research paper of media violence

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here .

Loading metrics

Open Access

Peer-reviewed

Research Article

Reactions to Media Violence: It’s in the Brain of the Beholder

* E-mail: [email protected]

Affiliations Department of Psychiatry, Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, New York, New York, United States of America, Department of Neuroscience, Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, New York, New York, United States of America

Affiliations Department of Psychiatry, Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, New York, New York, United States of America, Laboratory of Neuroimaging, National Institute on Alcohol Abuse and Alcoholism, Bethesda, Maryland, United States of America

Affiliation Department of Psychiatry, Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, New York, New York, United States of America

Affiliation Applied Mathematics and Statistics, SUNY, Stony Brook, New York, United States of America

Affiliation Laboratory of Neuroimaging, National Institute on Alcohol Abuse and Alcoholism, Bethesda, Maryland, United States of America

Affiliation Medical Department, Brookhaven National Laboratory, Upton, New York, United States of America

• Nelly Alia-Klein, 
• Gene-Jack Wang, 
• Rebecca N. Preston-Campbell, 
• Scott J. Moeller, 
• Muhammad A. Parvaz, 
• Wei Zhu, 
• Millard C. Jayne, 
• Chris Wong, 
• Dardo Tomasi, 

• Published: September 10, 2014
• https://doi.org/10.1371/journal.pone.0107260
• Reader Comments

Media portraying violence is part of daily exposures. The extent to which violent media exposure impacts brain and behavior has been debated. Yet there is not enough experimental data to inform this debate. We hypothesize that reaction to violent media is critically dependent on personality/trait differences between viewers, where those with the propensity for physical assault will respond to the media differently than controls. The source of the variability, we further hypothesize, is reflected in autonomic response and brain functioning that differentiate those with aggression tendencies from others. To test this hypothesis we pre-selected a group of aggressive individuals and non-aggressive controls from the normal healthy population; we documented brain, blood-pressure, and behavioral responses during resting baseline and while the groups were watching media violence and emotional media that did not portray violence. Positron Emission Tomography was used with [ 18 F]fluoro-deoxyglucose (FDG) to image brain metabolic activity, a marker of brain function, during rest and during film viewing while blood-pressure and mood ratings were intermittently collected. Results pointed to robust resting baseline differences between groups. Aggressive individuals had lower relative glucose metabolism in the medial orbitofrontal cortex correlating with poor self-control and greater glucose metabolism in other regions of the default-mode network (DMN) where precuneus correlated with negative emotionality. These brain results were similar while watching the violent media, during which aggressive viewers reported being more Inspired and Determined and less Upset and Nervous , and also showed a progressive decline in systolic blood-pressure compared to controls. Furthermore, the blood-pressure and brain activation in orbitofrontal cortex and precuneus were differentially coupled between the groups. These results demonstrate that individual differences in trait aggression strongly couple with brain, behavioral, and autonomic reactivity to media violence which should factor into debates about the impact of media violence on the public.

Citation: Alia-Klein N, Wang G-J, Preston-Campbell RN, Moeller SJ, Parvaz MA, Zhu W, et al. (2014) Reactions to Media Violence: It’s in the Brain of the Beholder. PLoS ONE 9(9): e107260. https://doi.org/10.1371/journal.pone.0107260

Editor: Jonathan A. Coles, Glasgow University, United Kingdom

Received: May 5, 2014; Accepted: August 7, 2014; Published: September 10, 2014

This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.

Data Availability: The authors confirm that, for approved reasons, some access restrictions apply to the data underlying the findings. All relevant brain and behavior data are provided in the supporting information files in excel format.

Funding: Funding was provided by (1) Brookhaven National Laboratory under contract DE-AC02-98CH10886, http://www.bnl.gov/world/ ; (2) National Institute of Mental Health: R01MH090134 (NAK), http://www.nimh.nih.gov/index.shtml ; and (3) National Institute of Mental Health NIDA and NIH K05DA020001 (JSF) and the National Institute of Alcohol Abuse and Alcoholism Intramural Program, http://www.drugabuse.gov/ and http://www.niaaa.nih.gov/ . The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exists.

Introduction

While visual media is replete with images of violence, only a small minority in the population engages in real-life violent behavior. Critically, whether a person will act violently depends on individual trait variations which play a prominent role in how visual media is experienced and processed [1] . Therefore, understanding the neurobiological underpinnings of those with aggressive personality traits above the documented norms, is an important prerequisite to the ongoing debate about media impact on behavior [2] . Enduring trait aggression reflects self-report of retaliatory motivation, with high face validity, where individuals endorse questions regarding the degree of their readiness to hurt others. It is emerging in the literature that aggressive individuals differ from non-aggressive individuals in their baseline, trait-like, neurobiological architecture [3] , suggesting involvement of the brain’s default mode network (DMN) [4] , [5] . The DMN forms a distributed circuit of connected brain systems that shows high and coherent metabolic activity or blood flow during awake yet passive resting states which may represent internal and self-referential processing [4] – [7] . The DMN includes regions typically spanning the posterior cingulate cortex (PCC) and precuneus, lateral inferior parietal gyrus (IPG), medial temporal gyrus (MTG), and ventromedial prefrontal cortex, including the orbitofrontal cortex (OFC) [8] . We hypothesize that at resting baseline, individuals with high trait aggression will exhibit different brain metabolism patterns in the DMN including its ventromedial prefrontal regions, revealing fundamentally different internal preoccupations than those with normative trait aggression.

Stimuli with violent themes can prime, or perhaps facilitate existing trait tendencies [1] , [9] . The General Aggression Model (GAM) [10] outlines the processes by which exposure to violence can cause aggressive behavior through the interplay of enduring traits that drive internal states, coupled with congruent visual stimuli from the environment (e.g., violent media). Therefore, according to GAM, chronic exposure to violent images in the media reinforces existing aggressive traits, thereby preparing the individual towards future violence [11] , [12] . The OFC is specifically involved in elements of aggressive behaviors [13] – [15] through its role in prioritizing emotional cues according to intrinsic salience [16] . Likewise, gray matter deficits in the OFC have been observed in individuals with aggressive and violent behavior [17] . As such, we predict involvement of the OFC since it appears to be specifically involved in response to repeated media violence [18] , [19] . Individual differences in brain and behavior during visual media viewing can be further understood in the context of self-reported affective states and autonomic responses (or lack thereof) [20] , [21] . For example, self-reported distress and systolic blood pressure changes were observed in response to viewing violent media [1] , [21] . Cortical representations of emotion-dependent autonomic response (e.g., blood pressure) have been shown in the OFC, anterior cingulate, and insula in response to viewing violent media in healthy controls [22] .

To test our hypotheses regarding baseline and media viewing differences as a function of trait aggression, we recruited a group of healthy aggressive individuals with a history of assault behavior and a group of non-aggressive healthy controls. Measurements of glucose metabolism with [ 18 F]fluoro-deoxyglucose using positron emission tomography (PET) were obtained at three conditions: at resting baseline, during exposure to violent media, and during exposure to emotional, non-violent media. Blood pressure (BP) and behavioral ratings of state affect were collected intermittently during the movie presentations. We expected that aggressive individuals would have a distinct intrinsic brain activity pattern at resting baseline and during passive viewing of the violent media compared to emotional media.

Ethics Statement

This research protocol was approved by the ethical review board of Stony Brook University and conducted accordingly. All participants provided written informed consent prior to participation. Approval number BNL-381.

Participants

A total of 54 males who responded to advertisement for healthy controls and healthy individuals with history of physical fights, were evaluated for their physical assault tendencies and other inclusion/exclusion criteria. Individuals were initially screened by phone and then seen at Brookhaven National Laboratory by a physician for general exclusion criteria which included current or past psychiatric disorders (e.g., drug abuse or dependence), neurological disease, significant medical illness, current treatment with medication (including over the counter drugs) and head trauma with loss of consciousness >30 minutes. Normal physical examination and laboratory tests were required for entry and pre-scan urine tests ensured the absence of any psychoactive drugs. Individuals were classified as aggressive (Ag) or non-aggressive (Na) depending on their responses on the Physical Aggression subscale of the Buss-Perry Aggression Questionnaire (the physical aggression subscale correlates strongly with peer ratings of aggression demonstrating its concurrent validity) [23] . Of these 54 participants, only individuals who reported physical fights in the last year and scored at or higher than 75 th percentile on the Physical Aggression scale (Ag, n = 12) or those who reported they did not engage in physical fights and scored at 50 th percentile or below on the Physical Aggression scale (Na, n = 13) were chosen for the study (mean age 25.15) [23] . As planned, the participants differed on Physical Aggression (Ag, mean ± standard error 33.5±1.2; Na, 14.5±1.0, p<.0001). They also differed significantly on the other subscales of the Buss-Perry: Verbal Aggression (Ag, 18.8±1.0; Na, 11.6±1.2, p<.0001), Anger (Ag, 23.7±1.5; Na, 9.6±0.6, p<.0001), Hostility (Ag, 23.1±2.0; Na, 11.8±0.9, p<.0001) and the total score (Ag, 99.5±3.8; Na, 47.5±2.7, p<.0001). The two groups did not differ on age, handedness [24] , socio-economic status [25] , estimates of verbal and non-verbal intelligence [26] , [27] , and depression symptoms [28] . Participants were asked about their media habits including the number of hours they watched TV per day on weekdays and on weekends ( Table 1 ). The participants were monetarily compensated for their participation. It is important to note that the staff performing the media exposure, imaging, nursing, and questionnaire completion, were blind to the subject’s assignment as aggressive or non-aggressive.

• PPT PowerPoint slide
• PNG larger image
• TIFF original image

https://doi.org/10.1371/journal.pone.0107260.t001

Personality and Behavioral Measures

In addition to the Buss-Perry Aggression Questionnaire, the Multidimensional Personality Questionnaire (MPQ) [29] , a three-factor structural model of personality was used. As listed in Table 1 , the MPQ models three higher order dimensions of personality: Negative Emotionality (NEM, or Neuroticism ) reflecting tendency toward emotional distress, alienation from others and aggressive behavior; Positive Emotionality (PEM, or extraversion) reflecting enduring positive affect through interpersonal engagement, and Constraint measuring tendencies toward self-control. Several lines of evidence have shown that high levels of NEM as Neuroticism are robustly associated with violence and aggression [30] . Similarly, individuals with elevated scores of NEM tend to experience/report more frequent negative emotions such as anger and anxiety, perceive their environment as hostile/unfair, and often exhibit poor coping mechanisms in a stressful situation [31] . The three NEM sub-scales include Stress Reaction which is linked to low frustration tolerance; Aggression which reflects the tendency to respond with retaliatory response style; and Alienation which is the most predictive primary scale of aggressive behavior. We also assessed attention and inhibitory control using a performance based measure, the Attention Network Task (ANT), that captures reaction-time performance on Alerting (response readiness), Orienting (scanning and selection), and Conflict (inhibitory control) in attention [32] .

Imaging Conditions and State Reactivity

There were three 40-minute imaging conditions: resting baseline, where participants were instructed to rest with eyes open, a video presentation of violent scenes, and a video presentation of emotional scenes not portraying violence. The two videos (violent and emotional) were edited from R-rated movies and documentary films. The violent media presentation contained 20 scenes of violent acts encompassing the depiction of intentional acts of violence from one individual to another (e.g. interpersonal, shootings, street fights). The emotional media presentation contained 19 emotionally intense and action filled but non-violent scenes (e.g. people interacting during a natural disaster, sudden failures during competitive sports). The length of each of the violent or emotional scenes was between 1–4 minutes; these scenes were separated by a black screen that appeared for 30 seconds which signaled the next scene. The level of valence and intensity of each of the violent and emotional scenes was evaluated internally in the laboratory (data not shown) for valence and intensity and sequenced to optimize with the dynamics of FDG uptake (most intense scenes during the first 10 minutes of FDG uptake period). During the movie presentations, state levels of emotional reactivity were assessed using the Positive and Negative Affective Schedule (PANAS) with adjectives of mood states (ranked from 1, slightly to 5, extremely) [33] . The PANAS was completed by the subjects 5 minutes before the media presentations, 10 minutes into the presentations, and at the end of the media presentations. Table 2 shows PANAS adjectives where differences were found between the groups at p<0.05 during the violent as compared to emotional media presentations. Systolic and diastolic BP was monitored with a compression cuff that operated automatically (Propaq Encore) on the participant’s non-dominant arm starting 5 minutes before the imaging and continued throughout the scanning sessions occurring at 5-minute intervals. For Figure 1 systolic BP data was first averaged within each group at each point in the time series during the violent and during the emotional media presentation. Then, the percentage changes in BP (delta) were calculated from the emotional to the violent media within each group [(violent-emotional)/emotional].

Ag (red) individuals show reduction in systolic blood pressure while watching the violent media versus Na (blue) individuals who show progressive increase in systolic blood pressure. Systolic blood pressure measures were averaged for each group at each time point and a percent change and a trend line were calculated (Y-axis). Error bars (joined and filled) reflect the standard deviation of the data that are presented.

https://doi.org/10.1371/journal.pone.0107260.g001

https://doi.org/10.1371/journal.pone.0107260.t002

PET Imaging

The 25 subjects were scanned 3 times with PET-FDG in counterbalanced order on separate days and under 3 conditions: resting baseline, violent scenes, non-violent emotional scenes. The scanning procedure is standardized and was described before [34] . The violent and neutral video presentations started 10 min prior to FDG injection and continued for a total of 40 min. PET imaging was conducted with a Siemens HR+ tomograph (resolution 4.5×4.5×4.5 mm 3 full-width half-maximum, 63 slices) in 3D dynamic acquisition mode. Static emission scan started 35 min after FDG injection and continued for the next 20 min. Arterialized blood was used to measure FDG in plasma. During the uptake period of FDG, subjects were resting with eyes open (no stimulation) or watching a movie (violent or emotional) in a quiet dimly lit room with a nurse by their side to ensure that they did not fall asleep. Metabolic rates were computed using an extension of Sokoloff’s model [35] . The emission data for all the scans were corrected for attenuation and reconstructed using filtered back projection.

Image and Data Analyses

Prior to the analysis, each participant’s PET image was mapped onto the Montreal Neurological Institute (MNI) template and smoothed via a Gaussian kernel with full width half maximum at 16 mm. Normalized metabolic images were analyzed using Statistical Parametric Mapping (SPM) [36] . The normalized images (relative images) were obtained by dividing the signal level of each voxel by the global mean, which was the average signal level of all voxels in the PET image. Analyses were performed in SPM8 with a flexible factor model design with one between-subject factor (Ag and Na groups) and one within-subject factor (baseline, violent, emotional conditions). Main effects of group were tested separately ( Figure 2 ) as well as group x condition interactions. The cluster threshold used was p<0.001, cluster extent >100; given the number of subjects, these parameters were chosen to ensure a minimum of t = 3.00 for each cluster reported. After the SPM results were obtained, cubic regions of interest (ROIs) with 125 voxels were centered at the peak coordinates of relevant activation clusters to compute average metabolic values within these ROIs. Pearson linear correlations were used to assess the association between average ROI measures and BP.

Left panel: Relative glucose metabolism (Y-axis) in Ag (red) and Na (blue) in response to the violent media. On the left of the dotted line are results from Ag>Na contrast and on the right of the dotted line are results from the Ag<Na contrast. Right panel: Glucose metabolism results in response to the emotional media Ag>Na. There were no significant results for Ag<Na. Standard error is presented in the corresponding error bars.

https://doi.org/10.1371/journal.pone.0107260.g002

The behavior and personality indices ( Table 1 ) were analyzed using independent-samples t-tests Bonferroni corrected for multiple comparisons [37] . The changes in BP (delta) were calculated from the emotional to the violent media within each group [(violent-emotional)/emotional] ( Figure 1 ). We tested whether the progressive change in systolic BP was significantly different between the groups with a general linear model (GLM), where time points and group were independent variables while the BP delta was the dependent variable. Two separate linear regression models were fitted within each group and used to test whether the delta in BP changed significantly over time and whether the slopes were significantly different between the groups. Analysis of PANAS responses to the violent and emotional media presentations was done by calculating differences in responses between violent and emotional presentations at 3 time points (pre, 10 min and end) using a GLM ( Table 2 ).

Traits, Inhibitory Control, and Resting Metabolism

As documented in Table 1 , the groups were not different on demographics and media exposure and no differences were found on MPQ personality traits of PEM which includes the subscales Well Being , Social Potency , Social Closeness and Achievement . Not surprisingly, the groups were substantially different on Negative Emotionality and inhibitory control. Individuals from the Ag group, reported more NEM, with high scores on the NEM subscales, Alienation , Aggression and Stress Reaction . The Ag group also demonstrated poor inhibitory control, reporting less self- Control on the MPQ and also showed increased latency to respond specifically in the Conflict condition of the ANT. This performance measure of inhibitory control correlated with self-reported aggression such that more latency as a result of conflict in attention was seen in those with more trait aggression as measured by two different self-report scales (Buss-Perry Physical Aggression scale r = .76, P<0.0001, and MPQ Aggression (r = .66, P<0.001).

The normalized brain metabolic measures were characterized by robust group effects at resting baseline, involving hyperactivity in the DMN and caudate, and dampened OFC metabolism in Ag as compared to Na ( Table 2 ). These resting metabolic measures in precuneus correlated positively across participants with NEM (R = .56, p<.01) and negatively with Control (R = −.46, 0<.05) whereas those in OFC showed the opposite pattern revealing a negative correlation with NEM (R = −.40, p<.05) and positive correlation with Control (R = .48, p<.05).

Glucose Metabolism and Mood Reactivity during Media Viewing

Listed in Table 2 are the main effects of group for each condition separately. These results show similar group differences at resting baseline than for the comparisons during violent media presentation, involving hyperactivity in the DMN and caudate, and dampened OFC metabolism in Ag than Na participants ( Figure 2, left panel ). While viewing the emotional media presentation, the only significant difference between groups was higher glucose metabolism in bilateral lingual gyrus in the Ag group ( Figure 2, right panel ). Group x condition interactions were not significant at our threshold or at a reduced threshold of p<0.005.

As documented in Table 3 , differences emerged between the groups in state reactivity 10 minutes into and at the end of the media presentations. During the violent media presentation as compared to the emotional media presentation, Ag participants when compared with the Na participants reported feeling less Upset ( Figure 3 ) and Nervous and more Inspired and Determined ( Table 3 ). In-line with the mood reactivity data, there were divergent responses between the groups in systolic BP across time. In the Na group, percent BP change progressively increased over time (t 16  = 3.26, p = 0.002) while in the Ag group, systolic BP progressively decreased (t 16  = −4.23, p = 0.0003) in response to the violent media as compared to emotional media ( Figure 1 ). A comparison of the trend lines between the groups shows that the trend lines were significantly opposite (F 1, 32  = 27.60, p<0.0001). Systolic and diastolic BP did not differ between the groups at resting baseline (p>0.05). Diastolic BP was not different between the groups in any of the conditions.

Self-report of being Upset immediately before, during, and at the end (EOV) of the violent media viewing. Standard error is presented in the corresponding error bars.

https://doi.org/10.1371/journal.pone.0107260.g003

https://doi.org/10.1371/journal.pone.0107260.t003

To examine the coupling of BP with glucose metabolism between the groups, we conducted ROI analyses to assess the correlation between regional metabolism during the violent media exposure and changes in systolic BP at time 37 (when most accentuated differences in BP were found between groups, as shown in Figure 1 ). In the Na, increases in BP were positively associated with increased metabolism in the right OFC (x = 22, y = 34, z = −26; r = 0.74; p<0.005) whereas the correlation was negative in (r = −0.56, p<0.005) ( Figure 4 ) in whom decreases in BP were also associated with metabolism in precuneus (R = −.81, p<.001). That is, in Na participants increases in BP were associated with higher metabolism in OFC whereas in Ag participants decreases in BP were associated with increased metabolism in the OFC and precuneus.

On the y-axis is response in the OFC response to violent media compared with emotional media; on the x-axis is systolic BP change between violent media compared with emotional media at time 37 into the media viewing.

https://doi.org/10.1371/journal.pone.0107260.g004

This study documented brain, behavior, and blood-pressure response as a function of trait aggression. Results showed that Ag had heightened traits of NEM and poor inhibitory control compared to Na. These constitutional differences between the groups were apparent in their brain function at resting baseline and during the violent media viewing, where Ag had higher relative metabolism in the retrosplenial DMN, and lower relative metabolism in OFC, gyrus rectus, and posterior cerebellum. While watching the violent compared to emotional media, the Ag viewers reported being more Inspired and Determined, less Upset and Nervous, and showed a progressive decline in systolic blood-pressure compared with controls in whom systolic BP increased. Furthermore, the BP findings were differentially coupled with glucose metabolism between the groups. While viewing violent media, increased blood-pressure in Na was associated with increased metabolism in OFC; in Ag, the observed reduced blood-pressure was associated with increased metabolism in this same region and also in the precuneus.

The Value of Pre-Selection Based on Abnormal Aggression Traits

In pre-selecting participants based on trait aggression this study revealed important baseline differences in brain and behavior compared with controls. Elevated trait aggression is found specifically in individuals with associated disorders, such as antisocial personality disorder and intermittent explosive disorder, as it has straightforward face validity [38] . In addition to elevated trait aggression, Ag also reported more Alienation and Stress Reaction and demonstrated poor inhibitory control, as measured by the ANT conflict [39] , which are part of externalizing behaviors in adults [40] . Studies show that inhibitory control (as documented here using the ANT) play an important role in violent media effects and aggression [41] . Similarly, high levels of NEM as Neuroticism have shown robust connections with violence and aggression [30] . These results on characterizing personality in trait aggression, lend support to the GAM theory, documenting the specificity of trait aggression in its effects on other personality traits [42] and their potential cognitive substrates. Those who endorse few or no aggression items, hence, the Na group, scored at the norms in NEM and PEM, demonstrating that it is normative to endorse very few aggression questions, providing an adequate control for Ag. Importantly, PEM and its subscales were comparable between the groups, perhaps validating a characterization of trait aggression specifically involving NEM while having normative PEM [42] . Supportive of the GAM theory on the role of traits in media viewing, these trait results are important in setting the context of brain metabolism comparisons between the groups.

Characterization of Trait Aggression through Resting Brain Metabolism

The most robust finding in this study is relative hyperactivity of the DMN during resting baseline with relative hypoactivity of the OFC and cerebellum in Ag compared to Na. The documented over-activity in components of the DMN may reflect a neural marker of enduring traits fostering inwardly directed attention to self-referential information stemming from years of social and cognitive learning [43] . Each of the DMN nodes and their network is associated with awareness and conscious information processing [44] , mental imagery, perspective taking, and autobiographical memory retrieval [45] – [47] needed to facilitate an enduring brain activity pattern of behavioral patterns (i.e., trait) [48] , [49] . Several studies mapped DMN regions with trait profiles; for example, Neuroticism (NEM in this study), was associated with lower volumetric measures and lower metabolism of the OFC [50] , [51] in line with our results of hypoactive OFC in Ag. Conducting direct correlations between resting metabolism and NEM as well as with trait Control , we found that the lower resting metabolism in the OFC the higher were NEM and lower Control scores. In contrast the higher resting metabolism in precuneus the higher was NEM and lower Control trait scores. Supporting this finding are recent findings of higher precuneus with reduced conscientiousness and openness [49] both associated with NEM and characteristic of those with high trait aggression.

Other over activated regions at baseline among Ag participants included the sensory motor area and caudate. One could speculate that this increased activity during rest would have a role in compromised responses during a cognitive task. A recent study proposed that striatal dopamine circuits, particularly the caudate, may provide a mechanism for the active suppression of the DMN under conditions that require increased processing of external stimuli (e.g., an attention demanding cognitive task) relative to internal, self-directed processing [52] . This might be related to a recent finding where heightened trait aggression is associated with reduced dopamine in striatum [53] and that striatal dopamine influences the DMN to affect shifting between internal states and cognitive demands [54] .

Brain Metabolism during Violent Media Viewing

The fusiform gyrus was uniquely activated during violent media viewing in Ag, perhaps suggesting increased attention to facial representation of socially relevant cues [55] . Aside from the fusiform activation, while viewing the violent media presentation, the Ag participants compared with the Na showed similar patterns of activation as they had during resting baseline. As such, it appears that DMN regions are active during passive viewing of visual stimuli (e.g., movie) [56] , [57] . We postulate that the violent media condition reflects congruence between the trait and the visual stimuli, such that the stimuli are syntonic (oscillating together) with internal processing, perhaps indicating personal experience with this material. Since resting baseline refers to mind wondering, it could be that participants in the Ag group have had aggressive thoughts that were instigating similar brain networks as during violent media viewing. A study in children during exposure to violent media documented engagement of the posterior cingulate and hippocampi, which was postulated to link memory and emotion to motor activation integrating existing aggression-related thoughts, thereby making them strongly accessible scripts over time [58] . The amygdala is a likely target for cortical arousal in violence viewing. Mathiak and Weber (2006) documented amygdala activation during active game-play in fMRI environment [59] . Their activation pattern showed signal decrease in the amygdala during players’ virtual violent behavior. Our study did not document amygdala responses possibly as a result of the passive nature of the viewing violent media or alternatively, amygdala was not documented because of the temporal resolution differences between PET and fMRI.

Hypoactivity of the Orbitofrontal Cortex

In our study, the Ag participants showed a pattern of reduced OFC activity relative to the Na in the both resting baseline and violent media conditions. The OFC plays a role in externalizing/impulsive behavior, and regulating emotional and social behavior [13] , [60] – [64] . Specific damage to the OFC is associated with impulsive and aggressive behavior [64] , and individuals with such damage show little control over their emotions as well as limited awareness of the moral implications of their actions, and poor decision making [65] . Impulsive aggressive personality disordered patients demonstrate impaired emotion regulation, and exhibit blunted prefrontal, including OFC, metabolism in response to a serotonergic challenge [66] . Deficits in the orbitofrontal lobes as represented by atrophy, lesion, or hypoactive metabolism have been observed across a number of psychiatric populations prone to aggression (e.g., antisocial personality disorder, psychopathy, borderline personality disorder, intermittent explosive disorder) [66] – [68] and suggest that OFC hypo-function may be a common mechanism underlying the pathophysiology of aggressive behavior in general (e.g., both impulsive and premeditated forms). Hypoactivity of the OFC in this study and its correlation with high NEM and low Control scores further support the reliable implication of OFC in the externalizing continuum.

This OFC hypoactivity is consistent with other studies where exposure to violent media is associated with decreased OFC activation. In a study that examined components of the fronto-parietal network in response to aggressive video cues, reduced levels of OFC activation were found [19] . It is possible that OFC hypoactivation reflects desensitization to violence and disrupts the process of moral evaluation of the violent visual stimuli [69] .

Familiarity with violent material could breed desensitization [69] – [71] . It could be that Ag have exhibited reduced inhibition and blunted evaluative categorization of violent stimuli as supported in other studies [71] such that they demonstrate a response (physiological/behavioral/cortical) that is suggestive of an overall desensitization to media violence [72] , [73] .

Under-reactive Emotional and Autonomic Response to Violent Media

There is further evidence in this study supporting the desensitization hypothesis. The Ag group reported being less Nervous and Upset and more Inspired and Determined during the media violence (compared with emotional media) while their systolic BP progressively decreased. In stark contrast, The Na mood and BP responses to the violent media may be associated with a threat evaluation producing sympathetic activation, resulting in BP increase in the Na group. In a study with healthy adolescents, participants viewing violent movie clips experienced increased BP compared to baseline; however, prior exposure to violence was associated with lowered BP [21] . Autonomic under-arousal to threat stimuli has been documented in individuals who exhibit low levels of fear [74] . Angered subjects permitted to commit aggression against the person who had annoyed them often display a drop in systolic blood pressure. They seem to have experienced a physiological relaxation, as if they had satisfied their aggressive urges [75] , [76] .

Indeed, the documented pattern of BP under-reactivity in Ag was associated with hypoactivations in the OFC ( Figure 3 ) and hyperactivation of the precuneus. Behaviorally-evoked changes in cardiovascular (e.g., blood pressure, heart rate) and cardiac-autonomic (e.g., heart rate variability) activity are correlated directly with neural activity within areas of the anterior cingulate cortex, OFC, medial prefrontal cortices, and the amygdala and often in interaction with activity in the insula, and relay regions of the thalamus and brainstem [22] , [77] , [78] . Based on neuroimaging and lesion evidence, a neurobiological model of cardiovascular reactivity shows that physiological and behavioral reactions are instantiated in the corticolimbic brains systems (e.g., medial/prefrontal corticies, insula, and amygdala) [79] . Afferent feedback, appraised by the OFC is integral in generation of somatic markers which trigger an emotional response, subsequently biasing overt behavior [80] . It is important to note here, that these results are relative to responses to emotional media viewing. It appears from our results that non-violent, yet emotionally salient action stimuli increase BP in the Ag individuals, whereas violent stimuli have the opposite effect of decreasing BP in these individuals. The specificity of hypo-response to violent content supports our assertion that the effects of violent media on individuals depend on theme-related traits, in this case aggression, and the brain of the beholder.

There are several limitations in this study that constrain our interpretation power and generalizability. First, there may have been too few participants in the study to ascertain group by condition interactions and to conduct correlations between trait and brain measures. Second, the inclusion of males only in this study was done to control for potentially differential emotional reaction patterns of activation as a function of sex. However, this approach prevents us from making any claims about female response to violent media. Future studies must include females. Third, the experimental design did not include an acute test of aggression following the media condition. Future studies could include such a test to document aggressive responses following violent media as a function of brain response during the violent media. Fourth, there are brain activity results during violent video games finding anterior cingulate involvement [59] , [81] . These results may not be comparable to this study since playing video games requires task-dependent active attention compared to passive attention maintained during movie viewing as we show in our results; therefore more studies are needed to distinguish responses to media sources requiring active attention such as video games from those requiring only passive attention as movie scenes [82] .

Acknowledgments

The authors gratefully acknowledge the contributions of all members of the Brookhaven PET team for advice and assistance in different aspects of this study.

Author Contributions

Conceived and designed the experiments: NAK NDV RZG JSF GJW. Performed the experiments: NAK MCJ CW DT. Analyzed the data: NAK MAP WZ CW. Contributed reagents/materials/analysis tools: WZ CW DT. Contributed to the writing of the manuscript: NAK SJM RPC RZG NDV MAP.

• View Article
• Google Scholar
• 25. Hollingshead AB (1975) Four-factor index of social status. Unpublished paper.
• 26. Wilkinson G (1993) The Wide-Range Achievement Test 3: Administration Manual. Wilminton, DE: Wide-Range Inc.
• 27. Wechsler D (1999) Wechsler abbreviated scale of intelligence: San Antonio, TX: Psychological Corporation.
• 28. Beck AT, Steer RA, Brown GK. (1996) Beck Depression Inventory Manual. 2nd ed. San Antonio, TX: The Psychological Corporation.
• 29. Tellegen A, Waller NG (1997) Exploring personality through test construction: development of the multidimensional personality questionnaire. In: Briggs SR, Cheek JM, editors. Personality measures: development and evaluation. Greenwich: JAI Press.
• 30. Blonigen DM, Krueger RF. (2007) Personaity & Violence: The unifying role of structural models of personality. In: Flannery DJ VA, Waldman ID (Eds.), editor. The Cambridge handbook of violent behavior. New York, NY: Cambridge University Press.
• 37. Stevens J (1992) Applied multivariate statistics for the social sciences. 2nd ed. Lawrence Erlbaum Associates: NewJersey.
• 41. Swing EL, Anderson CA (2014) The role of attention problems and impulsiveness in media violence effects on aggression. Aggress Behav.
• 49. Sampaio A, Soares JM, Coutinho J, Sousa N, Goncalves OF (2013) The Big Five default brain: functional evidence. Brain Struct Funct.
• 69. Funk JB (2005) Children’s exposure to violent video games and desensitization to violence. Child Adolesc Psychiatr Clin N Am 14: 387–404, vii–viii.
• 70. Huesmann LR, Kriwil L (2007) Why observing violence increases the risk of violent behavior in the observer; (Ed.) DF, editor. Cambridge, UK: Cambridge University Press.
• 74. Raine A (1996) Autonomic nervous system factors underlying disinhibited antisocial and violent behavior. Annals New York Academy of Sciences: 46–60.
• 79. Gianaros PJ, Sheu L (2009) A review of neuroimaging studies of stressor-evoked blood pressure reactivity: Emerging evidence for a brain-body pathway to coronary heart disease risk. NeuoImage 47.
• 80. Damasio A (1994) Descartes’ Error: Emotion, Reason, and the Human Brain. New York: HarperCollins.

The role of media violence in violent behavior

Affiliation.

• 1 Institute for Social Research, University of Michigan, Ann Arbor, Michigan 48106-1248, USA. [email protected]
• PMID: 16533123
• DOI: 10.1146/annurev.publhealth.26.021304.144640

Media violence poses a threat to public health inasmuch as it leads to an increase in real-world violence and aggression. Research shows that fictional television and film violence contribute to both a short-term and a long-term increase in aggression and violence in young viewers. Television news violence also contributes to increased violence, principally in the form of imitative suicides and acts of aggression. Video games are clearly capable of producing an increase in aggression and violence in the short term, although no long-term longitudinal studies capable of demonstrating long-term effects have been conducted. The relationship between media violence and real-world violence and aggression is moderated by the nature of the media content and characteristics of and social influences on the individual exposed to that content. Still, the average overall size of the effect is large enough to place it in the category of known threats to public health.

Publication types

• Research Support, U.S. Gov't, P.H.S.
• Aggression / psychology*
• Mass Media*
• Public Health*
• United States
• Video Games
• Violence / psychology*

Disclaimer » Advertising

• HealthyChildren.org

• Previous Article
• Next Article

INTRODUCTION

Recommendations, council on communications and media executive committee, 2009–2010, former executive committee members, contributors, media violence.

• Split-Screen
• Article contents
• Figures & tables
• Supplementary Data
• Peer Review
• CME Quiz Close Quiz
• Open the PDF for in another window
• Get Permissions
• Cite Icon Cite
• Search Site

Council on Communications and Media; Media Violence. Pediatrics November 2009; 124 (5): 1495–1503. 10.1542/peds.2009-2146

Download citation file:

• Ris (Zotero)
• Reference Manager

Exposure to violence in media, including television, movies, music, and video games, represents a significant risk to the health of children and adolescents. Extensive research evidence indicates that media violence can contribute to aggressive behavior, desensitization to violence, nightmares, and fear of being harmed. Pediatricians should assess their patients' level of media exposure and intervene on media-related health risks. Pediatricians and other child health care providers can advocate for a safer media environment for children by encouraging media literacy, more thoughtful and proactive use of media by children and their parents, more responsible portrayal of violence by media producers, and more useful and effective media ratings. Office counseling has been shown to be effective.

Although shootings in schools around the world periodically prompt politicians and the general public to focus their attention on the influence of media violence, the medical community has been concerned with this issue since the 1950s. 1 – 3   The evidence is now clear and convincing: media violence is 1 of the causal factors of real-life violence and aggression. Therefore, pediatricians and parents need to take action. 4  

In 1972, the US Surgeon General issued a special report on the public health effects of media violence that was based on a growing and nearly unanimous body of evidence. 5   Ten years later, the National Institute of Mental Health issued a comprehensive review of the research on media violence and its effects, which outlined concerns about children's psychological health. 6   At a Congressional public health summit in July 2000, the American Academy of Pediatrics (AAP) was joined by the American Medical Association, the American Academy of Child and Adolescent Psychiatry, and the American Psychological Association in issuing an unprecedented joint statement on the impact of entertainment violence on children. 7   Also in 2000, the Federal Bureau of Investigation released a report on shootings in schools that stated that media violence is a risk factor. 8   In 2003, a panel of media-violence experts convened by the National Institute of Mental Health, at the request of the US Surgeon General, published its comprehensive report on the effects of media violence on youth, which revealed media violence to be a significant causal factor in aggression and violence. 9   Most recently, in 2007, the Federal Communications Commission (FCC) released its report on violent television programming and its effects on children and agreed with the Surgeon General that there is “strong evidence” that exposure to media violence can increase aggressive behavior in children. 10   The weight of scientific evidence has been convincing to pediatricians, with more than 98% of pediatricians in 1 study expressing the personal belief that media violence affects children's aggression. 11   Yet, the entertainment industry, the American public, politicians, and parents all have been reluctant to accept these findings and to take action. 4   The debate should be over. 9 , 12  

American children between 8 and 18 years of age spend an average of 6 hours and 21 minutes each day using entertainment media (television, commercial or self-recorded video, movies, video games, print, radio, recorded music, computers, and the Internet). 13   Children between 0 and 6 years of age spend an average of almost 2 hours each day using screen media (television, movies, computers). 14 , 15   Televisions are also commonly present in bedrooms, with 19% of infants, 29% of 2- to 3-year-olds, 43% of 4- to 6-year-olds, and 68% of children 8 years and older having a television in their bedrooms. 13 , 15 , 16   The effects of having a television in a child's bedroom are only beginning to be studied, but the early indications are alarming. Children with a television in their bedroom increase their television-viewing time by approximately 1 hour per day. 13 , 17   Their risk of obesity increases 31%, 17   and their risk of smoking doubles. 18   In addition, if children have a television in their bedroom, parents are less able to monitor what is seen; parents are less able to have consistent rules for children's media use; children participate in fewer alternative activities such as reading, hobbies, and games; and children perform more poorly in school. 19 , 20  

A large proportion of children's media exposure includes acts of violence that are witnessed or “virtually perpetrated” (in the form of video games) by young people. By 18 years of age, the average young person will have viewed an estimated 200000 acts of violence on television alone. 21   The National Television Violence study evaluated almost 10000 hours of broadcast programming from 1995 through 1997 and revealed that 61% of the programming portrayed interpersonal violence, much of it in an entertaining or glamorized manner. 22   The highest proportion of violence was found in children's shows. Of all animated feature films produced in the United States between 1937 and 1999, 100% portrayed violence, and the amount of violence with intent to injure has increased through the years. 23   In a study of the top-rated PG-13 films of 1999–2000, 90% contained violence, half of it of lethal magnitude. 24   An estimated 12% of 22 million 10- to 14-year-olds saw 40 of the most violent movies in 2003. 25   More than 80% of the violence portrayed in contemporary music videos is perpetrated by attractive protagonists against a disproportionate number of women and blacks. 26   Similarly, teenagers' music has become more violent, especially rap music. 3 , 27 , 28   And, as teenagers increasingly use the Internet, they are exposed to violence there as well; a survey of more than 1500 10- to 15-year-olds revealed that 38% had been exposed to violent scenes on the Internet. 29   Video games also are filled with violence. A recent analysis of the Entertainment Software Ratings Board (ESRB) ratings of video games revealed that more than half of all games are rated as containing violence, including more than 90% of games rated as appropriate for children 10 years or older (E10+ and T ratings). 30  

Prolonged exposure to such media portrayals results in increased acceptance of violence as an appropriate means of solving problems and achieving one's goals. 2 , 3 , 9   American media, in particular, tend to portray heroes using violence as a justified means of resolving conflict and prevailing over others. 24 , 31   Television, movies, and music videos normalize carrying and using weapons and glamorize them as a source of personal power. 22 , 32   Children in grades 4 through 8 preferentially choose video games that award points for violence against others, and 7 of 10 children in grades 4 through 12 report playing M-rated (mature) games, with 78% of boys reporting owning M-rated games. 33 , 34   Of 33 popular games, 21% feature violence against women. 35   Because children have high levels of exposure, media have greater access and time to shape young people's attitudes and actions than do parents or teachers, replacing them as educators, role models, and the primary sources of information about the world and how one behaves in it. 36  

After the tragic shootings at Columbine High School in 1999, the Federal Trade Commission (FTC) investigated whether the motion picture, music, and video-game industries specifically advertised and marketed violent material to children and adolescents. Working with industry-provided documents, the FTC determined that, despite the fact that their own rating systems found the material appropriate only for adults, these industries practiced “pervasive and aggressive marketing of violent movies, music, and electronic games to children,” such as promoting R-rated movies to Campfire girls. 37  

Studies have revealed that children and adolescents can and do easily access violent media that are deemed inappropriate for them by the various rating systems and parents. 13 , 38 , 39   In a study of PG-, PG-13-, and R-rated films, the rating did not even predict the frequency of violence in the various films. 39   Many parents find the entertainment industry's media-rating systems difficult to use. 40   The movie ratings are used by approximately three quarters of parents, but only about half of parents say they have ever used the video-game ratings, the television ratings, or the music advisories to guide their choices. 41   Many parents find the ratings unreliably low, with an objective parental evaluation revealing as many as 50% of television shows rated TV-14 to be inappropriate for their teenagers. 42   At the same time, most parents do not even know that their television is equipped with a V-chip (“V” for “viewer” control), and only 20% of parents actually use it. 40   Video games with higher ratings may actually attract more young children (the “forbidden-fruit” hypothesis). 43   The various media ratings are determined by industry-sponsored ratings boards or the artists and producers themselves. They are age based, which assumes that all parents agree with the raters about what is appropriate content for children of specific ages. Furthermore, different rating systems for each medium (television, movies, music, and video games) make the ratings confusing, because they have little similarity or relationship to one another. The AAP offers an informational brochure that pediatricians can offer to parents and children to help them use the various rating systems to guide better media choices. 44  

Research has associated exposure to media violence with a variety of physical and mental health problems for children and adolescents, including aggressive and violent behavior, bullying, desensitization to violence, fear, depression, nightmares, and sleep disturbances. Consistent and significant associations between media exposure and increases in aggression and violence have been found in American and cross-cultural studies; in field experiments, laboratory experiments, cross-sectional studies, and longitudinal studies; and with children, adolescents, and young adults. 9 , 45 – 47   The new Center on Media and Child Health at Harvard lists more than 2000 research reports. 48   The strength of the association between media violence and aggressive behavior found in meta-analyses 9 , 49   is greater than the association between calcium intake and bone mass, lead ingestion and lower IQ, and condom nonuse and sexually acquired HIV infection, and is nearly as strong as the association between cigarette smoking and lung cancer 50   —associations that clinicians accept and on which preventive medicine is based without question.

Children are influenced by media—they learn by observing, imitating, and adopting behaviors. 51   Several different psychological and physiologic processes underlie media-violence effects on aggressive attitudes, beliefs, behaviors, and emotions, and these processes are well understood. 2 , 3 , 9   Furthermore, because children younger than 8 years cannot discriminate between fantasy and reality, they may be especially vulnerable to some of these learning processes and may, thereby, be more influenced by media violence. 52 , 53   However, even older adolescents and young adults are adversely affected by consumption of media violence, demonstrating that the ability to discriminate between fantasy and reality does not inoculate one from the effects of media violence. 54 , 55  

Some research has indicated that the context in which media violence is portrayed and consumed can make the difference between learning about violence and learning to be violent. 3   Plays such as Macbeth and films such as Saving Private Ryan treat violence as what it is—a human behavior that causes suffering, loss, and sadness to victims and perpetrators. In this context, with helpful adult guidance on the real costs and consequences of violence, appropriately mature adolescent viewers can learn the danger and harm of violence by vicariously experiencing its outcomes. Unfortunately, most entertainment violence is used for immediate visceral thrills without portraying any human cost and is consumed by adolescents or children without adult guidance or discussion. Furthermore, even if realistic portrayals of harmful consequences of violence reduce the typical immediate short-term aggression-enhancement effect, there still exists the potential long-term harm of emotional desensitization to violent images. 9 , 47 , 54   Other studies have shown that the more realistically violence is portrayed, the greater the likelihood that it will be tolerated and learned. 3 , 56   Titillating violence in sexual contexts and comic violence are particularly dangerous, because they associate positive feelings with hurting others. 57 , 58   One study of nearly 32000 teenagers in 8 different countries, for example, revealed that heavy television-viewing was associated with bullying. 59  

In addition to modeling violent behavior, entertainment media inflate the prevalence of violence in the world, cultivating in viewers the “mean-world” syndrome, a perception of the world as a dangerous place. 60 – 62   Fear of being the victim of violence is a strong motivation for some young people to carry a weapon, to be more aggressive, and to “get them before they get me.” 61   For some children, exposure to media violence can lead to anxiety, depression, posttraumatic stress disorder, 56 , 63   sleep disturbances and nightmares, 56 , 64   and/or social isolation. 65   Some have defended media violence as an outlet for vicariously releasing hostility in the safety of virtual reality. However, research that has tested this “catharsis hypothesis” revealed that after experiencing media violence, children and young adults behave more aggressively, not less. 66 – 68   Numerous studies have shown that an insidious and potent effect of media violence is to desensitize all of us to real-life violence. 69 – 72  

Interactive media, such as video games and the Internet, are relatively new media forms with even greater potential for positive and negative effects on children's physical and mental health. Exposure online to violent scenes has been associated with increased aggressive behavior. 29   Studies of these rapidly growing and ever-more-sophisticated types of media have indicated that the effects of child-initiated virtual violence may be even more profound than those of passive media such as television. In many games, the child or teenager is “embedded” in the game and uses a “joystick” (handheld controller) that enhances both the experience and the aggressive feelings. Three recent studies directly compared the effects of interactive (video games) and passive (television and movies) media violence on aggression and violence; in all 3 cases, the new interactive-media-violence effect was larger. 54   Correlational and experimental studies have revealed that violent video games lead to increases in aggressive behavior and aggressive thinking and decreases in prosocial behavior. 62 , 73 – 76   Recent longitudinal studies designed to isolate long-term violent video-game effects on American and Japanese school-aged children and adolescents have revealed that in as little as 3 months, high exposure to violent video games increased physical aggression. 54 , 77   Other recent longitudinal studies in Germany and Finland have revealed similar effects across 2 years. 78 , 79   On the other hand, there is also good evidence that prosocial video games can increase prosocial attitudes and behavior. 80  

Children learn best by observing a behavior and then trying it. The consequences of their behavioral attempts influence whether they repeat the behavior. All violent media can teach specific violent behaviors, the circumstances when such behaviors seem appropriate and useful, and attitudes and beliefs about such behavior. In this way, behavioral scripts are learned and stored in memory. 47   Video games provide an ideal environment in which to learn violence and use many of the strategies that are most effective for learning. 81   They place the player in the role of the aggressor and reward him or her for successful violent behavior. Rather than merely observing only part of a violent interaction (such as occurs in television violence), video games allow the player to rehearse an entire behavioral script, from provocation, to choosing to respond violently, to resolution of the conflict. 54 , 62 , 82   Children and adolescents want to play them repeatedly and for long periods of time to improve their scores and advance to higher levels. Repetition increases their effect. In addition, some youth demonstrate pathologic patterns of video-game play, similar to addictions, in which game play disrupts healthy functioning. 81 , 83   Advances in the measurement of brain function have been applied to the study of media violence. Several studies have linked media-violence exposure to decreases in prefrontal cortex activity associated with executive control over impulsive behavior. 84  

Interpersonal violence, for victims and perpetrators, is now a more prevalent health risk than infectious disease, cancer, or congenital disorders for children, adolescents, and young adults. Homicide, suicide, and trauma are leading causes of mortality in the pediatric population. In 2004, unintentional injuries claimed 17741 lives, homicides claimed 5195 lives, and suicide claimed 4506 lives among 5- to 24-year-olds. 85   Of all deaths by homicide or suicide, fully half were gun related, making gun violence a leading killer of children and adolescents. 86   For young black males, homicide is the leading cause of death, accounting for nearly 45% of all deaths. The homicide rate for black males is 2.7 to 15.8 times higher than for other racial/ethnic groups at the same age. 87   Although violent crime rates have decreased by more than 50% between 1994 and 2004 for young people 12 to 24 years of age, they remain higher at this age than at any other age. 87   Furthermore, the proportion of youth admitting to having committed various violent acts within the previous 12 months has remained steady or even increased somewhat in recent years. 88   In the 2007 National Youth Risk Behavior Survey, 18% of students in the 9th through 12th grades reported carrying a weapon to school in the month preceding the survey, and more than one third had been in a physical fight in the year before the survey. 85   An estimated 30% of 6th- through 10th-graders report either bullying other students or being targets of bullies. 89   A recent large study of New York City students found that nearly 10% of girls and more than 5% of boys reported a lifetime history of being sexually assaulted, and 10% of both boys and girls reported experiencing dating violence in the previous year. 90   Although exposure to media violence is not the sole factor contributing to aggression, antisocial attitudes, and violence among children and adolescents, it is an important health risk factor on which we, as pediatricians and members of a compassionate society, can intervene. Some research has suggested that interventions of the types discussed below can reduce media-violence consumption and its effects on children and adolescents. 2 , 3 , 54 , 91 , 92  

Pediatricians must become cognizant of the pervasive influence that the wide and expanding variety of entertainment media have on the physical and mental health of children and adolescents. 4 , 93   Residency training conferences, grand rounds, and continuing medical education courses are all important venues that should be used for teaching pediatricians about the effects of media on children and adolescents.

Pediatricians should ask at least 2 media-related questions at each well-child visit: (1) How much entertainment media per day is the child or teenager watching? (2) Is there a television set or Internet connection in the child's or teenager's bedroom? 4 , 93   For all children, healthy alternatives such as sports, interactive play, and reading should be suggested. 94   When heavy media use by a child is identified, pediatricians should evaluate the child for aggressive behaviors, fears, or sleep disturbances and intervene appropriately. 95 , 96  

Pediatricians should encourage parents to adhere to the AAP media recommendations 11 , 95   :

Remove televisions, Internet connections, and video games from children's bedrooms.

Make thoughtful media choices and coview them with children. Coviewing should include discussing the inappropriateness of the violent solutions offered in the specific television show, movie, or video game and helping the child to generate nonviolent alternatives. Parents tend to limit sexual content more than violent content, 38   yet research has indicated that the latter is potentially more unhealthy. 2 , 3  

Limit screen time (including television, videos, computer and video games) to 1 to 2 hours per day, using the V-chip, and avoiding violent video games (defined as games that include intentional harm to other game characters, including cartoonish or unrealistic violence as well as realistic or gory violence). Counseling about limiting screen time has been shown to be effective in office settings. 97   For example, just a minute or two of office counseling about media violence and guns could lead to less violence exposure for more than 800000 children per year. 97   Parents also need to be reminded that they are important role models in terms of their own media use.

Avoid screen media for infants or toddlers younger than 2 years. 98   There have been no studies to indicate that screen time contributes positively to infant development, 99 , 100   and there are now 7 studies that have documented possible language delays among children younger than 2 years who are exposed to television or videos. 100 – 108  

Pediatricians and other child health professionals should ensure that only nonviolent media choices be provided to patients in outpatient waiting rooms and inpatient settings.

On a local level, pediatricians should encourage parents, schools, and communities to educate children to be media literate as a means of protecting them against deleterious health effects of media exposure. 93 , 109 , 110   Research has demonstrated that media education and thoughtful media use can reduce violent behavior in children. 9 , 92 , 111  

On state and national levels, pediatricians should work with the AAP and their AAP chapters and districts to collaborate with other health care organizations, educators, government, and research-funding sources to keep media violence on the public health agenda. Media violence is often characterized in the public domain as a values issue rather than what it truly is: a public health issue and an environmental issue. A recent revealed found that two thirds of parents actually favor increased governmental oversight of the media when children and teenagers are concerned. 40  

Pediatricians should advocate for more child-positive media. Pediatricians should support and collaborate with media producers, applying our expertise in child health and development toward creating child-friendly and truthful media. The AAP makes the following recommendations to the entertainment industry:

Avoid the glamorization of weapon-carrying and the normalization of violence as an acceptable means of resolving conflict.

Eliminate the use of violence in a comic or sexual context or in any other situation in which the violence is amusing, titillating, or trivialized.

Eliminate gratuitous portrayals of interpersonal violence and hateful, racist, misogynistic, or homophobic language or situations unless explicitly portraying how destructive such words and actions can be. Even so, violence does not belong in media developed for very young children.

If violence is used, it should be used thoughtfully as serious drama, always showing the pain and loss suffered by victims and perpetrators.

Music lyrics should be made easily available to parents so they can be read before deciding whether to purchase the recording.

Video games should not use human or other living targets or award points for killing, because this teaches children to associate pleasure and success with their ability to cause pain and suffering to others.

Play of violent video games should be restricted to age-limited areas of gaming arcades; the distribution of videos and video games and the exhibition of movies should be limited to appropriate age groups.

Pediatricians should advocate for a simplified, universal, content-based media-rating system to help parents guide their children to make healthy media choices. Content should be rated on the basis of research about what types of media depictions are likely to be harmful to children, rather than simply on what adults find offensive. Just as it is important that parents know the ingredients in food they may feed to their children, they should be fully informed about the content of the media their children may use. 4 , 30 , 112 , 113  

Gilbert L. Fuld, MD, Chairperson

Deborah Ann Mulligan, MD, Chair-elect

Tanya Remer Altmann, MD

Ari Brown, MD

Dimitri A. Christakis, MD

Kathleen Clarke-Pearson, MD

Benard P. Dreyer, MD

Holly Lee Falik, MD

Kathleen G. Nelson, MD

Gwenn S. O'Keeffe, MD

Lead author

Regina M. Milteer, MD

Donald L. Shifrin, MD

Michael Brody, MD

American Academy of Child and Adolescent Psychiatry

Brian Wilcox, PhD

American Psychological Association

Craig A. Anderson

Douglas A. Gentile

Gina Ley Steiner

Veronica Laude Noland

This document is copyrighted and is property of the American Academy of Pediatrics and its Board of Directors. All authors have filed conflict of interest statements with the American Academy of Pediatrics. Any conflicts have been resolved through a process approved by the Board of Directors. The American Academy of Pediatrics has neither solicited nor accepted any commercial involvement in the development of the content of this publication.

Advertising Disclaimer »

Citing articles via

Email alerts.

Affiliations

• Editorial Board
• Editorial Policies
• Journal Blogs
• Pediatrics On Call
• Online ISSN 1098-4275
• Print ISSN 0031-4005
• Pediatrics Open Science
• Hospital Pediatrics
• Pediatrics in Review
• AAP Grand Rounds
• Latest News
• Pediatric Care Online
• Red Book Online
• Pediatric Patient Education
• AAP Toolkits
• AAP Pediatric Coding Newsletter

First 1,000 Days Knowledge Center

Institutions/librarians, group practices, licensing/permissions, integrations, advertising.

• Privacy Statement | Accessibility Statement | Terms of Use | Support Center | Contact Us
• © Copyright American Academy of Pediatrics

This Feature Is Available To Subscribers Only

Sign In or Create an Account

This paper is in the following e-collection/theme issue:

Published on 27.3.2024 in Vol 26 (2024)

This is a member publication of University of Toronto

Bridging and Bonding Social Capital by Analyzing the Demographics, User Activities, and Social Network Dynamics of Sexual Assault Centers on Twitter: Mixed Methods Study

Authors of this article:

Original Paper

• Jia Xue 1, 2 , PhD   ; 
• Qiaoru Zhang 3 * , BA   ; 
• Yun Zhang 3 * , MI   ; 
• Hong Shi 3 * , MI   ; 
• Chengda Zheng 3 * , MI   ; 
• Jingchuan Fan 1 , MSW, MPH   ; 
• Linxiao Zhang 1 , MA, MSc   ; 
• Chen Chen 3 , PhD   ; 
• Luye Li 4 , PhD   ; 
• Micheal L Shier 1 , PhD  

1 Factor Inwentash Faculty of Social Work, University of Toronto, Toronto, ON, Canada

2 Faculty of Information, University of Toronto, Toronto, ON, Canada

3 Artificial Intelligence for Justice Lab, University of Toronto, Toronto, ON, Canada

4 Department of Sociology, Anthropology, Social Work, and Criminal Justice, Seton Hall University, South Orange, NJ, United States

*these authors contributed equally

Corresponding Author:

Jia Xue, PhD

Factor Inwentash Faculty of Social Work

University of Toronto

246 Bloor Street West

Toronto, ON, M5S 1V4

Phone: 1 416 946 5429

Email: [email protected]

Background: Social media platforms have gained popularity as communication tools for organizations to engage with clients and the public, disseminate information, and raise awareness about social issues. From a social capital perspective, relationship building is seen as an investment, involving a complex interplay of tangible and intangible resources. Social media–based social capital signifies the diverse social networks that organizations can foster through their engagement on social media platforms. Literature underscores the great significance of further investigation into the scope and nature of social media use, particularly within sectors dedicated to service delivery, such as sexual assault organizations.

Objective: This study aims to fill a research gap by investigating the use of Twitter by sexual assault support agencies in Canada. It seeks to understand the demographics, user activities, and social network structure within these organizations on Twitter, focusing on building social capital. The research questions explore the demographic profile, geographic distribution, and Twitter activity of these organizations as well as the social network dynamics of bridging and bonding social capital.

Methods: This study used purposive sampling to investigate sexual assault centers in Canada with active Twitter accounts, resulting in the identification of 124 centers. The Twitter handles were collected, yielding 113 unique handles, and their corresponding Twitter IDs were obtained and validated. A total of 294,350 tweets were collected from these centers, covering >93.54% of their Twitter activity. Preprocessing was conducted to prepare the data, and descriptive analysis was used to determine the center demographics and age. Furthermore, geolocation mapping was performed to visualize the center locations. Social network analysis was used to explore the intricate relationships within the network of sexual assault center Twitter accounts, using various metrics to assess the network structure and connectivity dynamics.

Results: The results highlight the substantial presence of sexual assault organizations on Twitter, particularly in provinces such as Ontario, British Columbia, and Quebec, underscoring the importance of tailored engagement strategies considering regional disparities. The analysis of Twitter account creation years shows a peak in 2012, followed by a decline in new account creations in subsequent years. The monthly tweet activity shows November as the most active month, whereas July had the lowest activity. The study also reveals variations in Twitter activity, account creation patterns, and social network dynamics, identifying influential social queens and marginalized entities within the network.

Conclusions: This study presents a comprehensive landscape of the demographics and activities of sexual assault centers in Canada on Twitter. This study suggests that future research should explore the long-term consequences of social media use and examine stakeholder perceptions, providing valuable insights to improve communication practices within the nonprofit human services sector and further the missions of these organizations.

Introduction

Use of social media by nonprofit organizations.

Social media platforms, including Twitter (subsequently rebranded as X [X Corp]), have gained popularity among nonprofit advocacy organizations as essential tools for communication and public engagement [ 1 , 2 ]. Nonprofit organizations are increasingly recognizing the strategic value of social media in fostering public engagement, securing donations, disseminating information, recruiting volunteers, and raising awareness about social issues [ 3 - 8 ]. Today, most large and mid-sized nonprofit organizations actively maintain at least 1 social media account, underscoring the extensive use of social media within the nonprofit realm [ 9 ].

Twitter, for instance, offers nonprofit organizations a platform to create profiles, establish networks, and engage socially through features such as tweeting, sharing multimedia content, replying, and retweeting [ 10 ]. Recognized as a cost-effective means of consistently reaching a broader audience [ 11 , 12 ], Twitter proves especially valuable for nonprofit organizations, often facing limited financial resources and dedicated communication staff [ 12 ], including sexual assault centers, to actively engage with key stakeholders and spark meaningful conversations [ 13 , 14 ]. Moreover, engaging in dialogues with other Twitter users forms a central aspect of communication for these organizations, facilitating increased supporter involvement, knowledge dissemination, and the creation of supportive communities [ 15 , 16 ]. Nonprofit organizations have successfully captured their followers’ attention by regularly tweeting, responding to specific tweets, and retweeting other users’ content [ 2 ]. This social media engagement can be harnessed by nonprofit organizations to share educational information and advocate for social causes [ 17 ].

Bridging and Bonding Social Capital

Social capital plays a critical role in understanding the effectiveness of nonprofit organizations, as it is embedded within their networks, enabling them to enhance their adaptive capabilities by consolidating shared interests and harnessing diverse resources [ 18 , 19 ]. Within the context of social capital, Putnam [ 20 ] distinguishes between 2 fundamental forms: bridging and bonding social capital. Bridging social capital encompasses the distant and weak connections between individuals from diverse backgrounds, facilitating information flow. This often manifests as nonmutual following relationships between organizations and a diverse public. In contrast, bonding social capital revolves around preexisting and robust ties that reinforce homogeneity among groups, fostering emotional and social support. An illustrative example of this is the mutual following relationships observed between similar organizations [ 21 , 22 ].

From a social capital perspective, relationship building is seen as an investment. It involves a complex interplay of tangible and intangible resources, both embedded within existing relationships and generated through the act of forging new ones [ 23 , 24 ]. The success of nonprofit organizations relies significantly on their capacity to establish high-quality relationships with key stakeholders, including donors, clients, grant makers, seekers, and the broader public [ 25 , 26 ]. The social capital of nonprofits comprises the wealth of resources intricately embedded within these strategic alliances and stakeholder relationships [ 27 ]. Xu and Saxton [ 26 ] propose that the effective acquisition of social capital, at an elevated level, relies on the scope and quality of stakeholder connections. Their study introduces and demonstrates the significance of 2 primary stakeholder engagement strategies: content-based and connection-based strategies. This study underscores that the attainment of social capital is less about the number of stakeholder engagements and more about the breadth of those engagements. This breadth includes diverse stakeholder connections.

Social Media–Based Social Capital

Social media–based social capital signifies the diverse social networks that organizations can foster through their engagement on social media platforms [ 26 ]. The potential of social media to nurture and sustain web-based–offline social capital is substantial, although its effectiveness varies across platforms and strategies [ 28 ]. Platforms such as Facebook (Meta Platforms), Twitter, and Instagram (Meta Platforms) offer distinctive usability features that influence the dynamics of bridging and bonding social capital among their users. An important study conducted by Phua et al [ 22 ] examined the impact of 4 major social networking sites (Facebook, Twitter, Instagram, and Snapchat [Snap Inc]) on the development of web-based bridging and bonding social capital among 297 users. Their findings indicate that Twitter users exhibit the highest levels of bridging social capital, followed by Instagram, Facebook, and Snapchat. Conversely, when it comes to bonding social capital, Snapchat users demonstrate the highest levels, followed by Facebook, Instagram, and Twitter. Furthermore, research suggests a direct correlation between the number of followers and the development of bonding social capital [ 21 ]. Another study by Xu and Saxton [ 26 ], focusing on 198 community foundations, reinforces the importance of social media engagement strategies tailored to multiple intersectoral stakeholders and diverse communication patterns, which substantially contribute to the development of social media–based social capital.

In the context of nonprofit organizations, studies by Henry and Bosman [ 29 ] and Lee and Shon [ 30 ] underscore the positive impact of web-based social capital generated through social networking sites on charitable outcomes. These studies reveal that the quantity of Twitter followers is linked positively with personal contributions, although not necessarily with full-time equivalent volunteers [ 30 ]. Moreover, Xu and Saxton [ 26 ], drawing from Twitter data consisting of 198 community foundations, highlight the pivotal role of stakeholder engagement diversity over connection quantities. They emphasize the significance of using multiple communicative cues, such as message elements, and targeting intersectoral and interregional stakeholders in the successful acquisition of social capital through social networking sites. Leveraging social media platforms offers numerous advantages to nonprofit organizations, including the engagement of a donor base within the general population [ 1 , 31 ], the facilitation of communication strategies through the dissemination of information to a broader global audience [ 32 ], and the support of advocacy efforts for social change and community mobilization [ 17 ]. Svensson et al [ 33 ] examined the Twitter use of sport-for-development organizations and identified varying levels of engagement across different entities, potentially limiting the cultivation of social media–based social capital within this sector. Investigating the extent of social media use serves as a valuable tool to inform recommendations aimed at enhancing nonprofits’ web-based presence and fostering social media–based social capital [ 2 , 4 , 34 ].

These findings underscore the great significance of further investigation into the scope and nature of social media use, particularly within sectors dedicated to service delivery, such as sexual assault programs and organizations, which share a common mission and focus of their efforts. The endeavor to augment social capital through social media within a given sector has the potential to expand the donor and volunteer base, engage the community in matters affecting everyone, and catalyze broader social change at the policy level by mobilizing concerned citizens.

Aim of the Study

This study aims to address the existing research gap surrounding the use of social media platforms such as Twitter by specific organizations, such as sexual assault support agencies. This study intends to investigate user activities, demographics, and social network structures within these organizations on Twitter. By doing so, we aim to contribute to a better understanding of the current state of social media adoption and social network structures within sexual assault organizations in Canada. In addition, this study provides valuable insights and recommendations for building social capital among the sexual assault organizations on social media. To achieve these goals, we formulated the following research questions (RQs):

• RQ1a: How prevalent are sexual assault centers in Canada with official Twitter accounts, and which provinces and territories have the highest number of centers actively using Twitter?
• RQ1b: What are the geographic locations of sexual assault centers with official Twitter accounts in Canada?
• RQ1c: In what years were the Twitter accounts of sexual assault centers in Canada established, and are there any differences in account creation among provinces and territories?
• RQ1d: What is the average age of these centers since establishing their official Twitter accounts, and do any differences in account creation exist among provinces and territories?
• RQ2a: How many sexual assault centers maintain an active Twitter account each year in each province or territory?
• RQ2b: What are the Twitter activity and posting patterns of sexual assault centers while they are active on Twitter?
• RQ2c: How do the Twitter activity and posting patterns vary across provinces and territories?
• RQ3a: What are the variations in network size, specifically in terms of followers and followings, among sexual assault centers in different provinces and territories in Canada?
• RQ3b: What is the relationship between followers and followings of these organizations on Twitter?
• RQ3c: What insights can be gained from the social network structure of sexual assault centers on Twitter?

This study used purposive sampling to select sexual assault centers in Canada. Our sampling frame was developed by combining lists of sexual assault centers by province and territory from 2 sources: the Canadian Association of Sexual Assault Centres website and the Sexual Assault Centres, Crisis Lines, and Support Services directory. After removing duplicates, our sample frame consisted of 350 sexual assault centers across 10 provinces and 3 territories, providing basic information such as center name, phone number, email, and website. Our inclusion criteria were that the sexual assault center had a Twitter account and had posted at least 1 tweet. To confirm eligibility, a research assistant manually searched the home page of these centers and Twitter pages and conducted Google searches. We determined that 127 organizations had Twitter accounts, but 3 of them had never tweeted anything. As a result, our final sample consisted of 124 Twitter accounts belonging to sexual assault centers across 9 provinces and the Yukon and Northwest Territories. It should be noted that there were no sexual assault centers in Prince Edward Island and Nunavut that used Twitter.

Twitter Handles’ Acquisition

We collected the Twitter account name, location (eg, Toronto, Ontario), and Twitter handle (eg, @ABCD) for each sexual assault center’s Twitter account. The Twitter handle represented as “@name” is used by followers when replying to, mentioning, and sending direct messages to an account. We identified 22 duplicate Twitter handles among the sampled centers. As a result, our final sampling list consisted of 113 unique Twitter handles obtained from 124 centers. We gathered this information directly from the home page of each sexual assault center’s Twitter account.

Data Collection

Acquisition of twitter ids.

To collect the data necessary for this study, we obtained Twitter IDs for the 113 unique Twitter handles in our sample. A Twitter ID (eg, 12345678) is a unique numeric value associated with each Twitter handle, and it cannot be changed. We converted each Twitter handle to its corresponding Twitter ID. To ensure the accuracy of our conversions, 2 research assistants verified the results using 3 different websites: TweeterID, CodeOfaNinja, and Comment Picker.

Collection of Tweets

We used the 113 Twitter IDs associated with the sampled 124 sexual assault centers to collect their corresponding tweets. To accomplish this, we used Twitter’s academic search application programing interface (API) full archive end point and timeline end point, which allowed us to retrieve tweets published as early as 2006 [ 35 ]. We accessed the Twitter API using the native rest API requests. Our data collection process was conducted on March 15, 2023. We downloaded all tweets posted by the sampled centers from the date of each account’s establishment to March 15, 2023. Our data set included 294,350 tweets from 124 sexual assault centers, crisis lines, or support services. We obtained a substantial portion of the total number of tweets published by each Twitter ID on Twitter, specifically, >93.54%.

Data Features

We collected several features for each individual tweet message, including the user ID (user_id_str), user account creation date (user_created_at), user location (user_location), username (user_name), user screen name (user_screen_name), tweet creation time (tweet_created_time), full text of the tweet (full_text), and full text of any retweeted status (retweeted_status_full_text).

On Twitter, users commonly use functions such as retweets, replies, mentions, and hashtags. Retweets refer to publicly shared tweets between users and their followers. Users can also add their own comments and media before retweeting. In addition, users can participate in conversations on Twitter by replying to other users and mentioning them in their tweets. Finally, hashtags allowed users to easily follow and search for topics of interest.

Data Analysis

Preprocessing of raw data.

To address our RQs, we preprocessed the raw data using the following steps:

• We removed URLs from the tweets.
• We removed all punctuation marks, with the exception of apostrophes, which are important for contextual meaning in certain words (eg, “We’re”).
• We removed any bigrams from the set if either of its elements belonged to the list of stop words. For instance, the phrase “The increasing awareness about sexual assault” would generate the bigrams “the increasing,” “increasing awareness,” “awareness about,” “about sexual,” and “sexual assault.” In this case, the stop words “the” and “about” would be removed from the bigrams “the increasing,” “awareness about,” and “about sexual,” leaving the bigrams “increasing awareness” and “sexual assault” in the set.

Descriptive Analysis

Descriptive analysis was used to calculate the number of sexual assault centers in each province, the number of centers created each year, and their average age. The age of each sexual assault center was determined by dividing the month of March 2023 by the establishment date of that center. For example, we used R’s difftime method (R Core Team) to calculate the age of center A’s Twitter account, which was created on March 12, 2009. By subtracting “2009-03-12” from “2023-03-15” to obtain the time difference, we determined that this center has an age of 14 years.

Geolocation Mapping of Sexual Assault Centers

We used the Twitter accounts’ IDs and sexual assault center locations to determine the actual locations of each tweet sent by the 124 centers. The locations were plotted and visualized on a map of Canada. One research assistant manually (QZ) retrieved center location information, including the city, region, and province, from the centers’ official websites and obtained the longitudes and latitudes of the cities where the centers were located. The Google API was used to calibrate the geolocations if the absolute distance discrepancy between manually identified geolocations and Google map geolocations was >6 km. Finally, we developed a Python script to automatically generate D3.js for mapping all the centers with their latitudes and longitudes (the script is available upon request). We used Figma to indicate the cities on the map [ 36 ].

Social Network Analysis

Social network analysis is one of the most effective techniques for visualizing and assessing network connectivity dynamics, offering insights into patterns of connection and disconnection among participants at a given moment. In our study, we used social network analysis to construct networks from Twitter accounts, where nodes represented accounts and directed edges symbolized follower relationships. We used Pyvis [ 37 ] and NetworkX in Python to create the network, resulting in 111 nodes and 995 edges, which is a visual representation of the relationships within the sexual assault center’s Twitter accounts. To delve deeper, we applied a range of metrics: (1) density, which quantifies the percentage of actual connections within the network; (2) degree centrality, which evaluates a node’s significance by examining its connections, distinguishing between incoming (in-degree) and outgoing (outdegree) connections [ 38 ]; (3) eigenvector centrality, which measures a node’s influence in a network by considering the relative scores to connected nodes and is based on the concept that connections to nodes with higher scores exert a more significant influence on determining the node’s score, in contrast to connections with nodes having lower scores [ 39 ]; (4) modularity, indicating the network’s community organization strength through clustering [ 40 , 41 ]; (5) betweenness centrality, evaluating an individual’s role as a bridge between unconnected entities, fostering vital connections among clusters, communities, and organizations [ 38 ]; and (6) closeness, gauging a node’s centrality in a connected graph by summing the shortest path lengths to all other nodes [ 42 ]. These metrics collectively provided a comprehensive understanding of the social network’s structure, shedding light on its various facets and opportunities for relationship cultivation and network analysis.

Ethical Considerations

The data set and analyses relied on publicly accessible secondary Twitter data; thus, no ethics approval or organizational consent was necessary. The study data presented in this manuscript were subjected to anonymization and deidentification procedures. All personally identifiable information, including but not limited to individual organizations’ identities, pictures, user-specific data, or tweets that have not been rephrased, have been meticulously removed from the data set to ensure complete anonymity.

Demographic Profile of Sexual Assault Centers on Twitter

Prevalence of sexual assault centers on twitter.

We found that 124 (35.4%) centers out of the 350 sampled sexual assault centers have an official Twitter account and have posted at least 1 tweet since their establishment. We investigated the locations of the 124 sexual assault centers in Canada that have official Twitter accounts. The province with the highest number of sexual assault centers was Ontario (n=34), followed by British Columbia (BC; n=24) and Quebec (n=23). These 3 provinces accounted for two-thirds (81/124, 65.3%) of all sampled sexual assault centers in Canada. Newfoundland and Labrador and Yukon have only 1 sexual assault center each. Figure 1 shows the prevalence of sexual assault centers in Canada that have posted tweets.

The Geographic Distribution of Sexual Assault Centers on Twitter

To visualize the distribution of sexual assault centers with Twitter accounts, we created geographic distribution maps for Ontario, BC, and Quebec, which had the highest number of centers in our sample. Additional geographic distributions of Twitter accounts in the remaining provinces and territories are presented in Multimedia Appendix 1 .

Our sample included 34 sexual assault centers from 27 cities in Ontario, shown in Figure 2 . Our analysis revealed that most of these centers were concentrated in the southeast region of the province, which is also where most of Ontario’s population resides [ 43 ]. Specifically, many centers were found in the cities of Toronto, Ottawa, Peterborough, Timmins, Brampton, and London, which have larger populations in Ontario.

British Columbia

A total of 24 sexual assault centers were identified in BC, spread across 15 cities, shown in Figure 3 . Vancouver had the highest number of centers in the province, followed by Surrey. As per population distribution, most of the population in BC resides in the southern part of the province [ 44 ], and similarly, most of the sampled centers are located in the southern region of BC.

In Quebec, there are 23 sexual assault centers located in 17 cities, with 5 centers located in Montreal, shown in Figure 4 . We found that nearly all the centers are situated in the southern region of Quebec, where most of the province’s population resides [ 45 ]. Notably, no centers were located in the northern region of the province.

Twitter Account Creation Year by Sexual Assault Centers on Twitter: Provincial and Territorial Differences

We analyzed the year of establishment of the Twitter accounts and assessed whether there were any differences in account creation among provinces and territories. Figure 5 shows our analysis results of the sampled sexual assault centers’ Twitter account creation year. The first Twitter account was created in 2009, and the number of sexual assault centers gradually increased from 2009 to 2012, peaking in 2012 with 23 new accounts. However, from 2013 to 2017, the number of Twitter accounts created by sexual assault centers decreased. In 2019, only 1 sexual assault center created Twitter accounts, and there were no new accounts in 2021. In 2022, we identified 3 centers that had established new Twitter accounts. Notably, all 3 centers that created new Twitter accounts had been in operation for >30 years, as confirmed by our examination of their official profiles.

In addition, we analyzed the distribution of the Twitter accounts created across provinces from 2009 to 2023 and found provincial differences. A total of 8 sexual assault centers from 4 provinces, including Alberta, BC, Nova Scotia, and Ontario, created their Twitter accounts in 2009. From 2009 to 2017, we observed a recurring trend among centers located in Ontario and BC, the 2 provinces with the highest number of centers in Canada, where they established new Twitter accounts on an annual basis. Ontario contributed to the newest Twitter accounts created in 2011 (n=5) and 2012 (n=13). In Quebec, the third-largest province in terms of the number of centers with Twitter accounts, all sexual assault centers began using Twitter after 2010, and remarkably, 7 new centers were created in that year alone. We also noted that most centers in Saskatchewan established their Twitter accounts in 2014. In addition, 3 sexual assault centers located in the Canadian territories also created their Twitter accounts. Specifically, in 2019, one organization in Yukon established a Twitter account, whereas in 2010, two centers located in the Northwest Territories created their Twitter accounts.

Average Age of Sexual Assault Centers on Twitter by Province and Territory

We calculated the age of each sexual assault center’s Twitter account since its creation. We determined the duration of time that each sexual assault center had its official Twitter account by subtracting the most recent month of collected tweets (March 2023) from the account creation date. Then, we computed the average length of time for all sexual assault centers in each province and presented the results in Table 1 .

Table 1 displays the average length of time, SD, and range of the duration of years after the establishment of Twitter accounts by sexual assault centers in each province. For example, 13 sexual assault centers in Alberta created Twitter accounts, and the average number of years since their accounts’ establishment was 10.26 (SD 2.39) years. The range of 5 to 14 indicated that the earliest Twitter account was created in 2009 (March 2023: 14 y), whereas the most recent account was established in 2018 (March 2023: 5 y). These 13 sexual assault centers accounted for 10.5% (n=124) of the total 124 sexual assault centers in Canada.

To obtain a comprehensive understanding of Twitter presence and engagement, we used the metadata of Twitter users related to sexual assault centers in each province, which we obtained from the Twitter timeline API. Specifically, we extracted data such as “followers_count,” “friends_count,” “favorites_count,” and “listed_count” to determine the total number of followers, following, favorites, and listed users, respectively, for each province. We also aggregated the collected data by province to calculate the total number of tweets posted for each province.

User Activity of Sexual Assault Centers on Twitter

Active twitter accounts in canadian sexual assault centers by province and territory.

We analyzed the data to determine the number of active Twitter accounts maintained by sexual assault centers in each Canadian province and territory each year. We defined an active account as one that posted at least 1 tweet in a given year. The results indicated a steady increase in the number of active Twitter accounts in Ontario and BC since 2009, as shown in Figure 6 . In Quebec, there was an increase in the number of centers from 3 in 2011 to 9 in 2015, but this trend reversed in the following years, indicating that many Twitter accounts became inactive. Although there was an increase in the number of registered Twitter accounts from 13 in 2015 to 16 in 2018, only 4 of these accounts remained active in 2020, down from 9 in 2015. Alberta ranks third or fourth in terms of active Twitter accounts.

In Manitoba, 2 sexual assault centers had active Twitter accounts until 2020. Only 1 center in New Brunswick registered a Twitter account in 2013 but was inactive in 2018 and 2020 while maintaining activity in other years. In Newfoundland and Labrador, the only sexual assault center remained active on Twitter from 2012 to 2020. Nova Scotia consistently showed an increasing trend in the number of active Twitter accounts from 2009 to 2020. Meanwhile, the only center in the Northwest Territories maintained an active Twitter account since 2010. In Saskatchewan, the number of active Twitter accounts increased to 6 in 2015 but decreased to 3 in 2020.

Twitter Activity and Posting Patterns of Sexual Assault Centers

We also examined the Twitter activity of sexual assault centers in Canada, investigating the popular times for tweeting and the number of tweets posted per month. Figure 7 shows the total number of tweets posted by all sampled centers aggregated by month. Over a 12-year period, these centers posted an average of 12,849 tweets per month. The most active month was November, with a total of 16,239 tweets, whereas the least active was July, with only 9079 tweets. March and May were the peak tweeting months, with >15,000 tweets, whereas August had the fewest tweets, with approximately 9500 tweets.

We analyzed the monthly average tweet count of sexual assault centers during their active status (n=92). We computed the average number of tweets sent by each center per month from the first tweet after creating their account to the last tweet during the data collection period. Results showed that a large majority of the centers have a relatively low tweeting frequency, with the highest frequency of centers (over 20) averaging between 0 to 10 tweets per month. The distribution is right-skewed, showing that as the average monthly tweeting volume increases, the number of centers engaging at the level decreases. A small number of centers tweet between 10 and 30 times per month, and very few centers exceed this range. There are occasional outliers, with one center in particular averaging a significantly higher number of tweets per month, at around 180. This center is an extreme outlier in comparison to the rest of the data set. Overall, our findings suggest that sexual assault centers tend to use Twitter moderately, with the bulk of them tweeting less than 20 times per month, and a very exceptional few tweeting much more frequently.

Comparative Analysis of Tweet Activity Across Provinces and Territories

To answer RQ5, we further analyzed the total number of tweets posted by centers in different provinces and territories each month and compared the number of tweets posted across provinces and territories each year. Figure 8 shows the annual tweet activity generated from all sexual assault center accounts across provinces and territories, with data collected until March 15, 2023. Our analysis of the tweet activity revealed a gradual increase in tweeting volume across provinces and territories until 2017. Notably, Ontario exhibited the highest frequency of tweet activity in 2016, with approximately 17,000 tweets. However, with the exception of accounts in BC and Nova Scotia, the tweet activity gradually declined from 2017 to March 2023, returning to activity levels last observed in 2013 or 2014. It is worth noting that some provinces and territories, such as New Brunswick and Northwest Territories, had <100 tweets in the peak month, and hence, they were not included in our figure. The findings suggest a potential decrease in Twitter activity among sexual assault centers in recent years.

Social Network Dynamics of Sexual Assault Centers in Canada

Network size.

The sexual assault organizations under investigation had an average follower count of 1543 (SD 1555) and an average following count of 819 (SD 876). The range of follower counts was quite diverse, starting at a minimum of 8 followers for one organization and reaching a maximum of 7228 followers for another organization. Similarly, the following counts also showed significant variation, with one center having the lowest count of 1 following, whereas another center had the highest following count of 4458. More detailed information about the Followers, followings, and measurement of social network analysis are in Multimedia Appendix 2 .

Sexual assault organizations across 11 provinces exhibited varying average follower counts, ranging from as low as 1 follower in New Brunswick to 17 followers in Ontario, as shown in Table 2 . The organization located in Ontario had the highest number of followers, totaling 46. In contrast, some sexual assault organizations in provinces such as Alberta and BC had no followers at all. Similarly, the average number of followings by these organizations across the 11 provinces and territories ranged from 0 in New Brunswick to 18 in Ontario. The organization with the most followings was also situated in Ontario, with a total of 38 followings, whereas several organizations in provinces such as New Brunswick and Quebec had no followings.

a N/A: not applicable.

Relationship Between Followers and Followings on Twitter

We analyzed the Twitter user network by exploring the connections between the followers and the following lists. Figure 9 shows the log-log plot of the correlation between followers and followings. Each point on the graph represents an individual user, with the x-axis representing the user’s followers and the y-axis indicating their following count. The plot demonstrates that as the number of followers continued to increase, the followings also indicated an increasing trend. At the middle of the plot, users with a medium number of followers have a high number of followings.

Analysis of Twitter Social Network Structure and Node Categorization

Figure 10 presents a full network map, illustrating the relationships between followers and followings among 111 sexual assault centers on Twitter. This graph features 111 nodes and 995 edges. Each node represented a sexual assault center on Twitter, and each edge is directional, with arrows symbolizing “x follows y” relationship. In this context, Y serves as the following node, implying that it is followed by other nodes, whereas X acts as the follower node, signifying that it follows other nodes.

Among the 111 sexual assault centers on Twitter, the average count of both followers and followings was approximately 9. There is a notable variability in these counts, with a SD of 9.15 for followings and 9.85 for followers. The maximum number of followings observed was 38, whereas the maximum number of followers reached was 46, with a minimum count of 0.

We categorized them into different categories to account for the variability in the number of followers. The graph illustrates nodes of various colors, each representing a specific range of followers. Red denotes those with <11 followers, green indicates 11 to 21 followers, blue represents 21 to 31 followers, purple signifies 31 to 41 followers, and yellow indicates those with ≥41 followers. Notably, there is a single green node, which stands out with 46 followers. The in-degree centrality is 0.42, which is the highest value among all the nodes, underscoring their significance. In addition, they possess a closeness score of 0.49, ranking them within the top 1% among all nodes, implying their high level of proximity to other nodes and less dependency on others for information transmission. Furthermore, their betweenness score was 0.076, signifying their involvement in a substantial number of shortest paths and positioning them among the top 7 nodes in the betweenness ranking. We classify the sexual assault centers that meet the criteria of high closeness, in-degree centrality, low betweenness, and >41 followers and followings as social queens .

The figure also draws attention to a cluster of sexual assault centers characterized by a smaller number of followers and followings. We classified these nodes as marginalized entities. These centers have a limited impact on information transmission on Twitter, leading to their closeness and other metrics registering at 0.

Identification of Modular Patterns and Key Nodes in Twitter Network

Figure 11 is a section from Figure 10 , in which distinct patterns emerge as certain nodes cluster together into modularities. As an example, we can examine a particular section of Figure 10 where a modularity forms—a small cluster consisting of several nodes that mutually follow each other. We incorporated eigenvector centrality to interpret this modularity. Within the bottom-right corner of this modularity, we find that accounts 24 and 72 (anonymous Twitter handle_names) exhibit relatively high values of eigenvector centrality. A higher eigenvector centrality score implies greater significance of the node when compared with its neighboring points. Furthermore, the importance of the node itself is directly linked to the significance of the neighboring nodes connected to it. Consequently, this specific node can be regarded as the “social queen” within this particular modularity.

Principal Findings

This study represents a pioneering effort to conduct a comprehensive analysis of Twitter’s social network, user activities, and demographics within the context of sexual assault centers in Canada. By mapping and analyzing their Twitter practices, this research contributes to a better understanding of the social media landscape of sexual assault support organizations in Canada. The findings underscore the potential of Twitter as a platform for sexual assault organizations to build social capital, enhance their influence, and expand their reach. Moreover, it highlights the need for tailored engagement strategies that consider regional disparities and the unique characteristics of each province and territory. Our findings align with the broader literature on social capital, specifically bridging and bonding social capital. Among the various social media platforms, Twitter emerges as a valuable data set to study sexual violence [ 46 ] as well as a notable facilitator of bridging social capital, consistent with previous research that has underscored Twitter’s ability to connect organizations with a diverse and expansive audience [ 22 ].

The results of this study reveal a substantial presence of sexual assault centers in Canada on Twitter, signifying their acknowledgment of Twitter as a valuable communication and social capital development tool. Out of the 350 sampled centers, 124 (35.4%) maintain an active Twitter presence, highlighting the significant proportion of sexual assault organizations in Canada that recognize Twitter’s efficacy as a communication medium for engaging with their stakeholders and the public. This trend aligns with the broader nonprofit sector, where most large and mid-sized nonprofits maintain at least 1 social media account [ 9 ]. However, it is worth noting that effective communication on Twitter may be hindered by a reliance on broadcasting rather than engaging in dialogue, as observed in previous research [ 47 ]. The success of acquiring social capital through social media appears to depend on the extent and quality of stakeholder connections, emphasizing the importance of diverse engagement strategies and the diversity and complexity of message elements [ 26 ]. In the context of sexual assault centers and support services, social media–based social capital holds significant potential for increasing the donor and volunteer base, engaging with the community on social issues, and promoting wider social change [ 33 ].

The geographic distribution of sexual assault centers with Twitter accounts highlights regional variations in Twitter use and engagement. Ontario, BC, and Quebec emerged as the provinces with the highest number of centers using Twitter, collectively accounting for two-thirds of all sampled centers in Canada, indicating higher levels of social capital in those regions owing to increased opportunities for information sharing, emotional kinship, trust, and social support [ 20 ]. The concentration of centers in these provinces aligns with their higher population densities and emphasizes the importance of social media platforms, such as Twitter, in reaching a broader audience consistently [ 43 ]. The southeast regions of Ontario and Quebec as well as the southern region of BC showed a higher concentration of sexual assault centers with Twitter accounts, likely reflecting the higher population densities in these areas. Furthermore, spatial distribution may influence the topics and issues addressed in their tweets. In BC, the distribution of sexual assault centers was also concentrated in specific regions, such as Vancouver. The content of tweets from centers in these regions may reflect local concerns and initiatives. It is essential to consider the unique characteristics and needs of each region when developing communication strategies and leveraging social media platforms for social capital enhancement.

The average age of sexual assault centers’ Twitter accounts was calculated to determine the duration of their presence on the platform. The findings showed that the average duration varied across provinces and territories, ranging from 5 to 14 years. The Northwest Territories had the longest average duration of 12.77 years, indicating a relatively early adoption and subsequent use of Twitter among sexual assault centers in the territory. In contrast, provinces or territories such as Yukon and New Brunswick had a shorter average duration. These variations in account age reflect differences in the timing of adoption and highlight the diverse trajectories of Twitter use among sexual assault centers across Canada. These differences may be influenced by organizational factors, regional context, or resource availability. Centers with older Twitter accounts may have accumulated more followers and established stronger web-based communities, whereas newer accounts may need to focus on building and expanding their web-based presence.

Patterns of account creation offer insights into the temporal dynamics of engagement and emphasize the need for continuous and consistent communication efforts. The recent decline in the creation of new Twitter accounts by sexual assault centers in recent years may signal a saturation point, where most centers have already established their Twitter presence. Alternatively, it could be attributed to factors such as resource constraints, changing organizational priorities, limited staff dedicated to communication practices, or a shift in focus to other social media platforms.

The examination of the average monthly tweet count for active sexual assault centers provides valuable insights into their Twitter activity levels. The results indicate variations in tweet frequency across provinces, with Ontario and BC consistently demonstrating higher tweet volumes compared with other provinces. This observation aligns with the higher number of active Twitter accounts and underscores the importance of ongoing engagement and dialogue with stakeholders through regular tweets.

The findings related to social network dynamics reveal the landscape of Twitter engagement among sexual assault organizations. On average, these organizations have amassed approximately 1543 followers, demonstrating their capacity to reach a substantial audience. Simultaneously, they follow an average of 819 other accounts, indicating their active involvement within the Twitter community. This indicates the potential for these organizations to disseminate information, provide support, and raise awareness about their critical missions. This study aligns with the perspective emphasizing the importance of bridging social capital facilitated by Twitter’s ability to connect with various stakeholders, including service recipients, donors, and the general public [ 26 ]. It highlights the potential and disparities in bridging social capital among sexual assault centers across provinces and territories in Canada, suggesting that these organizations can better leverage Twitter to establish connections beyond their immediate constituencies. This aligns with the notion that social media platforms such as Twitter can extend an organization’s reach and promote the flow of information across various stakeholders [ 22 ].

The findings also uncovered regional disparities in Twitter engagement among sexual assault organizations in Canada. Sexual assault organizations across Canada’s provinces exhibited varying degrees of Twitter activity. Although some provinces, such as Ontario, displayed robust engagement, others, such as New Brunswick, had limited presence and following. Our observation resonates with previous studies (eg, [ 9 ]) that have emphasized the role of regional context in shaping nonprofit organizations’ social media use. These regional disparities suggest the need for tailored strategies to maximize the impact of Twitter engagement, considering the unique characteristics and needs of each province.

Within the context of nonprofit organizations, research has indicated a positive relationship between follower count and bonding social capital [ 21 ]. Our study aligns with this perspective by demonstrating a positive association between followers and followings. As follower counts increase, there is a corresponding increase in followings, indicating a proactive approach by organizations to engage with their audience. This observation underscores the importance of reciprocity and interaction on Twitter. This suggests that a larger number of followers on Twitter can contribute to increased financial support, in line with the positive impact of web-based social capital generated through social networking sites on charitable outcomes [ 29 , 30 ].

Within the intricate network of sexual assault centers on Twitter, we identified nodes with distinct characteristics. Some organizations emerged as “social queens,” characterized by high in-degree centrality, closeness scores, and low betweenness, coupled with substantial followers and followings. These “social queens” play pivotal roles in information transmission, networking, and community building. This finding suggests that organizations can strategically use Twitter to enhance their influence and reach within their fields of operation. However, this study also highlights the presence of marginalized entities with limited follower counts, which may face challenges in impacting information transmission on Twitter. This underscores the importance of proactive engagement strategies for organizations seeking to maximize their impact through social media. Sexual assault organizations can benefit from a comprehensive understanding of their social network structures, enabling them to identify opportunities to strengthen their social capital, expand their donor base, and effectively engage the community.

Limitations

This study had some limitations. First, the findings are specific to sexual assault centers in Canada and may not be applicable to other countries or regions owing to cultural, social, and organizational differences. Each country or region may have unique characteristics that influence the use of Twitter and other social media platforms by sexual assault organizations. Second, the study focuses solely on Twitter data and does not consider the use of other platforms such as Facebook, Instagram, or Snapchat, which could provide additional insights into communication strategies and social media practices. Therefore, the findings of this study may not provide a comprehensive understanding of the organizations’ overall social media use. Third, the study is cross-sectional, providing a snapshot of Twitter use at a specific time, and does not capture longitudinal changes or trends. A longitudinal study would offer more detailed insights into the evolution of Twitter practices and the effectiveness of communication strategies used by these organizations. For example, we lack information about growth and changes in the number of followers over time. The follower counts remained static at the time of data collection. Future studies may need to explore how follower counts evolve dynamically to gain deeper insights. Fourth, the study primarily focused on describing Twitter use at the organizational level rather than evaluating the effectiveness or outcomes of the communication strategies used. Further research is required to assess the impact and outcomes of social media use in this context. Fifth, this study did not have access to demographic information related to the organizational size of the nonprofits, which typically includes factors such as the number of employees, volunteers, or annual budget. Unfortunately, Twitter does not provide access to such data, resulting in its absence from this study. Finally, the study does not delve deeper into the content of tweets posted by sexual assault organizations on Twitter. Future studies could explore the thematic analysis of tweets, sentiment analysis to understand the emotional tone of their messages, and the effectiveness of specific content strategies used by these organizations to engage their audience and advocate for their cause.

Conclusions

In conclusion, this study provides valuable insights into the current use and social structure of Twitter by sexual assault centers, crisis lines, and support services in Canada. The findings highlight the widespread adoption of Twitter among these organizations and the potential for leveraging social media platforms to build social capital. By recognizing regional disparities, identifying key players, and understanding the dynamics of followers and followings, sexual assault organizations can better navigate the Twitter landscape to further their missions of promoting awareness and support for survivors of sexual assault. Further research in this area can explore the long-term impact of social media use on organizational outcomes and stakeholder perceptions into enhancing social capital within the nonprofit sector and beyond, providing additional guidance for effective communication practices in the nonprofit human services sector and ultimately contributing to the broader goals of these organizations.

Conflicts of Interest

None declared.

Sexual assault centers with official Twitter accounts in Alberta, New Brunswick, Newfoundland and Labrador, Nova Scotia, Manitoba, Saskatchewan, Northwest Territories, and Yukon.

Followers, followings, and measurement of social network analysis.

• Guo C, Saxton GD. Getting attention: an organizational-level analysis. In: The Quest for Attention. Redwood City, CA. Stanford University Press; 2020.
• Guo C, Saxton GD. Speaking and being heard: how nonprofit advocacy organizations gain attention on social media. Nonprofit Volunt Sect Q. 2018;47(1):5-26. [ CrossRef ]
• Bhati A, McDonnell D. Success in an online giving day: the role of social media in fundraising. Nonprofit Volunt Sect Q. 2020;49(1):74-92. [ CrossRef ]
• Campbell DA, Lambright KT. Are you out there? internet presence of nonprofit human service organizations. Nonprofit Volunt Sect Q. 2019;48(6):1296-1311. [ CrossRef ]
• Jung K, Valero JN. Assessing the evolutionary structure of homeless network: social media use, keywords, and influential stakeholders. Technol Forecast Soc Change. Sep 2016;110:51-60. [ CrossRef ]
• Maxwell SP, Carboni JL. Social media management. Nonprofit Manag Leadersh. Sep 19, 2016;27(2):251-260. [ CrossRef ]
• Zhou H, Ye S. Legitimacy, worthiness, and social network: an empirical study of the key factors influencing crowdfunding outcomes for nonprofit projects. Voluntas. Jun 4, 2018;30:849-864. [ CrossRef ]
• Zhou H, Ye S. Fundraising in the digital era: legitimacy, social network, and political ties matter in China. Voluntas. Apr 01, 2019;32:498-511. [ CrossRef ]
• Nah S, Saxton GD. Modeling the adoption and use of social media by nonprofit organizations. New Media Soc. 2013;15(2):294-313. [ CrossRef ]
• Ellison NB, Vitak J, Gray R, Lampe C. Cultivating social resources on social network sites: Facebook relationship maintenance behaviors and their role in social capital processes. J Comput Mediat Commun. Jul 01, 2014;19(4):855-870. [ CrossRef ]
• Li C, Bernoff J. Groundswell: Winning in a World Transformed by Social Technologies. Brighton, MA. Harvard Business Review Press; 2011.
• Park H, Rodgers S, Stemmle J. Analyzing health organizations' use of Twitter for promoting health literacy. J Health Commun. 2013;18(4):410-425. [ CrossRef ] [ Medline ]
• Lovejoy K, Waters RD, Saxton GD. Engaging stakeholders through Twitter: how nonprofit organizations are getting more out of 140 characters or less. Public Relat Rev. Jun 2012;38(2):313-318. [ CrossRef ]
• Xue J, Chen J, Chen C, Hu R, Zhu T. The hidden pandemic of family violence during COVID-19: unsupervised learning of tweets. J Med Internet Res. Nov 06, 2020;22(11):e24361. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Lovejoy K, Saxton GD. Information, community, and action: how nonprofit organizations use social media. J Comput Mediat Commun. Apr 01, 2012;17(3):337-353. [ CrossRef ]
• Xue J, Macropol K, Jia Y, Zhu T, Gelles RJ. Harnessing big data for social justice: an exploration of violence against women‐related conversations on Twitter. Human Behav Emerg Technol. Jul 26, 2019;1(3):269-279. [ CrossRef ]
• Guo C, Saxton GD. Tweeting social change: how social media are changing nonprofit advocacy. Nonprofit Volunt Sect Q. 2014;43(1):57-79. [ CrossRef ]
• Kapucu N, Demiroz F. A social network analysis approach to strengthening nonprofit collaboration. J Appl Manag Entrep. Jan 2015;20(1):87-101. [ CrossRef ]
• Putnam RD. Tuning in, tuning out: the strange disappearance of social capital in America. PS Political Sci Politics. 1995;28(4):664-683. [ FREE Full text ] [ CrossRef ]
• Putnam RD. Bowling Alone: The Collapse and Revival of American Community. New York, NY. Simon & Schuster; 2001.
• Hofer M, Aubert V. Perceived bridging and bonding social capital on Twitter: differentiating between followers and followees. Comput Hum Behav. Nov 2013;29(6):2134-2142. [ CrossRef ]
• Phua J, Jin SV, Kim J. Uses and gratifications of social networking sites for bridging and bonding social capital: a comparison of Facebook, Twitter, Instagram, and Snapchat. Comput Hum Behav. Jul 2017;72:115-122. [ CrossRef ]
• Bourdieu P. Distinction: a social critique of the judgment of taste. In: Food and Culture, 4th Edition. Milton Park, UK. Routledge; 2017.
• Lin N. Building a network theory of social capital. In: Social Capital, 1st Edition. Milton Park, UK. Routledge; 2001.
• Lai CH, Fu JS. Humanitarian relief and development organizations’ stakeholder targeting communication on social media and beyond. Voluntas. Mar 05, 2020;32:120-135. [ CrossRef ]
• Xu W, Saxton GD. Does stakeholder engagement pay off on social media? a social capital perspective. Nonprofit Volunt Sect Q. Aug 02, 2018;48(1):28-49. [ FREE Full text ] [ CrossRef ]
• Doerfel ML, Atouba Y, Harris JL. (Un)obtrusive control in emergent networks: examining funding agencies’ control over nonprofit networks. Nonprofit Volunt Sect Q. 2017;46(3):469-487. [ CrossRef ]
• Williams JR. The use of online social networking sites to nurture and cultivate bonding social capital: a systematic review of the literature from 1997 to 2018. New Media Soc. 2019;21(11-12):2710-2729. [ CrossRef ]
• Henry R, Bosman L. Strategic management and social media: an empirical analysis of electronic social capital and online fundraising. In: Olivas-Luján MR, Bondarouk T, editors. Social Media in Strategic Management. Bingley, UK. Emerald Group Publishing Limited; 2013;43-62.
• Lee YJ, Shon J. Nonprofits’ online social capital and charitable support. J Nonprofit Public Sect Mark. 2023;35(3):290-307. [ CrossRef ]
• Saxton GD, Wang L. The social network effect: the determinants of giving through social media. Nonprofit Volunt Sect Q. Apr 24, 2013;43(5):850-868. [ CrossRef ]
• Seo H, Vu HT. Transnational nonprofits’ social media use: a survey of communications professionals and an analysis of organizational characteristics. Nonprofit Volunt Sect Q. Feb 28, 2020;49(4):849-870. [ CrossRef ]
• Svensson PG, Mahoney TQ, Hambrick ME. Twitter as a communication tool for nonprofits: a study of sport-for-development organizations. Nonprofit Volunt Sect Q. Oct 16, 2014;44(6):1086-1106. [ CrossRef ]
• Yang A, Liu W. Coalition networks for the green new deal: nonprofit public policy advocacy in the age of social media. Nonprofit Volunt Sect Q. Sep 20, 2022;52(5):1284-1307. [ CrossRef ]
• Makice K. Twitter API: Up and Running. Sebastopol, CA. O'Reilly Media, Inc; Mar 2009.
• Amr T, Stamboliyska R. Practical D3.js. New York, NY. Apress; 2016.
• Interactive network visualizations. Pyvis. URL: https://pyvis.readthedocs.io/en/latest/ [accessed 2024-03-01]
• Conducting a social network analysis. Converge. URL: https://www.converge.net/toolkit/conducting-a-social-network-analysis [accessed 2024-03-01]
• Bonacich P. Some unique properties of eigenvector centrality. Soc Netw. Oct 2007;29(4):555-564. [ CrossRef ]
• McCurdie T, Sanderson P, Aitken LM. Applying social network analysis to the examination of interruptions in healthcare. Appl Ergon. Feb 2018;67:50-60. [ CrossRef ] [ Medline ]
• Newman ME, Girvan M. Finding and evaluating community structure in networks. Phys Rev E Stat Nonlin Soft Matter Phys. Feb 2004;69(2 Pt 2):026113. [ CrossRef ] [ Medline ]
• Closeness centrality (centrality measure). GeeksforGeeks. URL: https://www.geeksforgeeks.org/closeness-centrality-centrality-measure/ [accessed 2024-03-01]
• Canada Ontario Density 2016.png. Wikimedia Commons. URL: https://commons.wikimedia.org/wiki/File:Canada_Ontario_Density_2016.png#file [accessed 2024-03-01]
• McPhee A. Canada British Columbia Density 2016.png. Wikimedia Commons. 2019. URL: https://commons.wikimedia.org/wiki/File:Canada_British_Columbia_Density_2016.png
• McPhee A. Canada Quebec Density 2016.png. Wikimedia Commons. 2019. URL: https://commons.wikimedia.org/wiki/File:Canada_Quebec_Density_2016.png
• Xue J, Zhang B, Zhang Q, Hu R, Jiang J, Liu N, et al. Using Twitter-based data for sexual violence research: scoping review. J Med Internet Res. May 15, 2023;25:e46084. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Stabile B, Grant A, Purohit H, Sharan Bonala S. Take back the tweet: social media use by anti‐gender‐based violence organizations. Sex Gend Policy. Apr 29, 2021;4:38-56. [ CrossRef ]

Abbreviations

Edited by A Mavragani; submitted 04.07.23; peer-reviewed by W Ceron; comments to author 04.12.23; revised version received 23.12.23; accepted 13.02.24; published 27.03.24.

©Jia Xue, Qiaoru Zhang, Yun Zhang, Hong Shi, Chengda Zheng, Jingchuan Fan, Linxiao Zhang, Chen Chen, Luye Li, Micheal L Shier. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 27.03.2024.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.

Effects of COVID-19 Shutdowns on Domestic Violence in the U.S.

This chapter examines the impact of COVID-19 shutdowns on domestic violence (DV) in the United States. Despite widespread concerns that pandemic shutdowns could increase DV, initial studies found mixed evidence that varied across data sources and locations. We review the evolving literature on the effects of the pandemic and highlight results from studies that examine multiple measures of DV across a common set of large cities. These studies show that the conflicting early results are due to opposite effects of pandemic shutdowns on two measures of DV in police data: an increase in domestic violence 911 calls and a decrease in DV crime reports. In theory, this divergence can come from either higher DV reporting rates, possibly because of additional media attention to DV and greater third-party calling, or from lower policing intensity for DV crimes. Prior evidence from police data and other sources supports the conclusion that the increase in calls came from greater reporting, while the incidence of criminal DV decreased. Finally, we present new evidence drawing on police and hospitals records from across the state of California to show that DV crimes and hospital emergency department (ED) visits were both lower during pandemic shutdowns.

We acknowledge financial support from the IZA COVID-19 Research Thrust and from the Bill and Melinda Gates Foundation, through the NBER Gender in the Economy Study Group Research Grants on Women, Victimization, and COVID-19. We have no competing interests to disclose. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.

MARC RIS BibTeΧ

Download Citation Data

Working Groups

More from nber.

In addition to working papers , the NBER disseminates affiliates’ latest findings through a range of free periodicals — the NBER Reporter , the NBER Digest , the Bulletin on Retirement and Disability , the Bulletin on Health , and the Bulletin on Entrepreneurship  — as well as online conference reports , video lectures , and interviews .

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• HHS Author Manuscripts
• PMC10177625

Violent Media in Childhood and Seriously Violent Behavior in Adolescence and Young Adulthood

Michele l. ybarra.

Center for Innovative Public Health Research, San Clemente, California

Kimberly J. Mitchell

Crimes against Children Research Center, University of New Hampshire, Durham, New Hampshire

Jay Koby Oppenheim

Independent Consultant, New York, New York

Associated Data

To quantify the relative odds of self-reported seriously violent behavior in adolescence and young adulthood given one’s self-reported violent media diet in childhood.

Baseline data were collected nationally online from 1,586 youth 10–15 years of age in 2006. Follow-up data were collected in 2010–2011 and 2016. Children reported the amount of music, video games, television, websites with real people, and cartoons that depicted “physical fighting, hurting, shooting, or killing.” Seriously violent behavior was assessed 5 and 10 years later.

887 adolescents completed the survey at baseline and 5-year follow-up. The relative odds of reporting seriously violent behavior over time were 2.45-fold higher ( P <.001) with each incremental increase in one’s baseline violent media diet. After adjusting for other potentially influential characteristics, results persisted (aOR = 1.70, P =.01). The relative odds also were elevated for those frequently exposed to violence in music (aOR = 3.28, p=0.03), television (aOR = 3.51, p<0.001), and video games (aOR = 3.27, p=0.02). 760 young adults completed measures at baseline and 10-year follow-up. The relative odds of seriously violent behavior increased 2.18-fold ( P =.001) with each incremental increase in one’s baseline violent media diet. After adjusting for other factors, the association persisted (aOR = 1.72, P=.03). Frequent exposure to violence in video games (aOR = 3.28, p=0.03) and television (aOR = 3.14, p=0.02) also were implicated.

Discussion:

Exposure to violent media in childhood may be one modifiable influence on seriously violent behavior in adolescence and adulthood, even for those who have other risk factors.

Youth violence is a significant public health issue that negatively affects individuals, families, and communities. 1 , 2 Estimated costs associated with youth violence in the United States is more than $20 billion anually. 3 Although juvenile arrests in 2019 were down 58% since 2010, 4 youth nonetheless account for a sizable proportion of perpetrators: 9% of all violent crimes were committed by juveniles, and 21% by 18–24-year-olds. 5

No single risk factor causes violent behavior. Instead, an accumulation of exposures increases one’s risk at each level of the social ecology (e.g., exposure to spousal abuse). 2 , 6 – 9 Because it could easily be modified, exposure to violent media has been researched for decades as a potential contributor to aggressive behavior. Cross-sectional and laboratory research frequently document linkages. 10 – 12 Studies that measure violent behaviors report similar effect sizes to those that measure aggression. 10 Although fewer in number, longitudinal studies also report linkages: Huesmann and Eron found that adult criminal and violent behavior was associated with exposure to television violence 15 years prior. 13 Findings were replicated in a Finnish sample. 14 Further, Anderson and colleagues found that frequent violent video game play predicted physical aggression three to six months later for children and adolescents in three separate cohorts, two from Japan and one from the United States. 15 Some exceptions are noted. 16 Coyne and colleagues looked at longer term associations between externalizing behavior and violent video game play and did not find a linkage over the 5-year observation period. 17 This may be because the measure reflecting externalizing behavior included items that did meet the definition of aggression.

Youth media use is nearly ubiquitous 18 : Music is by far the most widely used medium in adolescence: 82% listen to music daily. 19 Most - 83% of adolescent girls and 97% of boys 13–17 years of age – also play video games; 95% own or have access to a smartphone, and 85% say they go online and exchange content. 20 Cross-sectional research by Ybarra and colleagues suggests that one’s general media violence diet may explain the increased odds of engaging in seriously violent behavior. 21 As such, it is important not just to examine the association that specific media may have but also the association that one’s violent media ‘diet’ across media may have with violent behavior over time.

The current study aims to fill noted research gaps. First, while extant research examines exposure to violence on television and in video games, exposures through other media, such as music, are less well studied yet constitute a large part of youth media diets. Second, much of the literature focuses on aggressive rather than violent behavior. Aggression is any behavior enacted by someone who intends to harm the other person when the other person does not want to be harmed. 22 , 23 Violence is a more severe type of aggression that carries with it the possibility of serious physical harm to the other individual. All violent behaviors are aggressive, whereas not all aggressive behaviors are violent. Third, few studies examine these linkages longitudinally, particularly between 5 and 10 years postexposure. Based on previous literature, we posit that violent media will predict violent behavior over time and that this will be particularly true for a general media diet as it reflects an accumulation of exposures.

Growing up with Media is a longitudinal study designed to study the association between violent media exposure in childhood and adolescence - particularly exposures to new media, including the Internet and seriously violent behavior. The survey protocol was reviewed and approved by the Centers for Disease Control and Prevention Institutional Review Board (IRB) for Waves 1–3 and by Chesapeake IRB for Waves 4–7 (subsequently acquired by Advarra IRB). Parents provided informed consent for their participation and permission for their child’s participation, and youth provided informed assent by reading the assent information and then clicking either “Yes, I want to take the survey” or “No, I do not want to take the survey.”

In 2006, 1,586 child-caregiver pairs were recruited through an email sent to randomly identified adult Harris Poll OnLine (HPOL) panel members who reported having a child living in their household. HPOL was the largest online panel at the time of recruitment, including four million members. Members were recruited through online advertising, advertising at conferences and events, and referrals.

Eligible adult caregivers reported having a child 10–15 years of age living in the household, speaking English, and being equally or more knowledgeable than other adults living in the household about their youth’s daily activities. Eligible youth participants were 10–15-year-olds who read English, lived in the household at least 50% of the time, and had used the Internet at least once in the last six months. Recruitment was balanced on youth age and sex; once a demographic ‘bin’ was filled (e.g., for 10–12-year-old girls), subsequent youth who met those criteria were marked ineligible.

Seriously violent behavior.

Seriously violent behavior, as defined by the US Department of Justice, 24 includes murder, aggravated assault, robbery, and sexual violence. Youth were coded as having engaged in past-year seriously violent behavior if they endorsed any of the following five behaviors: (1) behaviors that would likely result in murder (i.e., stabbing or shooting someone); (2) aggravated assault (i.e., threatening someone with a weapon; attacking someone resulting in the need for medical care); (3) robbery (i.e., using a knife or gun or some other kind of weapon like a bat to get something from someone else); and (4) sexual assault (kissing, touching, or doing anything sexual with another person when it was not wanted by that person). This last item was written to be developmentally appropriate for 10–15-year-olds. Because it may include behaviors that extend beyond rape, a sensitivity analysis was conducted to examine the results when this measure of sexual assault was excluded.

Exposure to violent media.

Youth reported the amount of violence they were exposed to across five different types of media: Television, computer and video games, music, websites of real people, and websites of cartoons. A similar question format was used for each medium: “When you [engage with media type], how many of them [show/talk about] physical fighting, hurting, shooting, or killing?” 25 Response options were captured on a four-point Likert scale [1 (almost none/none of them) – 4 (almost all / all of them)].

To reflect a general violent media diet, a factor score that included all five media, was estimated using maximum likelihood [Eigenvalue = 1.69, factor loadings ranged from .47 - .69, α = 0.70, Kaiser-Meyer-Olkin ranges from 0.71 – 0.78].

For specific mediums, a categorical measure was created based upon data distributions to reflect those who reported that: (1) none/almost none, (2) some, or (3) many or almost all/all of each of the medium they consumed depicted violence. Because of low cell stability, for all longitudinal analyses, baseline exposure to violence on television was dichotomized to compare none/almost none or some versus many or almost all/all; baseline exposures to real people engaging in violence online was dichotomized to none/almost none versus some, many, almost all/all. Wave 7 longitudinal analyses included a measure of baseline exposures to cartoons engaging in violence online dichotomized to none/almost none versus some, many, almost all/all.

Background variables.

Youth age and sex were reported by caregivers; race and ethnicity were reported by youth. At the individual level, because trait anger can be increased by media violence 26 , we include youths’ self-reported baseline propensity to respond with anger, measured by the 10-item State-Trait Anger Expression Inventory (STAXI-CA) T-Anger scale (α = 0.86). 27 At the peer level, baseline exposure to externalizing peers was measured by asking youth the number of close friends they had who “have been arrested or done things that could get them in trouble with the police.” 28 At the family level, youth were asked if: “Ever, in real life, have you seen one of your parents get hit, slapped, punched, or beat up by your other parent, or their boyfriend or girlfriend?” 29

Randomly identified adults were emailed a link to a brief online survey that assessed their eligibility. Ineligible adults were thanked for their participation; eligible adults were invited to complete a longer 5-minute survey after obtaining informed consent. They then forwarded their survey link to their child, who provided assent and completed the, on average, 21-minute survey. Youth were encouraged to return to the survey later if they were not in a space where their responses could be kept private from others, including their caregiver.

Data were collected online in 2006 (Wave 1), 2007–2008 (Wave 2), 2008 (Wave 3), 2010–2011 (Wave 4), 2011–2012 (Wave 5), 2012–2013 (Wave 6), and 2016 (Wave 7). In this paper, we examine data from baseline (Wave 1) and five years later (Wave 4, n=887); and baseline and 10 years later (Wave 7, n=779). Incentives were $10 in Wave 1 and increased to $40 in Wave 7. The Wave 1 survey response rate (31%) is consistent with well-conducted surveys using online panels at the time of baseline recruitment. The response rate at Wave 4 was 56% (i.e., 887/1586), and at Wave 7, 49% (i.e., 779/1586).

As the recruitment target, data were weighted statistically to reflect the population of adults with children ages 10 to 15 years old in the United States according to adult age, sex, race/ethnicity, region, education, household income, and child age and sex. Using data collected from random digit dial samples, propensity score weighting also was applied to adjust for adult respondents’ propensity to be online. The weight also adjusted for nonresponse across waves.

Plan of analysis

Rates of within-wave missingness were very low: Race (1.2%) had the highest rate of declination to answer. For all dichotomous variables, “decline to answer” was recoded as “symptom absent” (e.g., not having been in a physical fight). Those who declined to answer the question about race were coded as White, the majority race. For continuous variables, “decline to answer” was recoded to the cohort mean. As a sensitivity analysis, models also were estimated with missing data imputed. Youth who did not respond to Wave 4 or Wave 7, respectively, were excluded from that specific longitudinal analysis.

Analyses were conducted using Stata 15. 30 First, co-relations of violent exposure across media were explored using a correlation matrix and Cronbach’s alpha, which reflects the inter-relatedness of the items. We also examined the percent of youth who reported varying patterns of exposure across media types. Next, to understand the long-term association between media violence and later violent behavior, we first estimated direct, unadjusted logistic regression odds (Model 1). We then estimated logistic regression odds that adjusted for baseline levels of seriously violent behavior, one’s propensity to respond to stimuli with anger, exposure to externalizing peers, exposure to caregiver spousal abuse, sex, age, race, ethnicity, and self-reported dishonesty in answering survey questions (Model 2). For each time point, six unadjusted and adjusted models were estimated: One for violent media diet and five for each of the specific types of violent media of interest.

On average, youth were 12.6 years of age (SE: 0.05) at baseline, 16.7 years of age (SE: 0.07) at 5-year and 22.1 years of age (SE: 0.07) at 10-year follow-up. As shown in Table 1 , those who completed Waves 4 and 7, respectively, versus those who did not, respectively, generally had similar baseline demographic characteristics; exposure to externalizing peers was of exception.

Responses at Wave 1 (baseline) for completers and noncompleters of Wave 4 (5 years) and Wave 7 (10 years), respectively; weighted data

Co-relation of violence exposure across media

The five indicators of exposure to violence in specific media were interrelated: Cronbach’s alpha, Wave 1 = 0.70 (unweighted data given the computation ability of Stata). As shown in Supplemental Table 1 , all media were significantly interrelated. The strongest correlations were noted for violence exposure in television and video games (0.46), and television and music (0.44). Although still significantly interrelated, violence exposure in video games and websites with real people was the least correlated (0.22).

As shown in Figure 1 , more than half (56%) of youth said that none of the media they consumed was mostly violent (i.e., many, almost all, or all of it depicted physical violence).

The percent of youth who reported a specific amount of exposure to physical fighting, shooting, or killing across media types for 10–15-year-olds in the United States (n = 1,586). Different colored bars denote the number of media to which a youth were exposed to a certain level of violence. Five media were aggregated: television, video games, music, websites with real people, and websites with cartoon figures.

Relative odds of seriously violent behavior five years after exposure to youth’s general violent media diet

A factor score was estimated to reflect one’s “violent media diet,” that is, the intensity within and across youth exposures to violent content in five mediums. As shown in Table 2 and Supplemental Figure 1 , the relative odds of reporting seriously violent behavior five years later were 2.45-fold higher ( p <0.001) with each incremental increase in one’s baseline violent media diet. After adjusting for other potentially influential characteristics, the relative odds of seriously violent behavior five years later rose 1.70-fold ( p =0.01) with each incremental increase in one’s violent media diet at baseline.

The relative odds of seriously violent behavior 5 years after exposure to violence by media type and overall media diet, weighted data (n=887)

OR: Odds ratio; aOR: Adjusted odds ratio. Models are adjusted for youth age, sex, race, ethnicity; and baseline seriously violent behavior and exposure to caregiver spousal abuse, propensity to respond to stimuli with anger, externalizing peers, and self-reported honesty in answering survey questions. Bolded text denotes p<0.05; italicized text denotes p<0.20.

Specific types of media also were implicated: Frequent childhood exposure to violence in television (OR = 4.44, p<0.001), music (OR = 5.91, p<0.001), video games (OR=6.73, p<0.001), websites with real people (OR = 2.39, p=0.03) and websites with cartoons (OR = 3.35, p=0.03) each was associated with significantly elevated odds of seriously violent behavior in adolescence. Findings persisted for music (aOR = 3.28, p=0.03), television (aOR = 3.51, p<0.001) and video games (aOR = 3.27, p=0.02) even after adjusting for other childhood influences on violent behavior. Importantly, too, “some” exposure in childhood was associated with seriously violent behavior in adolescence for both music (aOR = 2.34, p=0.05) and video games (aOR = 2.72, p=0.02).

Longitudinal associations a decade later

As shown in Table 3 and Supplemental Figure 1 , the relative odds of seriously violent behavior 10 years after one’s exposure in childhood increased 2.18-fold ( p =0.001) with each incremental increase in one’s violent media diet. After adjusting for other factors, the association persisted (aOR = 1.72, p=0.03). As with adolescence, frequent childhood exposure to violence in music (OR = 4.48, p=0.008), television (OR = 4.26, p=0.001) and video games (OR = 5.38, p=0.001) each were associated with seriously violent behavior in adulthood. This longitudinal association persisted for video games (aOR = 3.28, p=0.03) and television (OR = 3.14, p=0.02) even after taking into account other potentially influential factors; violence depicted in music also was implicated (aOR = 2.85, p=0.13).

The relative odds of seriously violent behavior 10 years after exposure to violence by media type, weighted data (n=760)

RG: Reference group; OR: Odds ratio; aOR: Adjusted odds ratio. Models are adjusted for youth age, sex, race, ethnicity, baseline seriously violent behavior, concurrent propensity to respond to stimuli with anger and self-reported honesty in answering survey questions. Bolded text denotes p<0.05; italicized text denotes p<0.20.

Contrary to other trends observed, exposure to violent websites that depicted cartoons at baseline was associated with lower odds of seriously violent behavior a decade later (aOR = 0.48, p=0.09). Given that this is in the opposite direction of other violent media exposures examined, it seems likely that this may be a statistical anomaly.

Findings were replicated when seriously violent behavior was defined without the measure of sexual assault ( Supplemental Table 2 ), and when missing data were imputed ( Supplemental Table 3 ).

In this national, longitudinal study of children initially 10–15 years of age, findings suggest that exposure to violence in specific mediums and a general diet of violent media across media in childhood are associated with seriously violent behavior in adolescence and adulthood. Measured both in intensity and diversity of exposure, as one’s violent media diet increases incrementally, so too do the odds of seriously violent behavior by 70%, over time. The increased odds are evident even after taking into account other factors that could explain violent behavior later in life, such as one’s violent behaviors in childhood, exposure to caregiver spousal abuse, one’s propensity to respond with anger, and association with peers who engage in activities that could get them in trouble with the police. Pediatricians should work with parents to identify a media consumption plan for their children that is realistic and associated with the least amount of violence as possible across the online, television, game, and music content they consume. Efforts to co-view content and talk with youth about what they are being exposed to in the media they are consuming also are likely useful. 31

Youth do not experience media in a vacuum: Exposure to violence in one medium correlates highly with exposure in another medium. This saturation of messaging may be reinforcing the idea that violence is an appropriate and common tool to address situational anger across environments and stimuli. Understanding how individual types of media are affecting youth behavior is important. Current findings suggest that it may be equally important to understand how influences across media together are affecting behavior. Findings further suggest that early, intense exposure to violence in specific media, namely music, video games, and television, may be related to seriously violent behavior in adolescence and adulthood. There appears to be a stepwise association such that those who report “some” exposure in childhood are differentially at risk than those with more intense (i.e., many, almost all/all) exposures. This suggests that if parents are unable to eliminate their children’s violent media exposure entirely, pediatricians could encourage them to reduce their exposure as much as possible, and that this may still have a positive impact.

Much of the research on exposure to violent media has focused on visual media, such as television, movies, and video games; 17 , 32 , 33 or aggregated exposure across types. 34 Less is known about aural influences, like violent music, although studies exist: In one longitudinal study of adolescents, listening to aggression in music was associated with increased aggression one year later. 35 The current study builds upon this nascent research by noting associations at 5- and 10-years post-exposure, and suggests that more research attention could be focused on the content of the music to which adolescents are listening. Given the ease of digital download of music combined with the widespread ownership of smart phones among today’s adolescents, this exposure may be more hidden and require additional effort by adults to co-experience and manage their children’s consumption.

Limitations

Self-report is a less rigorous measure than objective measures of exposure to violent media. Given the length of the survey and the multitude of questions and topics queried however, it seems unlikely that youth were able to determine the study hypotheses, thereby introducing demand characteristics. Additionally, youth report the intensity of exposure to, and not the amount of time spent with, violent media. For example, some youth who primarily play violent video games may do so for 2 hours a week, whereas others may do so for 40 hours a week. This may result in an underestimate of the association between exposure and behavior. 36

Although community-based research facilitates a wider view into youth behavior than other sources, such as juvenile justice data, self-report is vulnerable to misreporting, particularly of behaviors deemed undesirable. Efforts were made to increase the validity of self-report (e.g., surveying youth online vs in person or over the telephone, reminding them their answers were private, adjusting for self-reported dishonesty in answering survey questions). The inclusion of a social desirability scale might have facilitated a more direct examination of the prevalence and impact of misreporting in the data. That said, one in twenty youth (5%) reported at least one of the seriously violent behaviors queried at baseline. This is generally consistent with base rates observed in other large self-reported surveys, 37 suggesting that under-reporting may not have been an issue in the present study.

Additionally, the multivariate models may be over-adjusting for confounders and report artificially attenuated effect sizes. 38 For example, trait anger can be increased by media violence exposure, 26 and is therefore likely interrelated with media violence exposure. Including trait anger in the multi-variate model, therefore, partially controls for prior effects that this exposure has had on behavior. Also, controlling for prior violent behavior also essentially adjusts for prior predictors of violent behavior. Moreover, youth who consume high levels of media violence may be more likely to spend time with externalizing peers. If true, then the current models may underestimate the association between media violence and violent behavior given that youth who were associating with such peers at baseline were less likely to participate in subsequent waves.

Moreover, although the data are national, they may not be representative. Survey weights were applied to adjust for this possibility. The national reach nonetheless affords a broader view of youth experiences than might not have been observed in a local setting. Moreover, given the study’s focus on mechanism, internal validity is more important than external validity. Finally, rates of attrition are suboptimal, although differential attrition generally was not apparent.

Implications

Since 2006, when baseline data were collected, technology has changed dramatically. A growing body of literature suggests that newer, peer-to-peer, and immersive technologies may positively affect health behavior change. 39 , 40 It stands to reason that a similar learning effect could be observed if content encouraged unhealthy behaviors, including violence. The current study supports this hypothesis with older technology. Future research should both replicate the current study and examine whether newer technologies are associated with an enhanced learning effect.

During childhood, exposure to violence across a variety of media, operationalized as one’s violent media “diet,” appears to be related to engaging in seriously violent behavior in adolescence and adulthood, even beyond one’s propensity to respond to situations with anger, having peers who are engaging in behaviors that could get them in trouble with the police, being exposed to caregiver spousal abuse, and engaging in violent behaviors as a child. Specific exposures to video games and television also appear to be associated with violent behavior over time; similar linkages are suggested for music. While findings should be replicated in other community-based samples, it seems reasonable to suggest that pediatricians might work with parents to identify a media consumption plan that minimizes children’s exposure to violence across media types and is realistic within the family milieu.

Implications and Contribution

In this national, longitudinal study, exposure to violent media at 10–15 years-old was associated with increased odds of seriously violent behavior 5 and 10 years later, adjusting for aggression, externalizing peers, and caregiver spousal abuse. This was true for a general ‘violent media diet;’ and video games, television and music.

Supplementary Material

Acknowledgements:.

We would like to thank the entire Growing up with Media study team from the Center for Innovative Public Health Research, Princeton Survey Research Associates International, Harris Interactive, Johns Hopkins Bloomberg School of Public Health, and the Centers for Disease Control and Prevention, who contributed to different parts of the planning and implementation of the study. Finally, we thank the families and youth for their time and willingness to participate in this study.

Funding/support

Research reported in this publication was supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health under Award Number R01HD083072, and by the Centers for Disease Control and Prevention under Award Numbers U49 CE000206; R01 CE001543. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the Centers for Disease Control and Prevention. Neither funder was involved in data analysis or manuscript preparation.

Conflict of interest disclosure:

The authors have no conflicts of interest to declare.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Contributor Information

Michele L. Ybarra, Center for Innovative Public Health Research, San Clemente, California.

Kimberly J. Mitchell, Crimes against Children Research Center, University of New Hampshire, Durham, New Hampshire.

Jay Koby Oppenheim, Independent Consultant, New York, New York.

The people's place in the city of bits and atoms

via Archinect

March 6, 2024

• #artificial intelligence
• #urban planning
• #sustainability
• Naroa Coretti Sánchez Research Assistant
• Ainhoa Genua Cervino
• Persuasive Electric Vehicle (PEV)
• City Science Network
• The MIT Autonomous Bicycle Project
• Autonomous micro-mobility simulation study
• Autonomous micro-mobility for food deliveries
• Media Lab Research Theme: Future Worlds
• Media Lab Research Theme: Life with AI

Share this article

By Niall Patrick Walsh

The city of the 21st century represents a confluence of bits and atoms; an organism in its own right that relentlessly spawns information and data about itself, its people, and the invisible flows that support them. What is the relationship between humans and the city in this new condition? What is its future? To explore these questions, we speak with architect, TED founder, and father of information architecture Richard Saul Wurman, 2025 Venice Biennale curator Carlo Ratti, and MIT Media Lab researchers Naroa Coretti and Ainhoa Genua.

Understanding cities

In 1976, Richard Saul Wurman chaired the national AIA Convention in Philadelphia. Operating under the convention theme ‘The American City: The Architecture of Information,’ Wurman set out a vision for cities built upon the relationship between urbanism, information, and people.

“Wouldn’t a city – any city – be more useful and more fun if everybody knew what to do in it, and with it?” the conference brochure asked. “As architects, we know it takes more than good-looking buildings to make a city habitable and usable. It takes information: information about what spaces do as well as how they look; information that helps people articulate their needs and respond to change. That’s what Architecture of Information is all about.”

Almost fifty years later, in a recent conversation with Wurman, I returned to the 1976 brochure with a mission of exploring how prevailing approaches to urban planning have, or have not, lived up to Wurman’s vision. As someone whose lifelong pursuit of ‘understanding’ includes founding the TED conference, pioneering the field of Information Architecture, and formulating organizational theories such as LATCH and A-NOSE, it is perhaps no surprise that Wurman’s views on urbanism emphasize the potential for cities as places of learning and understanding.

Best New Ideas in Money: The road ahead for public transit

A discussion of the future of public transportation, with a highlight of the Autonomous Bicycle project led by PhD student Naroa Coretti.

Could self-driving bikes change how we use cities?

Kent Larson, head of the Media Lab’s City Science group, and PhD student Naroa Coretti Sánchez talk about the group's work and philosophy.

What is the MIT Autonomous Bicycle Project?

The MIT Autonomous Bicycle Project proposes that autonomous bicycles could serve as an alternative to current bike sharing models.

City Science team publishes two papers in Communications in Transportation Research

The two papers relate to project work from the City Science group around the impact of shared autonomous micro-mobility systems.

Help | Advanced Search

Computer Science > Computer Vision and Pattern Recognition

Title: mm1: methods, analysis & insights from multimodal llm pre-training.

Abstract: In this work, we discuss building performant Multimodal Large Language Models (MLLMs). In particular, we study the importance of various architecture components and data choices. Through careful and comprehensive ablations of the image encoder, the vision language connector, and various pre-training data choices, we identified several crucial design lessons. For example, we demonstrate that for large-scale multimodal pre-training using a careful mix of image-caption, interleaved image-text, and text-only data is crucial for achieving state-of-the-art (SOTA) few-shot results across multiple benchmarks, compared to other published pre-training results. Further, we show that the image encoder together with image resolution and the image token count has substantial impact, while the vision-language connector design is of comparatively negligible importance. By scaling up the presented recipe, we build MM1, a family of multimodal models up to 30B parameters, including both dense models and mixture-of-experts (MoE) variants, that are SOTA in pre-training metrics and achieve competitive performance after supervised fine-tuning on a range of established multimodal benchmarks. Thanks to large-scale pre-training, MM1 enjoys appealing properties such as enhanced in-context learning, and multi-image reasoning, enabling few-shot chain-of-thought prompting.

Submission history

Access paper:.

• Download PDF
• Other Formats

References & Citations

• Google Scholar
• Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

• Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .