al., 2024).

To date, most empirical studies have treated image-based sexual abuse as a single, unitary phenomenon. In many cases, research relies on one or two global items asking whether youth have “ever” experienced image-based abuse, an approach that obscures the wide variety of image-based sexual abuse patterns across behaviors, motivations, and relational contexts (Flynn et al., 2023). Even when multiple behaviors are assessed, these experiences are typically combined into a single composite score, and analyses remain variable-centered. This approach has been essential for establishing that image-based sexual abuse is widespread and linked to significant distress (Gewirtz-Meydan et al., 2025), but it obscures important heterogeneity in how digital sexual harm unfolds and contrasts to typological work that emphasizes image-based sexual abuse as a heterogeneous and multi-dimensional form of harm. A small number of recent studies have begun to disaggregate image-based sexual abuse experiences, for instance, by parsing different forms of digital sexual harassment or by identifying high-risk subgroups based on mental health and adversity profiles (Mitchell et al., 2025a, b). However, these efforts remain largely person-level and rarely focus on the specific incident-level constellations of behaviors and contexts that characterize youth image-based sexual abuse. As a result, prevention and policy discussions may treat image-based sexual abuse as a monolithic problem, which can contribute to generalized alarm about any sexual image-sharing among adolescents while making it difficult to identify which patterns are most severe and which youth are most vulnerable. Rather than treating all image-based sexual abuse incidents as equivalent, the present study derives subtypes based on the patterns of co-occurring behaviors within specific incidents and compares the resulting classes on incident characteristics, prior victimization, and sociodemographic vulnerabilities. This analytic strategy moves beyond binary or reductionist classifications and captures the multifaceted and heterogeneous nature of image-based sexual abuse among youth, highlighting constellations of risk that may warrant more targeted clinical, legal, and policy responses.

Current Study

There is a lack of differentiation in existing research on image-based sexual abuse, which has typically treated this as a single, uniform phenomenon rather than a set of behaviorally and contextually distinct experiences. To address this gap, the current study identifies unique constellations of image-based sexual abuse behaviors that occurred before age 18 using an incident-level latent class approach. Anticipated dimensions included the degree of coercion, non-consensual redistribution, involvement of older or adult partners, and the presence of threats or commercial elements. Beyond identifying behavioral profiles, the study also examined how classes differed in incident-level characteristics (e.g., relationship with person responsible, intent to harm), prior victimization and adversity, and sociodemographic vulnerabilities. By clarifying how distinct forms of image-based sexual abuse are organized and how they relate to individual and contextual factors, this study provides a more developmentally informed foundation for future theory-building and empirical work on youth image-based harm.

Methods

Participants

This study utilized data from an online survey aimed at understanding the characteristics and impact of image-based sexual abuse involvement. The study was conducted in the United States and designed to oversample sexual and gender minority individuals and ensure sociodemographic diversity through targeted enrollment quotas. Specifically, recruitment aimed for approximately 50% of participants to be assigned female at birth, 80% to identify as sexual minorities (romantic or sexual attraction to someone that differs from the societal norm of heterosexuality), at least 20% to identify as Hispanic, 20% as Black, and a minimum of 20% to identify as gender minorities (someone whose gender identity differs from the sex they were assigned at birth). Once these quotas were met, individuals matching those characteristics were no longer eligible to continue to the next part of the survey (consent document). Data collection occurred between June 28, 2023, and April 1, 2024.

A total of 8,596 participants started the survey. During data cleaning, cases were excluded due to declined consent (n = 227), suspected duplication or fraudulent entries (i.e. bots or scammers) (n = 289), incomplete survey responses (n = 1,701), or missing screener data (n = 161). An additional 48 participants with ≥ 90% survey completion were retained. After removing 11 participants with missing data on key latent class indicators, the final sample used for the analyses comprised 6,255 individuals. For descriptive purposes, 29 additional cases with excessive missing data on key study constructs were excluded, resulting in a final sample in the overall study of 6,226 participants. The present analyses focus specifically on the subsample of participants who reported image-based sexual abuse involvement (46%). This subset includes 2,854 individuals reporting on 4,205 unique incidents and forms the basis for all analyses presented in this manuscript. Descriptive characteristics of these 2,854 participants are provided in Table 1.

Table 1.

Demographic characteristics of the analytic sample (N = 2,854)

Characteristic	%	n
Sex assigned at birth
Male	23.7	676
Female	73.6	2100
Intersex	1.2	33
Not reported	1.6	45
Gender identity
Exclusively cisgender	71.4	2037
Transgender (any)	9.4	267
Non-binary	14.7	418
Other gender minority identity	0.6	17
Questioning	0.8	24
Not reported	3.2	88
Sexual identity
Exclusively heterosexual	35.6	1015
Gay or lesbian (any)	15.2	433
Bisexual	44.3	1264
Other sexual minority identity	1.8	52
Questioning	0.7	20
Not reported	2.5	70
Race
White	64.7	1848
Black	5.6	159
Asian	5.0	142
Hawaiian or Pacific Islander	0.3	9
American Indian or Alaska Native	1.4	40
Other race	4.4	125
Two or more races	12.4	353
Not reported	6.2	178
Ethnicity
Not of Hispanic or Latino origin	75.1	2141
Hispanic or Latino origin	21.1	603
Not reported	3.8	109
Household income
Lower than average family	42.4	1211
About as the average family	39.0	1113
Higher than the average family	12.5	357
Not reported	6.1	173
Highest level of education
Less than high school	5.3	152
High school, GED, trade school	28.2	805
Some college	29.8	850
Bachelor’s degree	17.3	494
Graduate school	15.9	454
Not reported	3.5	99
Age (SD)	M = 22.16	SD = 3.1

Procedure

Participants aged 18 to 28 were recruited from across the United States using paid, targeted advertisements on Facebook and Instagram. These recruitment messages did not disclose eligibility requirements or mention any financial compensation. Instead, they included general calls to action such as “make a difference” or “have your voice heard” to encourage interest. Eligibility criteria required participants to (1) be between 18 and 28 years old, (2) have proficiently with the English language, and (3) currently residing in the United States. In addition to these criteria, the recruitment strategy was designed to oversample individuals likely to have experienced image-based sexual abuse. This was achieved through two screening items assessing the likelihood that someone had either (1) been asked for a sexual photo or video or (2) sent one themselves. Both questions were rated on a 5-point scale from 1 (not at all likely) to 5 (extremely likely). Initially, those who answered a likelihood of “2” or above on the screener questions were directed to the full survey. This criterion was modified to “3” during recruitment to maximize the number of participants with image-based sexual abuse.

Because this was a convenience sample, the findings are not intended to reflect population-level estimates. Instead, the sampling approach was designed to explore risk factors and behavioral patterns among individuals at heightened risk for image-based sexual abuse. Recruiting through social media enabled access to a broader and more diverse population, including individuals from marginalized groups that are typically underrepresented in panel surveys and school-based samples, and who are known to be at higher risk for image-based sexual abuse (Parton & Rogers, 2025; Turner et al., 2024). Interested individuals completed an eligibility screener and, if eligible (57% of completed screeners), were directed to a secure online survey hosted on the Qualtrics platform.

Participants spent an average of 29.8 min completing the survey (26.7 min for participants with no image-based sexual abuse involvement and 33.7 min for those with, p = .04). Those who completed the entire survey and provided a valid mailing address (which was confirmed via smarty.com) in the United States (non-P.O. box) were sent a $15 Amazon gift card as a token of appreciation. Requiring a physical mailing address effectively minimized fraudulent attempts from participants outside the United States which is a frequent issue in online incentive-based studies (Pratt-Chapman et al., 2021). Several additional safeguards were implemented to identify and prevent fraudulent entries. The survey link was never distributed directly to individuals upon request to help reduce fraudulent entries; access to the study was only through the advertisements. Qualtrics features were used to restrict multiple submissions and to detect suspicious activity. These included allowing only one submission per IP address, checking time zone consistency, using RECAPTCHA scoring (< 0.5 flagged as bot-like), and utilizing RelevantID’s Duplicate Score (≥ 75 flagged as likely duplicates) and Fraud Score (≥ 30 flagged as potential scammers). Entries falling into these flagged ranges were automatically redirected out of the survey. Manual quality checks were also employed after data collection, including identifying inconsistent or contradictory responses (e.g., inconsistent demographics like age, race, and sex between the screener and main behavioral survey responses), monitoring straight-lining behavior, and flagging unrealistically short completion times. Entries not passing the quality checks were deleted from the dataset. This study was conducted with the approval of the University of New Hampshire Institutional Review Board.

Measures

Image-based sexual abuse (latent profile items)

This was measured using a comprehensive set of self-report items capturing a wide range of image-based sexual abuse experiences (Gewirtz-Meydan et al., 2025). These twelve experiences served as indicators in the latent class analysis to identify distinct patterns of image-based sexual abuse involvement in which someone: (1) took a sexual picture/video of you without your permission, (2) made a sexual picture/video of you without your permission, (3) shared a sexual picture/video of you without your permission, (4) threatened you into giving them a sexual picture/video of yourself, (5) tried to force you into giving them a sexual picture/video of yourself, (6) strongly pressured you into giving them a sexual picture/video of yourself, (7) threatened to share a sexual picture/video of you to get you to give them another sexual picture/video, (8) threatened to share a sexual picture/video of you to get you to have a sexual relationship with them, (9) threatened to share a sexual picture/video of you to get you to pay them money, (10) threatened to share a sexual picture/video of you to get you to do something else, (11) shared a sexual picture/video with an individual five or more years older while under age 18, and (12) made, sent or posted sexual pictures/videos of yourself in exchange for money, drugs, or other valuable items. Each item had a yes/no response option. These items were developed and validated in a series of prior studies (Finkelhor et al., 2023; Gewirtz-Meydan et al., 2025).

Incident characteristics

These items (Turner et al., 2025) included whether there was any commercial element involved (e.g., the image was exchanged for money, goods, or services), whether the incident lasted one month or longer, and whether it occurred on four or more separate occasions. Additional items assessed whether the person responsible intended to cause harm, whether an actual image or video was shared, and whether the media depicted explicit sexual activity. Participants were also asked whether they disclosed the incident to anyone. Relationship with the primary person responsible included: a juvenile dating partner, a juvenile friend, a juvenile acquaintance, a stranger or unknown individual, an adult, or a relative. A separate item assessed whether the incident involved two or more people responsible.

Peer norms related to sexual image sharing

This was measured using two items assessing participants’ retrospective perceptions of their friends’ behaviors prior to age 18 (Finkelhor et al., 2021). These items captured the extent to which sharing sexual images or videos was perceived as common within participants’ peer groups during adolescence. The first item asked respondents to estimate how prevalent it was among their friends to share sexual images or videos with individuals they personally knew, reflecting perceived normative behavior within familiar or trusted social circles. The second item asked participants to report how many of their friends shared sexual images or videos with individuals they did not know. Together, these items were intended to capture perceived peer norms across both lower- and higher-risk forms of sexual image sharing. Responses were provided on a four-point scale ranging from none of them (1) to most or all of them (4).

Childhood victimization

Childhood victimization was assessed using eight items from the Juvenile Victimization Questionnaire (Finkelhor et al., 2005), each capturing a different type of victimization that participants may have experienced before the age of 18. Participants responded using a binary scale (0 = no, 1 = yes). The victimization experiences included peer exclusion, peer name-calling, assault, witnessing domestic violence, caregiver physical abuse, caregiver emotional abuse, witnessing violence, and dating violence. Responses were summed to create a total victimization count score ranging from 0 to 8, with higher scores reflecting greater exposure to multiple different forms of victimization.

Adverse childhood experiences

Childhood adversity was measured using 11 items, capturing participants’ exposure to non-violent but significant stressors during childhood (Turner & Butler, 2003). Participants responded using a binary scale (0 = no, 1 = yes), indicating whether they had experienced each adversity prior to the age of 18. The items covered experiences such as serious accidents or chronic medical conditions, parental divorce, caregiver incarceration, financial instability, family substance use, loss of a close family member, and homelessness or foster care placement. A total adversity count score (0 to 11) was calculated by summing the number of endorsed items, with higher scores reflecting greater exposure to adverse childhood experiences.

Sexual intercourse

This was measured by asking how old they were when they had sexual intercourse for the first time (Finkelhor et al., 2021). Participants filled in a specific age, and this was grouped into the following categories for analysis: never had sexual intercourse, 13 years or younger, 14–15 years, 16–17 years, and 18 years or older.

Puberty

Participants were asked whether they went through puberty: before other kids your age, at the same time as other kids your age, after other kids your age, or have not gone through puberty (Finkelhor et al., 2021). Responses of before other kids your age and not gone through puberty were coded as reflecting early puberty for analyses.

Sexual and gender identity

Gender identity prior to age 18 was measured through a checklist of options including cisgender girl, cisgender boy, transgender girl (assigned male at birth), transgender boy (assigned female at birth), non-binary, and “none of these describe me.” Those selecting “none of these” were prompted with additional categories. For analytic purposes, participants were categorized as either identifying with a gender minority identity (any identity other than cisgender boy/girl) or not. Sexual orientation prior to age 18 was assessed through a question with initial response options of gay, lesbian, bisexual, and heterosexual. Those who indicated that none of these described them were presented with a secondary list. Responses were recoded into a binary variable distinguishing between sexual minority (any identity other than heterosexual) and heterosexual identities (coded as yes/no).

Demographic characteristics

Several characteristics were captured through self-report survey items covering race, ethnicity, geographic location, household income, and educational attainment. Race and ethnicity were assessed separately: participants selected a racial category from predefined options (White, Black, Asian, Native Hawaiian or Pacific Islander, American Indian or Alaskan Native, or another non-White identity) and separately indicated whether they identified as Hispanic or Latinx. For analysis, racial/ethnic minority identity was computed as a binary variable (yes/no) based on whether participants identified with any non-White racial group and/or as Hispanic/Latinx. Household income was measured by asking participants to categorize their income level as below average, about average, or above average. Responses indicating “below average” were coded as reflecting lower socioeconomic status. Education level was assessed using a question about the highest level of schooling completed, ranging from “6th grade or below” to “some college completed.” For the present analysis, lacking a high school diploma was coded as a binary variable (yes/no). These variables were used as indicators of structural disadvantage and social vulnerability in auxiliary analyses.

Data Analysis

The sample comprised 4,205 unique incidents of image-based sexual abuse occurring prior to age 18. Latent class analysis was applied to identify distinct classes of the 12 types of image-based sexual abuse. Latent class analysis is a person-centered statistical method used to identify unobserved subgroups within a population based on patterns of responses across multiple observed variables (Collins & Lanza, 2010). Unlike variable-centered approaches that examine relationships among variables averaged across all individuals, latent class analysis groups individuals who share similar response profiles into distinct classes. Latent class analysis allows for the empirical identification of whether these behaviors cluster into meaningful subtypes that may have different dynamics.

Missing data were minimal (0.13% of observations) across 25 different patterns. To evaluate the missing data mechanism, a non-parametric test (Jamshidian & Jalal, 2010), which assesses whether data are missing completely at random, meaning missingness is unrelated to any observed or unobserved values, was conducted. Hawkins’s test was significant (χ²median = 43.70, p < .001), but the Anderson-Darling test was non-significant (Tmedian = 4.85, p = .236). When Hawkins’s test is significant, but the Anderson-Darling test is not, this suggests the data deviate from multivariate normality but are consistent with missing completely at random. Prior to the primary analyses, distributional assumptions were also evaluated. Anderson-Darling normality tests indicated that all continuous measures significantly deviated from normality (all A > 43.28, all p < .001). The presence of multivariate outliers was assessed using the Minimum Covariance Determinant approach (Leys et al., 2018) implemented via the Routliers R package. Ninety-six observations were identified as multivariate outliers. Rather than excluding these cases, a robust maximum likelihood estimation was conducted which provides standard errors that are robust to non-normality.

Latent class analysis was conducted using Mplus Version 8.10 (Muthén & Muthén, 1998–2023). Because the 4,205 incidents were reported by 2,854 unique participants (i.e., some participants reported multiple incidents), standard errors were adjusted for non-independence using cluster-robust estimation. This approach accounts for the nested structure of the data and prevents artificially deflated standard errors. The 12 binary indicators of image-based sexual abuse were modeled as categorical variables using a logit link function, and parameter estimation was performed using robust maximum likelihood. To ensure identification of the global maximum likelihood solution and avoid convergence to local maxima, 500 random starting values with the 100 best solutions carried forward to final optimization were conducted. All models converged successfully with replicated best log-likelihood values (Muthen & Muthen 2025).

One- through seven-class solutions were estimated and compared using multiple criteria. First, information criteria was examined including the Bayesian Information Criterion, Sample-size Adjusted Bayesian Information Criterion, Consistent Akaike Information Criterion, and Approximate Weight of Evidence, with lower values indicating better fit. These indices were plotted across class solutions (Fig. 1) to identify an “elbow point”, a point of diminishing returns where adding classes yields minimal improvement in fit, similar to scree plot interpretation in factor analysis. Second, the Vuong-Lo-Mendell-Rubin adjusted likelihood ratio test was examined with significant p-values indicating the k-class model fits significantly better. The bootstrapped likelihood ratio test was not conducted because of the cluster-adjusted estimation (Nylund et al., 2007). Third, Bayes Factor indices provided pairwise comparisons between adjacent models, with > 10 indicating strong evidence for the more complex model (Kass & Raftery, 1995), while the correct model probability estimates the posterior probability that each model is the true model. Fourth, entropy (ranging from 0 to 1) summarizes classification accuracy, with values above 0.80 indicating adequate class separation and values above 0.90 indicating excellent separation (Celeux & Soromenho, 1996). Finally, a minimum class size of 5% of the sample (approximately 210 incidents) was set to ensure stable parameter estimates, and classes were evaluated for theoretical interpretability and distinctiveness.

Fig. 1.

Scree plot of information criteria as a function of the number of classes

After identifying the optimal class solution, how the classes differed on auxiliary variables not used in class formation was explored. These “distal outcomes” included incident characteristics (e.g., relationship with person responsible, commercial involvement, intended harm), childhood background (victimization history, adversity), and demographic characteristics (gender identity, sexual orientation, race/ethnicity). The term “distal” indicates these variables are conceptualized as correlates of class membership rather than indicators defining the classes. The Bolck, Croons, and Hagenaars (2004) method was used for these comparisons which is a three-step method that (1) estimates the latent class model using only the class indicators, (2) calculates classification weights that adjust for measurement error in class assignment, and (3) uses these weights when comparing classes on auxiliary variables (Vermunt & Magidson, 2021). This separation ensures that classes are defined solely by the substantive indicators of image-based sexual abuse, while still accounting for classification uncertainty when examining distal outcomes. In the distal analysis, the two developmental covariates were modeled as predictors of class membership via multinomial logistic regression, while all other auxiliary variables were treated as distal outcomes. These developmental measures of age at first sexual experience (categorized as: never, ≤ 13, 14–15, 16–17, or ≥ 18 years) and early puberty onset (yes/no) were specifically included as covariates. These variables were selected because early sexual debut and pubertal timing have been linked to increased risk for sexual victimization (Skoog & Özdemir, 2016). Other developmental measures such as relationship status and sexual health history were not available in the dataset and are acknowledged as a limitation.

Cluster-robust standard errors were maintained throughout. For each distal variable, pairwise Wald tests comparing all class pairs (6 comparisons per variable × 23 variables = 138 tests) was conducted. Given the large number of statistical tests, an adaptive false discovery rate correction with q = 0.10 was applied (Benjamini & Hochberg, 2000). Of the 138 comparisons, 102 remained statistically significant after this correction. All findings reported below survived FDR correction unless otherwise noted. This approach controls the expected proportion of false discoveries while maintaining statistical power. Effect sizes for all pairwise comparisons were conducted. For binary distal outcomes, risk differences (the absolute difference in probability between classes), relative risks (the proportional increase in probability), and odds ratios with 95% confidence intervals are reported. For continuous distal outcomes, Hedges’ g, a bias-corrected standardized mean difference, with 95% confidence intervals, is reported where values of 0.20, 0.50, and 0.80 represent small, medium, and large effects, respectively (Cohen, 1988).

Results

Latent Class Analysis

The results are summarized in Table 2. As is often the case in applied research (Nylund-Gibson & Choi, 2018), the fit indices did not converge on a single optimal solution. The scree plot of information criteria (Fig. 1) revealed an apparent “elbow” at the four-class solution: while Bayesian Information Criterion decreased from 27,014 (1-class) to 23,663 (2-class) to 221,770 (3-class) to 21,007 (4-class), representing a 90.8% cumulative improvement, subsequent reductions were notably smaller (20,731 for 5-class, 20,600 for 6-class, 20,510 for 7-class). Although the correct model probability, Bayes Factor, and Sample-size Adjusted Bayesian Information Criterion marginally favored the seven-class solution, entropy, which reflects how accurately individuals can be classified into their most likely class, dropped considerably when moving beyond four classes, declining from 0.981 (4-class) to 0.913 (5-class) to 0.905 (6-class) to 0.910 (7-class). This decline indicates increasingly blurred boundaries between classes at higher solutions (although above the recommended threshold of 0.80). Additionally, the seven-class model included a class comprising only 1.9% of the sample (n = 81), falling below the pre-specified 5% minimum threshold and raising concerns about replicability and stable parameter estimation.

Table 2.

Fit indices for 1- to 7-Class solutions

	1 Class	2 Classes	3 Classes	4 Classes	5 Classes	6 Classes	7 Classes
Information Criteria
Akaike Information Criterion	26,938	23,504	21,529	20,683	20,325	20,112	19,939
Bayesian Information Criterion	27,014	23,663	21,770	21,007	20,731	20,600	20,510
Sample-size Adjusted Bayesian Information Criterion	26,976	23,583	21,649	20,845	20,528	20,356	20,224
Consistent Akaike Information Criterion	27,026	23,688	21,808	21,058	20,795	20,677	20,600
Approximate Weight of Evidence Criterion	27,150	23,946	22,201	21,585	21,457	21,474	21,531
Likelihood Ratio Tests
Vuong-Lo-Mendell-Rubin Adjusted Likelihood Ratio Test p	—	< 0.001	< 0.001	< 0.001	< 0.001	< 0.001	< 0.001
Model Comparison
Bayes Factor (vs. k − 1)	—	> 10	> 10	> 10	> 10	> 10	> 10
correct model probability	0.00	0.00	0.00	0.00	0.00	0.01	0.99
Classification Quality
Entropy	—	0.983	0.986	0.981	0.913	0.905	0.910
Class Size
Smallest class n	4,205	1,987	948	428	440	321	81
Smallest class %	100.0	47.3	22.5	10.2	10.5	7.6	1.9
Model Fit
Log-likelihood	−13,457	−11,727	−10,726	−10,291	−10,099	−9,979	−9,879
Parameters (k)	12	25	38	51	64	77	90

Note. For information criteria, lower values indicate better fit; bolded values indicate the best-fitting model by that index. Entropy ranges from 0 to 1, with values > 0.80 indicating adequate classification accuracy and > 0.90 indicating excellent accuracy; the 4-class solution is bolded as having the highest entropy among solutions with adequate class sizes. The minimum acceptable class size was set at 5% of the sample (n ≈ 210); values meeting this threshold are bolded. The 4-class solution was selected based on entropy, theoretical interpretability, and adequate class sizes (see text for details)

Balancing statistical fit with theoretical interpretability, the four-class solution was selected. This model demonstrated excellent classification quality: entropy was 0.981, indicating near-perfect class separation, and average posterior probabilities for most likely class membership ranged from 0.93 to 1.00 (Class 1: 0.97; Class 2: 1.00; Class 3: 1.00; Class 4: 1.00), well above the recommended 0.70 threshold. The smallest class comprised 428 incidents (10.2%), exceeding the 5% minimum threshold. It also captures four theoretically distinct victimization pathways: coercive production, non-consensual redistribution, threat-based exploitation, and age-discrepant image sharing. To further evaluate model selection, the theoretical differentiation of the 5- and 6-class solutions was examined from that of the 4-class solution. Although these models showed only marginal improvements in information criteria and maintained entropy above the 0.80 threshold (0.913 and 0.905, respectively), the additional classes did not represent theoretically distinct victimization types. Specifically, the 5-class solution split the “Coerced Production and Distribution” class into two subgroups both comprising exceptionally high rates of being pressured to provide a picture, alongside elevated prevalence of threats to obtain a picture and attempts to elicit one forcibly. While these subgroups differ in the intensity of coercive tactics, they represent variations in degree rather than qualitatively different mechanisms of victimization; both involve people compelling victims to create images. The 6-class solution retained this split and further divided the “Re-Distributors” class based on the prevalence of image-taking and sharing, both involve non-consensual distribution.

The selected 4-class solution profiles are presented in Fig. 2 and include the following groups: “Coerced Production and Distribution” (n = 2,008, 47.7%), “Re-Distributors” (n = 820, 19.5%), “Threat-Based Exploitation” (n = 428, 10.2%), and “Non-Coerced Image Sharing with Older Individuals” (n = 948, 22.5%). Participants in the “Coerced Production and Distribution” class reported exceptionally high rates of being pressured to provide a picture, alongside elevated prevalence of threats to obtain a picture and attempts to elicit one forcibly. The “Re-Distributors” class was characterized by high rates of taking, producing, and sharing pictures without the target’s permission. The third class, “Threat-Based Exploitation,” showed elevated rates of threats to exchange a picture for another image, for sex, for money, or for other gains, as well as of sending a picture or video in return for money or goods. The final class, “Non-Coerced Image Sharing with Older Individuals,” comprised participants who shared a picture or video with someone at least five years older while they themselves were under 18 years of age.

Fig. 2.

The selected 4-class solution. Values in parentheses represent the group’s sample sizes

Differences between Classes in Covariates

Using the “Non-Coerced Image Sharing” class as the comparison group, early puberty emerged as a significantly differentiating factor. Specifically, individuals who experienced early puberty were less likely to be in the “Coerced Production” class (OR = 0.82, 95% CI [0.72, 0.94], p = .005) and more likely to be in the “Re-Distributors” class (OR = 1.21, 95% CI [1.03, 1.42], p = .022). This suggests that early physical maturation may be associated with different pathways into image-based sexual abuse, potentially increasing exposure to peer-based redistribution contexts while being less characteristic of coercion-based victimization. Age at first sexual experience did not significantly predict class membership (all p > .35).

Differences between Classes in Distal Outcomes

Beyond the behaviors that defined each class, how the four groups differed on a range of incident characteristics, relationship with the person responsible, childhood experiences, and demographic factors was examined. Significant differences emerged across all 23 variables examined; however, minority status and gender minority status showed only minimal differentiation, with a single significant pairwise comparison each (both comparing “Coerced Production” to “Non-Coerced Image Sharing”). Means, standard errors, and significance indicators are presented in Table 3, with Supplementary Figs. 1–19 displaying all comparisons. Effect sizes of these differences are presented in Supplementary Table 1.

Table 3.

Class-Specific proportions and means for distal outcomes with significant pairwise comparisons

Variable	Coerced Production (C1)	Re-Distributors (C2)	Threat-Based Exploitation (C3)	Non-Coerced Image Sharing (C4)	Significant Comparisons
Characteristics of Primary Person Responsible
Juvenile dating partner	15.7%	26.6%	11.3%	4.0%	2 > 1,3,4; 1 > 3,4; 3 > 4
Juvenile friend	5.9%	9.9%	5.0%	1.5%	2 > 1,3,4; 1,3 > 4
Juvenile acquaintance	7.0%	7.6%	5.8%	0.5%	1,2,3 > 4; 2 > 3
Stranger/unknown	51.4%	50.1%	72.9%	73.4%	3,4 > 1,2
Adult	54.9%	45.9%	67.2%	74.5%	4 > 3 > 1 > 2
Relative	1.5%	5.5%	3.5%	2.2%	2 > 1,3,4; 3,4 > 1
Two or more people	17.2%	30.2%	23.6%	0.0%	2 > 1,3 > 4; 1,3 > 4
Incident Characteristics
Commercial element	10.5%	18.3%	52.4%	17.9%	3 > 1,2,4; 2,4 > 1
Intended harm	23.3%	49.0%	55.4%	25.0%	3 > 2 > 1,4; 4 > 1
Actual image/video involved	68.6%	100.0%	83.7%	100.0%	2,4 > 3 > 1
Explicit sexual content	46.1%	75.0%	70.7%	78.2%	2,3,4 > 1; 4 > 2 > 3
Duration ≥ 1 month	34.3%	25.0%	27.8%	42.4%	4 > 1 > 2,3
Frequency ≥ 4 times	43.6%	27.7%	31.1%	52.0%	4 > 1 > 2,3
Disclosed to someone	39.8%	47.0%	41.9%	30.1%	2 > 1,3 > 4
Childhood History
Peer norms (M)	2.26	2.52	2.54	2.46	2,3,4 > 1; 2,3 > 4
Childhood victimization (M)	4.02	4.45	4.56	4.03	2,3 > 1,4
Childhood adversity (M)	3.92	4.48	4.64	3.96	2,3 > 1,4
Demographics
Sexual minority	64.5%	64.5%	64.4%	70.8%	4 > 1,2,3
Female at birth	81.0%	76.7%	66.9%	67.1%	1 > 2 > 3,4
Gender minority	25.6%	26.6%	25.8%	28.3%	4 > 1
Race/ethnic minority	39.7%	40.7%	41.1%	42.4%	4 > 1
Low income	42.8%	46.5%	47.9%	46.8%	2,3,4 > 1
No high school diploma	19.4%	21.1%	22.6%	21.8%	3,4 > 1

Note. Values represent percentages for binary variables and means (M) for continuous variables (Peer norms, Childhood victimization, Childhood adversity). All estimates are BCH-weighted to account for classification uncertainty. Significant comparisons reflect pairwise Wald tests surviving Benjamini-Hochberg false discovery rate correction at q < 0.10. Class abbreviations: C1 = Coerced Production and Distribution; C2 = Re-Distributors; C3 = Threat-Based Exploitation; C4 = Non-Coerced Image Sharing with Older Individuals. “>” indicates significantly higher proportion/mean; notation condensed for readability (e.g., “2,3 > 1,4” indicates C2 and C3 are both significantly higher than C1 and C4)

Who was responsible: Peers versus adults

The classes differed markedly in the age and relationship with the primary person responsible. In the “Re-Distributors” class, the person responsible was most often someone the victim knew from their peer group: over one-quarter (26.6%) involved a juvenile dating partner, and additional incidents involved juvenile friends (9.9%) or acquaintances (7.6%). This peer-based pattern was strikingly absent in the “Non-Coerced Image Sharing” class, where juveniles were rare (dating partners: 4.0%; friends: 1.5%; acquaintances: 0.5%). Instead, this class was defined by adults (74.5%) and strangers or unknown individuals (73.4%); rates substantially higher than in “Re-Distributors” (45.9% adults; 50.1% strangers). In practical terms, this means “Re-Distributors” incidents typically occurred within adolescent social networks, whereas “Non-Coerced Image Sharing” involved adults, often strangers met online, engaging minors in what victims may have perceived as consensual relationships despite the inherent power imbalance.

Commercial exploitation and intent to harm

The classes also differed in whether the person responsible appeared motivated by financial gain or a deliberate desire to cause harm. The “Threat-Based Exploitation” class stood apart, with over half of incidents (52.4%) involving some commercial element, such as demanding payment or selling images, compared to only 10.5% in “Coerced Production” and approximately 18% in the other classes. Similarly, the person responsible in “Threat-Based Exploitation” were most often perceived as intending to harm the victim (55.4%), followed by “Re-Distributors” (49.0%), while “Coerced Production” and “Non-Coerced Image Sharing” showed lower rates of perceived malicious intent (23.3% and 25.0%, respectively). These findings suggest that “Threat-Based Exploitation” represents a more predatory, financially motivated form of abuse, consistent with what is commonly termed “sextortion,” whereas “Coerced Production” may more often involve people seeking images for personal use rather than commercial distribution or deliberate harm.

Nature of the abuse: Threats versus actual images

An important distinction emerged in whether incidents involved actual intimate images or remained at the level of threats and demands. In both “Re-Distributors” and “Non-Coerced Image Sharing,” 100% of incidents involved actual images or videos being created or shared. In contrast, roughly one-third of “Coerced Production” incidents (31.4%) and one-sixth of “Threat-Based Exploitation” incidents (16.3%) involved threats, pressure, or demands without actual image distribution. This indicates that the harm in these latter classes sometimes stems from the coercion and fear itself, even when the people responsible do not ultimately obtain or distribute images. When images were involved, explicit sexual content (depicting sexual activity rather than nudity alone) was lowest in “Coerced Production” (46.1%) compared to 70–78% in the other classes.

Duration, frequency, and disclosure

The “Non-Coerced Image Sharing” class showed a troubling pattern: these incidents were more likely to be ongoing (42.4% lasting one month or longer) and repeated (52.0% occurring four or more times) compared to other classes. Yet paradoxically, victims in this class were least likely to have disclosed to anyone (30.1%), compared to disclosure rates of 40–47% in other classes. This pattern suggests that many of these relationships, often between an adult and a minor, persisted over time, with victims potentially not recognizing the interactions as abusive or feeling reluctant to report what they may have perceived as a consensual relationship.

Childhood history

Victims in the “Threat-Based Exploitation” and “Re-Distributors” classes reported more adverse childhood experiences than those in the other two classes. On average, “Threat-Based Exploitation” victims reported 4.56 types of childhood victimization and 4.64 types of childhood adversity measured, compared to approximately 4.0 types of each in “Coerced Production” and “Non-Coerced Image Sharing.” While these differences were statistically significant, the effect sizes were small to medium (Hedges’ g ranging from 0.20 to 0.29), indicating meaningful but modest differences. These findings align with research suggesting that prior victimization may increase vulnerability to certain forms of exploitation, particularly those involving commercial elements or peer networks.

Peer norms

Victims’ perceptions of whether their peers share sexual material differed across classes. Those in the “Coerced Production” class reported the lowest peer norms around sharing (M = 2.26 on a scale where higher scores indicate greater peer approval), significantly lower than all other classes (M = 2.46–2.54). This suggests that coerced production often occurred in contexts where sharing sexual images was not normalized among the victim’s peer group; the person’s demands ran counter to the victim’s social environment rather than being facilitated by it.

Gender and sexual orientation

Victims assigned female at birth were overrepresented across all classes but most prominently in “Coerced Production” (81.0%), followed by “Re-Distributors” (76.7%), with lower rates in “Threat-Based Exploitation” and “Non-Coerced Image Sharing” (both approximately 67%). This suggests that coercive tactics to obtain images disproportionately target girls and young women. Sexual minority youth were overrepresented in the “Non-Coerced Image Sharing” class (70.8%) compared to all other classes (approximately 64–65%), potentially reflecting heightened vulnerability to adults who exploit sexual minority youth seeking connection or identity affirmation. Notably, racial minority, ethnic minority, and gender minority status showed minimal variation across classes, with only the “Coerced Production” class differing significantly from “Non-Coerced Image Sharing” on these variables. Socioeconomic indicators (low income, educational attainment) showed a consistent pattern: the “Coerced Production” class reported lower rates than other classes, but overall variation was modest, indicating that image-based sexual abuse affects youth across demographic backgrounds.

Multiple people responsible

A defining feature of the “Non-Coerced Image Sharing” class was the complete absence of multiple people responsible (0%), in stark contrast to more substantial rates in “Re-Distributors” (30.2%), “Threat-Based Exploitation” (23.6%), and “Coerced Production” (17.2%). This reinforces the characterization of “Non-Coerced Image Sharing” as typically involving a single older individual cultivating a relationship with a minor, whereas the other forms, particularly redistribution, often involve images being shared among multiple people.

Summary

Taken together, these patterns paint distinct portraits of each class. “Threat-Based Exploitation” represents the most severe and predatory form, characterized by commercial motives, deliberate harm, strangers, and victims with elevated histories of childhood adversity, a profile consistent with organized sextortion. “Non-Coerced Image Sharing with Older Individuals” involves adults targeting minors in isolated, ongoing interactions that victims may not recognize as abusive, with notably low disclosure and high rates of sexual minority youth. “Re-Distributors” reflects peer-based abuse within adolescent social networks, often involving dating partners and multiple people responsible who share images among themselves. “Coerced Production and Distribution” shows a distinct pattern: primarily female victims pressured to create images in contexts where such behavior is not peer-normalized, with lower rates of commercial exploitation or overt intent to harm but nonetheless representing a violation of autonomy and trust.

Discussion

Research on image-based sexual abuse has increasingly recognized that treating it as a single, uniform construct limits understanding of the behavioral patterns, relational contexts, and developmental risks that shape how adolescents experience image-based harm. Addressing this gap requires approaches that capture variation within incidents rather than assuming that all forms of image-based sexual abuse are equivalent in severity, context, or vulnerability. The goal of the current study was to identify distinct configurations of image-based sexual abuse behaviors that occurred before age 18 and to examine how these incident-level patterns were associated with key contextual and individual factors. The analysis revealed four conceptually meaningful profiles that differed in behavioral severity, relational dynamics, involvement of adults versus peers, and links to prior adversity and minority status. Together, these findings demonstrate that image-based sexual abuse in adolescence is not a homogeneous phenomenon but a set of diverse and developmentally situated patterns of harm, each reflecting different mechanisms and risk pathways.

The most prevalent group, Coerced Production and Distribution, reflected experiences in which adolescents were pressured to create or share sexual images, often within peer contexts. Although many participants reported an actual sexual image or video, this group had the lowest rates of commercial elements and explicit intent to harm. These dynamics align with developmental features of adolescence, such as susceptibility to peer influence, relational insecurities, and the role of digital communication in social belonging (Fredlund et al., 2018). Incidents in this class also persisted over time, yet disclosure remained low, suggesting that coercion may have been subtle, normalized, or perceived as part of a romantic or social exchange (Gauthier, 2023). This class included the highest proportion of participants assigned female at birth, consistent with gendered power dynamics and dominant sexual scripts that position girls as responsible for sustaining intimacy and accommodating partner expectations (Henry & Beard, 2024; DeKeseredy & Schwartz, 2016). This pattern parallels broader research showing that girls and young women experience image-based sexual abuse within wider contexts of cumulative victimization and gendered vulnerability (Powell et al., 2020). These findings underscore the need for prevention efforts that address relational pressure, consent negotiations, and peer norms.

The second largest group, Non-Coerced Image Sharing with Older Individuals, involved adolescents sharing sexual content with individuals at least five years older. Incidents typically involved one adult or stranger and were not perceived as overtly coercive. However, the consistent presence of actual sexual images or videos, frequent explicit content, and prolonged involvement raise concerns related to power imbalance and grooming. Developmentally, adolescents may lack the cognitive capacity to evaluate consent and risk in age-discrepant relationships (Lippert et al., 2022). This class contained the highest proportion of sexual minority youth, consistent with minority stress theory (Meyer, 2003), which posits that stigma, isolation, and limited access to affirming environments heighten vulnerability. Sexual and gender minority adolescents may also turn to older partners for validation or belonging, a dynamic documented in research on online exploitation among sexual and gender minority youth (Turner, Finkelhor, Mitchell et al., 2024). These findings emphasize the need for identity-affirming support and education tailored to sexual and gender minority youth (Meechan-Rogers et al., 2022).

The Image Re-Distribution class captured non-consensual taking, making, or sharing of sexual images, most often by juvenile dating partners, friends, or acquaintances. This reflects a peer-based form of harm that challenges binary victim–perpetrator models and demonstrates the social rather than exclusively predatory nature of much image-based sexual abuse (Quayle et al., 2018). While the incidents were generally shorter in duration, they involved sexually explicit content and often multiple people responsible. High disclosure in this class may indicate eventual recognition of harm or a willingness to come forward when the person responsible is a peer rather than an anonymous adult. Greater representation of girls in this class aligns with research on gendered dynamics in peer victimization and the social consequences of image redistribution (Maheux et al., 2020). These findings underscore the importance of school-based, peer-focused education on digital consent and accountability.

The smallest class, Threat-Based Exploitation, represented the most severe incidents, characterized by explicit threats, blackmail, and transactional demands. These incidents frequently involved commercial elements, unknown adults, or multiple people responsible and reflected dynamics associated with sextortion and digitally facilitated sexual exploitation (Walsh & Tener, 2022; Wolak et al., 2018). Participants in this class reported the highest levels of prior victimization and adversity, highlighting how cumulative vulnerability intersects with predatory online behavior (Paradiso et al., 2024). The lower proportion of female participants in this class may reflect gendered patterns of online targeting, underreporting among girls, or heightened exposure among boys and gender-diverse youth to financially motivated exploitation. These findings underscore the need for integrated responses involving child protection systems, platform safety mechanisms, and specialized law enforcement.

Taken together, the four classes illustrate how image-based sexual abuse emerges from the interplay of developmental processes, gendered norms, sexual scripts, power imbalances, and minority stress. This incident-level perspective highlights that all adolescents do not experience image-based sexual abuse in the same way and that distinct patterns carry different implications for prevention, education, clinical practice, and policy.

Implications

The identification of four distinct configurations of image-based sexual abuse underscores that prevention and response efforts must be tailored to the specific relational and behavioral dynamics characterizing each class. For youth in the Coerced Production and Distribution class, where experiences often emerge within peer and dating contexts and coercion is subtle or normalized, school- and community-based efforts centering peer norms, consent communication, and digital pressure resistance may be particularly effective. For the Image Re-Distribution class, where peers engage in non-consensual taking or sharing, platform-level interventions such as rapid takedown mechanisms, user reporting pathways, and restorative or accountability-oriented responses within schools and youth-serving environments may offer meaningful avenues for harm reduction. The Threat-Based Exploitation class reflects a pattern aligned with sextortion and digitally facilitated sexual exploitation, highlighting the need for coordinated law enforcement and clinical pathways, anti-sextortion education, and stronger safeguards within digital platforms to detect high-risk patterns. For the Non-Coerced Image Sharing with Older Individuals class, where age-discrepant exchanges may involve grooming or emotional manipulation, prevention efforts may benefit from developmental and sexual and gender minority affirming supports, online grooming detection tools, and increased access to safe spaces and identity-affirming resources for sexual and gender minority youth.

Strengths and Limitations

This study has several strengths. It draws on a large and diverse sample, which adds to the growing body of research in this field. By employing an incident-centered analytic strategy, the study was able to identify distinct subtypes of image-based sexual abuse involvement, revealing important differences in severity, dynamics, and context that are often obscured by binary or aggregate approaches. The use of a comprehensive and behaviorally specific set of image-based sexual abuse experiences enabled a more refined and differentiated mapping of the various forms of image-based sexual abuse encountered by adolescents. In addition, incorporating incident-level variables offered critical insights into the contextual characteristics of each class, thereby enhancing the clinical applicability and policy relevance of the findings.

Nonetheless, several limitations must be acknowledged. The cross-sectional and retrospective nature of the data limits causal inference and is subject to potential recall bias. Reliance on self-reported measures may introduce social desirability or underreporting biases. Although the study encompassed a broad range of image-based sexual abuse experiences, it did not directly assess participants’ motivations, perceptions of consent or legality, or the evolution of their emotional responses over time (e.g., initial enthusiasm followed by shame or regret). Including these dimensions could offer important context for interpreting the findings and developing a better understanding of the subjective experience of image-based sexual abuse involvement. In addition, while the study identified important patterns related to gender and sexual identity, the survey length did not permit an in-depth exploration of the underlying mechanisms shaping these findings, such as gender norms, internalized attitudes about sexuality, or minority stress processes. Finally, the study did not explore legal outcomes or the involvement of formal systems such as law enforcement or child protection, which may be relevant in more severe cases of abuse.

Future Research Directions

Future research would benefit from longitudinal designs that trace developmental trajectories and long-term consequences associated with distinct image-based sexual abuse profiles. Additional work is needed to clarify how adolescents’ sense of agency, peer and cultural norms, and platform-specific digital affordances shape the interpretation and escalation of image-based harm. Integrating qualitative approaches could deepen understanding of meaning making processes, motivations, and emotional responses within these incidents. The role of minority stress, particularly among sexual and gender minority youth, also warrants further investigation given its relevance to vulnerability in age discrepant and exploitative digital interactions.

Across all patterns of image-based harm, intersections with technology platforms, child protection systems, and law enforcement remain essential. Platform architecture and safety tools influence both opportunities for exploitation and the speed of response, while child protection and policing frameworks shape disclosure pathways and outcomes. Future work could be strengthened by incorporating incident diaries or timeline methodologies, linking self-reported incidents with real-world platform takedown data, and testing intersectional mechanisms to clarify how gender, sexuality, race, ethnicity, and socioeconomic status contribute to differing vulnerability across image based sexual abuse profiles.

Conclusion

A central challenge in image-based sexual abuse research has been the tendency to treat image-based harm as a single, uniform phenomenon, obscuring the diverse ways in which adolescents encounter and are affected by these experiences. By examining detailed incident-level data, the present study addressed this gap and revealed four distinct patterns of image-based sexual abuse that differed in behavioral severity, relational context, involvement of adults versus peers, intent to harm, and links to prior adversity and minority status. These findings demonstrate that image-based sexual abuse in adolescence encompasses multiple developmentally situated patterns of harm rather than one homogeneous experience. The profiles identified in this study clarify which configurations of behaviors and relationships tend to cluster together. In doing so, the study contributes to a more precise and developmentally informed understanding of image-based sexual abuse in adolescence, showing that the nature and meaning of image-based harm depend heavily on the context in which it emerges.

Supplementary Information

Below is the link to the electronic supplementary material.

Biographies

Kimberly J. Mitchell

is a Research Professor of Psychology at the University of New Hampshire and a Senior Research Scientist at the Crimes against Children Research Center. Her areas of research include youth exposure to violence generally, with specific emphasis in the areas of technology-facilitated crimes against children, child sex trafficking, exposure to suicide, bias and hate crimes, and drug endangered children.

Ateret Gewirtz-Meydan

is an Associate Professor at the School of Social Work at the University of Haifa and the Director of the Science of Sex Research Lab. Her research explores the complex interplay between trauma, sexuality, and relationships, with a particular focus on how sexual and relational outcomes are shaped by experiences of childhood and adult trauma

Lisa M. Jones

is an Associate Research Professor of Psychology and senior researcher at the Crimes against Children Research Center at the University of New Hampshire. Here areas of research include translational research, assisting youth serving organizations on using research to design programs and evaluating new approaches to help protect youth from violence andvictimization.

David Finkelhor

is Director of Crimes against Children Research Center, Professor of Sociology, and University Professor at the University of New Hampshire. His core areas of interest have been in child maltreatment and family violence, dating back in 1977.

Heather A. Turner

is Professor of Sociology and Senior Research Associate at the Crimes against Children Research Center at the University of New Hampshire. Dr. Turner’s research program has concentrated on social stress processes and mental health, including the effects of violence, victimization, and other forms of adversity on the social and psychological development ofchildren and adolescents

Authors' Contributions

KJM conceived of the study, participated in its design, acquisition of data, acquired funding for the study, and drafted the manuscript. AGM conceived of the study, participated in the study coordination and design, and drafted the manuscript. LMJ drafted the manuscript and acquisition of data. DF contributed to conception and design and critically revising the manuscript for important intellectual content. HAT contributed to conception and design and critically revising the manuscript for important intellectual content. All authors read and approved the final manuscript.

Funding

All phases of this study were supported by National Institute of Justice grant 5PNIJ- 21-GG-02983-MUMU. The research presented in this paper is that of the authors and does not reflect the official policy of the US Department of Justice, Office of Justice Programs.

Data Availability

The datasets generated and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.

Declarations

Informed Consent

All participants provided their informed consent to take part in this study.

Conflicts of Interest

The authors declare no competing interests.

Research involving Human Participants and/or Animals

This research involved human subjects and was conducted with the approval of the University of New Hampshire Institutional Review Board (IRB-FY2002-212), approved 12/14/2021.