Video Game Use as Risk Exposure, Protective Incapacitation, or Inconsequential Activity Among University Students
Comparing Approaches in a Unique Risk Environment
Abstract
Abstract. While there is extensive literature exploring the possible negative effects of video games, and many such studies using college student samples, there is little research on how video game use impacts the unique risk environment of college students. This study focuses on the unique risk aspects of the college and university environment with a preregistered survey comparing three competing models of video games’ possible role (games as risk, incapacitation, or inconsequential) in predicting alcohol and substance use, sexual risk, interpersonal violence, bullying victimization, suicide, disordered eating, and exercise to provide a baseline measure of what role, if any, video games play in the college and university risk environment. Video game play was most consistently associated with outcomes related to suicide and interpersonal violence, and more sporadically associated with some other outcomes.
One of the most prolific and prominent topics in media psychology scholarship over the past few decades has been the effects of video games on their users. Hundreds of studies have examined questions such as whether violent video games increase users’ aggressive behavior or whether video games are addictive for their users. Much of the research investigating the potential impact of video games on their users has involved either laboratory experiments or surveys. The former have tested causal links between video game play and artificial measures conceptually related to more important real-world outcomes, such as the effects of violent video games on administration of noise blasts to a confederate (see Elson, Mohseni, Breuer, Scharkow, & Quandt, 2014). The latter have tended to examine correlations between self-reported video game play and self-reported outcomes such as school performance (e.g., Gentile, Choo, Liau, Sim, Li, & Fung, 2011) and risk behavior (e.g., Hull, Brunelle, Prescott, & Sargent, 2014).
It suffices here to say that the results and interpretation of all of this research are mixed. Opinions vary widely among scholars about the strength of evidence for negative effects of video games, so much so that there is even animated disagreement about the extent of the existing scholarly consensus on the issue (see Bushman, Gollwitzer, & Cruz; Griffiths et al., 2016; Ivory et al., 2015; Petry et al., 2014; Quandt et al., 2015). Many of these disagreements revolve around the extent to which a host of methodological issues affect the validity and generalizability of research in the area. One of many such issues is the heavy use of college student samples in research on games’ effects (Ferguson & Olson, 2014), particularly given that much of the public discussion about societal effects of video games concerns children. Interestingly enough, however, few of these studies of college students address the unique health risks of that population, with studies instead generalizing from their postsecondary student samples to broader societal groups.
This is a missed opportunity, considering that (a) at least some video game use is nearly universal among college and university students and a majority of them play regularly (Jones, 2003), and (b) several risk behaviors are known to be particularly prevalent among college and university students – in some cases higher than the general population despite college and university students’ youthful age and low prevalence of many risk factors for poor health (e.g., low socioeconomic status; Centers for Disease Control & Prevention, 1997). It is possible that the social effect of video games on college students’ health is negligible among other risk factors, as has been argued about the effect of video games in general (e.g., Ferguson & Olson, 2014). It is also possible that some video games are a contributing factor to certain unhealthy behaviors among college students, as has also been argued in the case of other populations (e.g., Gentile & Bushman, 2012; Hull et al., 2014). A third possibility, given that exposure to the social environment of colleges and universities introduces a constellation of health risks (Weitzman, Nelson, & Wechsler, 2003), is that time spent playing video games is a protective factor against campus risks simply because spending time with games causes users to spend less time in high-risk settings. This incapacitation approach has been examined in video game sales and crime data (Cunningham, Engelstätter, & Ward, 2016) but has been somewhat unexplored in laboratory and survey research.
This study’s purpose is twofold: (a) to provide baseline survey data to indicate what potential role video games may have in the unique health risk environment of college and university campuses, and (b) to test the competing predictions of the risk, incapacitation, and inconsequential approaches to the possible role of video games in the postsecondary risk environment.
Literature Review
Health Risks in Colleges and Universities
Health risks can be influenced by a range of factors, which include not only individuals’ overt risk-taking behaviors but also circumstances that expose individuals to the risk of victimization by others. A commonly identified risk behavior among college and university students is abuse of alcohol and other substances, which is itself a risk and is also understood to be an important contributor to other associated health risks among students (Weitzman et al., 2003). Campus alcohol use is associated with negative outcomes for both drinkers and nondrinkers, and is implicated as a factor in injuries, fights, sexual assaults, and risky sexual behavior (Abbey, 1991; Cooper, 2002) – all of which are behaviors identified as substantial risks for college students (Centers for Disease Control and Prevention, 1997). Health risks such as tobacco use, drug abuse, eating disorders, and decline in physical activity are often adopted in college, bullying remains prevalent in college, and college students attempt suicide at a higher rate than the general population (Chappell et al., 2004; O’Neill, 2007; Taylor et al., 2006).
Video Games as Added Risk, Incapacitating Distraction, or Inconsequential Activity in the Postsecondary Environment
The friction between different findings and interpretations of the literature about video game effects complicates predictions of what role, if any, video game use might play in the unique problems of the college and university risk environment. For several of the most prominently identified health risks in college, competing predictions are plausible based on three competing conceptual mechanisms:
- 1.Video game exposure as an added risk influencing negative outcomes along with other factors, whether because of the effects of message content or the effects of game play setting, as predicted by popular models of learned negative game effects (Gentile & Bushman, 2012; Hull et al., 2014).
- 2.Time spent with video games as an incapacitating distraction from the dangerous elements of the university social environment, as predicted by economic approaches to video game use as time displacement from other activities (Cunningham et al., 2016).
- 3.Video game use as an inconsequential activity not uniquely influential in the face of greater risk factors, as predicted by models focusing on behavior as a product of biology and long-standing traits with minimal influence of short-term environmental triggers (e.g., Ferguson & Olson, 2014).
Tobacco, Alcohol, and Substance Use
In the case of abuse of alcohol and other substances, this friction between these three competing approaches to video game effects is apparent. There are findings indicating that video games and media that idealize and celebrate risk may promote risk-taking among users (e.g., Hull et al., 2014); this effect is not unique to oft-violent action games (a genre that includes first-person shooters and fighting games), but has been observed with racing games as well (Fischer, Greitemeyer, Kastenmüller, Vogrincic, & Sauer, 2011). Additionally, playing games, whether alone or with friends, may co-occur with substance abuse. Meanwhile, however, the incapacitation argument that time spent playing games may be time not spent in the alcohol-drenched college social environment, as well as the inconsequential argument that game effects are not consistently observed on meaningful outcomes, both remain relevant to alcohol and substance abuse – although for these approaches, game content and genre are not important to effects. Therefore, we propose competing hypotheses based on the competing approaches in turn:
Overall video game use, action game use, and sports and racing game use will be positively associated with use of tobacco, alcohol, and other substances (risk).
Overall video game use will be negatively associated with use of tobacco, alcohol, and other substances (incapacitation).
Overall video game use, action game use, and sports and racing game use will not be associated with use of tobacco, alcohol, and other substances (inconsequential).
Sexual Risk
For sexual risk, the risk approach again predicts genre-specific increases in risk, as well as effects of overall game use given the prevalence of risk-taking in popular video games (Hull et al., 2014), the incapacitation approach predicts broad reduction in risk, and the inconsequential approach predicts that video game play will be irrelevant:
Overall video game use, action game use, and sports and racing game use will be positively associated with sexual risk (risk).
Overall video game use will be negatively associated with sexual risk (incapacitation).
Overall video game use, action game use, and sports and racing game use will not be associated with sexual risk (inconsequential).
Interpersonal Violence
The same thread of competing hypotheses is plausible for interpersonal violence, although with a unique focus on violent action games given the literature’s focus on aggression as a possible effect of exposure to violent games (e.g., Bushman et al., 2015; Hull et al., 2014). Here, the risk approach predicts unique effects of oft-violent action video game use, although the generally high prevalence of violence in video games would be consistent with the risk prediction as well:
Overall video game use and action game use will be positively associated with carrying weapons and fighting (risk).
Overall video game use will be negatively associated with carrying weapons and fighting (incapacitation).
Overall video game use, action game use, and sports and racing game use will not be associated with carrying weapons and fighting (inconsequential).
Bullying Victimization
While bullying victimization is a concern on college and university campuses, it is less clear what the competing approaches might predict in terms of game effects given that bullying victimization is not instigated by the user. Therefore, we ask:
Are overall video game use, action game use, and sports and racing game use associated with bullying victimization?
Suicide
It is also unclear how suicide can be interpreted as a risk affected by game messages or time displacement. Therefore, we ask:
Are overall video game use, action game use, and sports and racing game use associated with suicide risk?
Disordered Eating
Evidence for the presence of idealized body images in video game characters (Lynch, Tompkins, van Driel, & Fritz, 2016) suggests a risk prediction that overall video game use would be associated with disordered eating. However, the incapacitation hypothesis is relevant if video game use distracts from unhealthy social influences, and the inconsequential approach also remains plausible:
Overall video game use will be positively associated with disordered eating (risk).
Overall video game use will be negatively associated with disordered eating (incapacitation).
Overall video game use will not be associated with disordered eating (inconsequential).
Exercise
In the case of exercise, the incapacitation hypothesis loses relevance as time displacement could detract from exercise opportunity. Therefore, only a risk hypothesis of game time detracting from exercise and an inconsequential hypothesis are proposed:
Overall video game use will be negatively associated with exercise (risk).
Overall video game use will not be associated with exercise (inconsequential).
Method
Design
An online survey of full-time students enrolled at colleges and universities in the United States measured participants’ age, sex, parents’ education level, self-reported overall video game use, video game use by genre, and risk-related behaviors pertaining to tobacco, alcohol, and substance use, sexual risk, interpersonal violence, bullying victimization, suicide, disordered eating, and exercise.
Participants
Participants were recruited for this study using Amazon’s Mechanical Turk crowdsourcing Internet marketplace, which has been found to contain a disproportionately high prevalence of college students (e.g., Levay, Freese, & Druckman, 2016; Ross, Irani, Silberman, Zaldivar, & Tomlinson, 2010). Recruiting materials indicated that only full-time college and university students 18 years and older were eligible to participate, and participants were asked to confirm their full-time enrollment status and list the college or university they were attending. To limit irrelevant, spurious, and careless responses, eligibility criteria based on Mechanical Turk participation history from Levay et al. (2016) were applied, and data from participants who provided disqualifying (e.g., not a full-time student) or impossible responses (e.g., smoking cigarettes 40 days per month), or who completed the instrument in fewer than 5 min, were excluded from analyses. Given the results of an a priori power analysis (multiple regressions with three predictors including two controls; f2 = .02 with power = .80; see Analysis Strategy section), the targeted number of valid participants was n = 543. Recruitment was conducted over a period of about 36 hr until the target number was reached after eliminating disqualified respondents. The resulting final sample comprised 553 respondents included in analyses (of 1,313 total respondents). Respondents were paid US $1. Respondents were 57.3% male (n = 317) with a median age of 23 years (M = 25.02, SD = 5.67). Respondents’ ethnic make-up was 68.2% White (n = 377), 13.6% Black, not Hispanic (n = 75), 8.7% Hispanic or Latino (n = 48), 6.5% Asian or Pacific Islander (n = 36), 1.4% American Indian or Alaskan Native (n = 8), and 1.6% other races (n = 9). Compared with the US population of postsecondary students, the sample of participants is proportionally more male, slightly more White, and older (National Center for Educational Statistics, 2016). Participants reported spending a mean of 6.41 hr per week (SD = 7.79) playing video games, with 84.45% (n = 467) reporting at least 1 hr of play per week.
Questionnaire Instrument and Measures
The Qualtrics online survey tool was used to deliver a questionnaire to participants. Measures of video game use and demographics are original. Measures of risk behaviors and parents’ education level were adapted from the Centers for Disease Control and Prevention’s (1997) National College Health Risk Behavior Survey (NCHRBS) and the American College Health Association’s (2015) National College Health Assessment.
Overall Weekly Video Game Use
One question asked how many hours participants spend playing video games (including games played on a game console, computer, or mobile device) in a typical week at college.
Weekly Video Game Use by Genre
A subsequent question presented only to participants who reported some weekly video game use measured the proportion of time spent in a typical week at college playing games from each a short and broad list of genres: action, role-playing, simulation, strategy, sports (including racing), and puzzle and trivia. Percentage responses for each genre were multiplied by overall weekly video game use to produce weekly video game use measures for each genre.
Tobacco, Alcohol, and Other Substance Use
In all, 12 questions asked how many times in a typical month at college participants (a) smoke cigarettes, (b) smoke electronic cigarettes, (c) use chewing tobacco or snuff, (d) have at least one drink of alcohol, (e) have five or more drinks of alcohol in a row, within a few hours, (f) use marijuana, (g) use any form of cocaine including powder, crack, or freebase, (h) use inhalants, (i) use other illegal drugs, (j) use steroids without a prescription, (k) take prescription drugs that were not prescribed to them, and (l) ride in a vehicle driven by someone who has been drinking alcohol.
Sexual Risk
Questions assessed general sexual risk behavior and sexual assault risk exposure specifically. One question asked the number of sexual partners with whom participants have sexual intercourse, oral sex, or anal sex in a typical month at college. Three questions asked how many times in a typical month at college participants have (a) sexual intercourse, oral sex, or anal sex, (b) sexual intercourse or anal sex using a condom, and (c) sexual intercourse or anal sex without using a condom. One question asked how many times in the past year while at college participants have become pregnant or gotten someone pregnant. Two questions asked how many times in the past year participants (a) had someone have sexual intercourse, oral sex, or anal sex with them without their permission or when too intoxicated to provide consent, and (b) had sexual intercourse, oral sex, or anal sex without being sure their partner gave permission or when the partner may have been too intoxicated to provide consent.
Fighting
Three questions asked how many times in the past year while at college participants have (a) carried a weapon such as a gun, knife, or club (not for work), (b) been in a physical fight, and (c) been in a physical fight in which they were injured and had to be treated by a doctor or nurse.
Bullying Victimization
Two questions asked how many times in the past year while at college participants have been (a) bullied physically, and (b) bullied electronically online.
Suicide
Two questions asked how many times in the past year while at college participants have (1) seriously considered suicide and (b) attempted suicide.
Disordered Eating
One question asked for participants’ perceptions of their weight (1 = very underweight, 3 = about the right weight, 5 = very overweight). Three questions asked how many times in a typical month at college participants (a) diet to lose weight or keep from gaining weight, (b) vomit or take laxatives to lose weight or keep from gaining weight, and (c) take diet pills to lose weight or keep from gaining weight.
Exercise
Four questions asked how many days in a typical month at college participants (a) participate in sports activities for at least 20 min that made them sweat or breathe hard, (b) do stretching exercises, (c) do strength exercises, and (d) walk or bicycle for at least 30 min at a time.
Demographic Measures
Three questions asked participants’ age, sex, and race. Sex was used as a control variable in analyses.
Parents’ Education
To control for an indirect indicator of socioeconomic status, two questions asked the education level of participants’ mother and father with four responses ranging from did not finish high school to graduated from college and a not sure option. The variable was treated as continuous in analyses with responses scored 1–4 and averaged across the two measures (Cronbach’s α = .671). Respondents who selected the not sure option for both parents did not receive a score.
Other Questions
Questions measuring class in school, television use, reading for school, and reading for pleasure were included, but not analyzed.
Analysis Strategy
While some outcome measures were conceptually related (e.g., three measures of tobacco use), they were not considered unidimensional enough to be examined in indices, and normality of data was not expected to be consistent across measures. Therefore, analyses were conducted individually for each measure. Each of a series of multiple regression analyses included the control variables of sex and parents’ education, the appropriate video game use predictor variable, and the appropriate outcome measure entered as the dependent measure.
For each of the risk hypotheses, a regression equation included the control measures and overall game use measure predicting each outcome measure, then with analyses repeated for any predicted game genre measure (e.g., action games and sports and racing games for H1a and H2a, action games for H3a). The incapacitation hypotheses were tested with the same regression analyses for overall game use, and the inconsequential hypotheses were tested with the same series of regressions for overall game use as well as additional regressions for the same game use genres predicted for the corresponding risk hypotheses. Corrections to critical alpha values were made to adjust for the number of measures included for each conceptual outcome variable using the Bonferroni method. Specifically, the traditional critical alpha of p < .05 was divided by the number of outcome measures examined for each conceptual variable (3 for tobacco use, 3 for alcohol use, 6 for other substance use, 5 for general sexual behavior, 2 for sexual assault, 3 for fighting, 2 for bullying, 2 for suicide, 4 for disordered eating, and 4 for exercise).
While hierarchical OLS regression may be appropriate for variables with normally distributed data, event count data are likely to be positively skewed and better fit by the Poisson regression model, which is modeled on the Poisson distribution rather than the normal distribution. That said, the Poisson model assumes the variance and mean of data to be equal, and sometimes fits event count data poorly. In such cases, the negative binomial regression model, which allows for estimation of mean and variance independently, may provide a superior fit. Finally, some count variables have a disproportionately high number of zero scores because certain circumstances may lead some cases to be in a structural zero group where the event never occurs (e.g., committed nondrinkers) while other circumstances predict event counts among cases where the event may occur (e.g., factors influencing number of drinks for sporadic or heavier drinkers). For these variables, zero-inflated Poisson regression models and zero-inflated negative binomial models create models of two processes for a given outcome: likelihood of zero or nonzero outcomes using logistic regression, and counts for the outcome using either Poisson or negative binomial regression, respectively (see Atkins & Gallop, 2007).
Given that some event count outcome measures were not expected to be normally distributed, an a priori analysis plan was developed to account for skewness. If skewness of the outcome measure did not exceed a threshold of ±2, a hierarchical OLS regression was used for that outcome measure with the control measures entered in the first step and the predictor measure added in the second step. If skewness did exceed ±2, then a series of regression models (each increasingly deviant from assumptions of normality and dispersion) were conducted and tested for fit in this order: Poisson model, negative binomial model, zero-inflated Poisson model, and zero-inflated negative binomial model. However, sparseness of data for some counts led all Poisson and negative binomial models to produce Hessian matrix singularity, yielding unstable parameter estimates and unreliable model fit data. Additionally, some zero-inflated negative binomial models did not produce results as optimization failed to converge. Thus, zero-inflated Poisson models with the control variables as covariates are reported for all analyses of outcome variables with skewness exceeding ±2.
Results
Tobacco, Alcohol, and Other Substance Use
Study data and materials are available at https://osf.io/evknr/?view_only=f5d377bfa06b47fda1354e5b33737b51. H1a predicted that overall video game use, action game use, and sports and racing game use would be positively associated with use of tobacco, alcohol, and other substances (risk), while H1b predicted that overall video game use would be negatively associated with substance use (incapacitation), and H1c predicted that overall video game use, action game use, and sports and racing game use would not be associated with substance use (inconsequential).
Tobacco
See Table 1 for a list of all study outcome measures, associated descriptive statistics, and a summary regression analysis results. A series of nine zero-inflated Poisson regression model equations were conducted with each combination of the three weekly game play variables predicting the three skewed (> ±2) tobacco use variables with the two control measures as covariates and a critical alpha value of .0167. Weekly action game play was a significant predictor in the zero-inflation model for e-cigarette use (coefficient estimate = −.069, SE = .027, z = −2.529, p = .011), as well as the count model for chewing tobacco use (coefficient estimate = −.073, SE = .024, z = −3.080, p = .002). No other coefficients were significant (p > .0167). Overall, results for tobacco use favor confirmation of H1c (inconsequential), except that action game play was associated with a decreased likelihood of never smoking e-cigarettes in a month (structural zero group) in partial support of H1a (risk). Action game play was associated with fewer days per month using chewing tobacco, partially consistent with H1b (incapacitation) but the predicted mechanism of H1b was not genre-specific.
Alcohol
A series of three hierarchical linear multiple regression equations and six zero-inflated Poisson regression equations were conducted with each combination of the three weekly game play variables predicting the one nonskewed (< ±2) and two skewed (< ±2) alcohol use variables with the two control measures as covariates and a critical alpha value of .0167. Weekly sports and racing game play was a predictor in the count model for having at least five drinks in a sitting (coefficient estimate = .091, SE = .017, z = 5.406, p < .001), as well as the zero-inflation model for riding with a drunk driver (coefficient estimate = −.316, SE = .087, z = −3.615, p < .001). No other coefficients were significant (p > .0167). Overall, results for alcohol use favor confirmation of H1c (inconsequential), except that sports and racing game play was associated with more days per month having five or more drinks in a sitting and a decreased likelihood of never riding with a drunk driver in a month in partial support of H1a (risk).
Other Substances
A series of 18 zero-inflated Poisson regression model equations were conducted with each combination of the three weekly game play variables predicting the six skewed (> ±2) substance use variables with the two control measures as covariates and a critical alpha value of .0083. Overall weekly game play was a significant predictor in the count model for using other illegal drugs (coefficient estimate = .046, SE = .015, z = 3.107, p = .002) Weekly action game play was a significant predictor in the count model for inhalant use (coefficient estimate = 2.579, SE = .855, z = 3.016, p = .003). Weekly sports and racing game play was a significant predictor in the count model for inhalant use (coefficient estimate = −2.577, SE = .728, z = −3.541, p < .001). No other coefficients were significant (p > .0083). Overall, results for other substance use favor confirmation of H1c (inconsequential), except that overall game play was associated with more days per month using illegal drugs and action game play was associated with more days per month sniffing inhalants, both in partial support of H1a (risk). Sports and racing game play was associated with fewer days sniffing inhalants, partially consistent with H1b (incapacitation), but the predicted mechanism of H1b was not genre-specific.
Sexual Risk
H2a predicted that overall video game use, action game use, and sports and racing game use would be positively associated with sexual risk (risk), while H2b predicted that overall video game use would be negatively associated with sexual risk (incapacitation), and H2c predicted that overall video game use, action game use, and sports and racing game use would not be associated with sexual risk (inconsequential).
General Sexual Behavior
A series of 15 zero-inflated Poisson regression model equations were conducted with each combination of the three weekly game play variables predicting the five skewed (> ±2) general sexual behavior variables with the two control measures as covariates and a critical alpha value of .01. Overall weekly game play was a significant predictor in the count model for number of times per month having sexual intercourse (coefficient estimate = .011, SE = .002, z = 5.991, p < .001), as well as the count model for number of times per month having sexual intercourse without a condom (coefficient estimate = .015, SE = .002, z = 7.158, p < .001). Weekly action game use was a significant predictor in the count model for number of times per month having sex without a condom (coefficient estimate = .036, SE = .005, z = 6.606, p < .001). Weekly sports was a significant predictor in the count model for number of sexual partners per month (coefficient estimate = .119, SE = .022, z = 5.293, p < .001), as well as the count model for number of times having sex per month (coefficient estimate = .049, SE = .011, z = 4.469, p < .001). No other coefficients were significant (p > .01). Overall, results for general sexual behavior favor confirmation of H2c (inconsequential), but several findings partially support H2a (risk): Overall game play was associated with having sex more times per month and having sex without a condom more times per month, while action game play was also associated with having sex more times without a condom per month and sports and racing game play was associated with more sexual partners per month and having sex more times per month.
Sexual Assault
A series of six zero-inflated Poisson regression model equations were conducted with each combination of the three weekly game play variables predicting the two skewed (> ±2) sexual assault variables with the two control measures as covariates and a critical alpha value of .025. Overall weekly game play was a significant predictor in the count model for number of times in the past year being sexually victimized without giving consent (coefficient estimate = .052, SE = .017, z = 2.946, p = .003). Weekly sports and racing game play was a significant predictor of the count model for number of times in the past year being sexually victimized without giving consent (coefficient estimate = .344, SE = .126, z = 2.737, p = .006), and weekly sports was a significant predictor of the count model for number of times in the past year having sex with others without obtaining consent (coefficient estimate = .391, SE = .131, z = 2.990, p = .003). No other coefficients were significant (p > .025). Overall, results for sexual assault are mixed. The findings from half the analyses partially support H1c (inconsequential), while the other half partially support H1a (risk): Overall weekly game play and sports and racing game play were associated with being sexually victimized without consent more often in the past year, while sports and racing game play was also associated with having sex with others without obtaining consent more often in the past year.
Interpersonal Violence
H3a predicted that overall video game use and action video game use would be positively associated with carrying weapons and fighting (risk), while H3b predicted that overall video game use would be negatively associated with carrying weapons and fighting (incapacitation), and H3c predicted that overall video game use and action game use would not be associated with carrying weapons and fighting (inconsequential). A series of six zero-inflated Poisson regression model equations were conducted with each combination of the two relevant weekly game play variables predicting the three skewed (> ±2) interpersonal violence variables with the two control measures as covariates and a critical alpha value of .0167. Overall weekly game play was a significant predictor in the count model for number of times carrying a weapon in the past year (coefficient estimate = .006, SE = .002, z = 2.957, p = .003), as well as the count model for number of times in the past year getting in a physical fight (coefficient estimate = .027, SE = .010, z = 2.616, p = .009), and the count model for number of times in the past year getting in a physical fight where someone needed medical attention (coefficient estimate = .069, SE = .019, z = 3.670, p < .001). Weekly action game play was also a significant predictor of the count model for times carrying a weapon in the past year (coefficient estimate = .081, SE = .004, z = 18.746, p < .001), and the count model for getting in a physical fight where a participant needed medical attention (coefficient estimate = .119, SE = .034, z = 3.490, p < .001.) No other coefficients were significant (p > .0167). Overall, results for interpersonal violence support H3a (risk), with overall game play associated with increased frequency of carrying a weapon, getting in a fight, and getting in a fight involving hospitalization in the past year, and action game play associated with increased frequency of carrying a weapon and getting in a fight involving hospitalization in the past year.
Bullying Victimization
RQ1 asked if overall video game use, action game use, and sports and racing game use are associated with bullying victimization. A series of six zero-inflated Poisson regression model equations were conducted with each combination of the three weekly game play variables predicting the two skewed (> ±2) bullying variables with the two control measures as covariates and a critical alpha value of .025. Overall weekly game play was a significant predictor in the count model for number of times bullied online in the past year (coefficient estimate = .031, SE = .002, z = 13.259, p < .001). Weekly action game play was a significant predictor of the count model for number of times bullied online in the past year (coefficient estimate = −.051, SE = .007, z = −7.590, p < .001). No other coefficients were significant (p > .025). Overall, results for bullying are mixed. Overall weekly play is associated with more frequent bullying victimization online, while action game play is associated with less frequent bullying victimization online, and neither game play measure is associated with risk of physical bullying.
Suicide
RQ2 asked if overall video game use, action game use, and sports and racing game use are associated with suicide risk. A series of six zero-inflated Poisson regression model equations were conducted with each combination of the three weekly game play variables predicting the two skewed (> ±2) suicide variables with the two control measures as covariates and a critical alpha value of .025. Overall weekly game play was a significant predictor in both the count model (coefficient estimate = .046, SE = .003, z = 17.576, p < .001) and the zero-inflation model (coefficient estimate = −.033, SE = .014, z = −2.367, p = .018) for number of times suicide was considered in the past year, and of the count model for number of times attempted suicide in the past year (coefficient estimate = .102, SE = .020, z = 5.089, p < .001). Weekly action game play was a significant predictor of the count model for number of times suicide was considered in the past year (coefficient estimate = .245, SE = .010, z = 23.856, p < .001), and of the count model for number of times suicide was attempted in the past year (coefficient estimate = .779, SE = .154, z = 5.068, p < .001). Weekly sports and racing game play was a significant predictor of the count model for number of times suicide was considered in the past year (coefficient estimate = −.288, SE = .085, z = −3.371, p < .001). No other coefficients were significant (p > .025). Overall, results for suicide risk suggest relationships with video game play. Overall weekly play is associated with more frequently considering suicide, a lower likelihood of never considering suicide in a year, and more frequently attempting suicide, and weekly action play is associated with more frequently considering and attempting suicide. By contrast, playing sports and racing games is associated with less frequently considering suicide.
Disordered Eating
H4a predicted that overall video game use would be positively associated with disordered eating (risk), while H4b predicted that overall video game use would be negatively associated with disordered eating (incapacitation), and H4c predicted that overall video game use would not be associated with disordered eating (inconsequential). A series of two hierarchical linear multiple regression equations and two zero-inflated Poisson regression equations were conducted with overall weekly game play predicting the two nonskewed (< ±2) and two skewed (< ±2) disordered eating variables with the two control measures as covariates and a critical alpha value of .0125. Overall video game play was a significant predictor of weight (B = .013, SE = .004, β = .138, t = 3.235, p = .001, R2 change = .018), as well as of the count model for days per month vomiting or taking laxatives (coefficient estimate = −.103, SE = .029, z = −3.576, p < .001), and the count model for days per month taking diet pills (coefficient estimate = −.025, SE .007, z = −3.466, p < .001). No other coefficients were significant (p > .0125). Overall, results for disordered eating tend to support H4c (incapacitation) with overall weekly play associated with fewer days vomiting or taking laxatives and fewer days taking diet pills. The finding that overall video game play is associated with higher weight may be ambiguous, as it may indicate a tendency toward greater likelihood of being overweight in support of H4a (risk) or a decreased tendency to be underweight in support of H4b (incapacitation).
Exercise
H5a predicted that overall video game use would be negatively associated with exercise behavior (risk), while H5b predicted that overall video game use would not be associated with exercise behavior (inconsequential). A series of four hierarchical linear multiple regression equations were conducted with overall weekly game play predicting the four nonskewed (< ±2) exercise variables with the two control measures as covariates and a critical alpha value of .0125. None of the video game use measures were a significant predictor of any of the exercise measures (p > .0125). Results support H5c (inconsequential).
Discussion
This study provides baseline information about the role of video games in an understudied set of risk factors – specifically, health risks prevalent in postsecondary student populations. The unique risk circumstances of college and university students merit attention given the population’s prevalent video game use and disproportionately high rate of some negative health outcomes. While correlational in nature, and thus not appropriate evidence for causal claims, this study provides a first step toward understanding whether video games may have a meaningful harmful or protective role in the college risk environment.
It is also important that research investigating video game use and behavioral outcomes be conducted using preregistration and open science practices. Even though this study cuts a broad conceptual swath by analyzing many potential outcomes and competing conceptual predictions, preregistered research with open data is an effective way to extend knowledge by ensuring that procedures are determined a priori rather than conducted or reported flexibly or selectively with prejudice toward one mechanism or outcome versus another. With the attention of the public and policymakers on researchers’ findings in this area, there is too much at stake for investigations to proceed any other way.
Our findings suggest that video game use may be a predictor of outcomes related to suicide and interpersonal violence, as well as for unprotected sex and sexual assault as predicted by the risk approach. It may be that video game play influences these behaviors, or that video game play is a marker of lifestyle and health status factors associated with these behaviors. In either case, more research is needed. If a strong body of further preregistered research replicates our findings of video game use variables predicting interpersonal violence, suicide, unprotected sex, and sexual assault (whether causally or as a lifestyle marker) colleges, universities, campus organizations, and video game producers may do well to consider targeting video game enthusiasts with preventative campaigns.
Video game play was largely unrelated with other outcomes analyzed here, most consistently exercise behavior and use of tobacco and other substances, in support of the inconsequential approach. This study adds little to speculation about video games influencing these behaviors, although more studies should be conducted before firm conclusions are drawn. Further, while there was evidence that video game play was associated with higher weight – which may conflate increased risk of being overweight with decreased risk of being underweight – video game use was associated with reduced rates of some disordered eating behavior in a manner consistent with the protective incapacitation approach. It may be useful to explore what mechanism may account for video game play as a marker for reduced disordered eating behavior. For other outcomes, there was little evidence for the incapacitation approach.
Despite the advantages of the preregistered approach of this study and its openly available data for further analysis, the study also has many limitations. This study cannot imply causal directions, nor does it isolate a conceptual mechanism by which video game use might be associated with outcomes. Also, the strength and robustness of observed associations in a practical setting is difficult to assess from one cross-sectional survey with limited covariates. Aside from the usual limitations involved with self-report data, this study’s use of the Mechanical Turk platform may affect its representativeness. While college and university students are plentiful among Mechanical Turk users, they may be different from the general college and university student population in many ways. This study should thus be replicated with other samples of postsecondary student populations. Further, the arbitrary 5-min minimum completion time for included responses may have been too conservative given the number of disqualified responses. While preregistered disqualification criteria are preferable to flexible criteria conducive to a “fishing expedition” in search of preferred findings, consistent best practices for a priori criteria can be developed as more preregistered studies are conducted. Several other possible control measures were not captured. Finally, many measures were imprecise. For example, the broad sports and racing category, measured in the questionnaire as sports including racing, was more general than measures in previous research targeting racing games as a specific predictor, and the measure of frequency of unprotected sex did not distinguish unprotected sex between committed partners from unprotected sex in riskier settings.
The risks facing the college and university student population are unique, and the roles of video game use in predicting those risks is unclear. This study represents an early broad step toward a better understanding of what risks game users face to greater or lesser degrees than their peers. Why some risks are unique to the game-playing population, and what can be done to alleviate them, is something we must investigate further.
Electronic Supplementary Material
The electronic supplementary material is available with the online version of the article at http://dx.doi.org/10.1027/1864-1105/a000210
- Survey Questionnaire Instrument Items (24).
Adrienne Holz Ivory (PhD, Human Development, Virginia Tech), is an assistant professor in the Department of Communication at Virginia Tech, USA.
James D. Ivory (PhD, Mass Communication, University of North Carolina at Chapel Hill), is an associate professor in the Department of Communication at Virginia Tech, USA.
Madison Lanier (BA, Communication and Political Science, Virginia Tech), is a master’s student in the Department of Communication at Virginia Tech, USA.
References
(1991). Acquaintance rape and alcohol consumption on college campuses: How are they linked? Journal of American College Health, 39, 165–169. doi: 10.1080/07448481.1991.9936229
. (2015). National college health assessment, Retrieved from http://www.acha-ncha.org/
(2007). Rethinking how family researchers model infrequent outcomes: A tutorial on count regression and zero-inflated models. Journal of Family Psychology, 21, 726–735. doi: 10.1037/0893-3200.21.4.726
(2015). There is broad consensus: Media researchers agree that violent media increase aggression in children, and pediatricians and parents concur. Psychology of Popular Media Culture, 4, 200–214. doi: 10.1037/ppm0000046
. (1997, November 14). Youth risk behavior surveillance: National college health risk behavior survey – United States, 1995. Morbidity and Mortality Weekly Report, 46(SS-6). Retrieved from http://www.cdc.gov/mmwr/PDF/ss/ss4606.pdf
(2004). Bullying in college by students and teachers. Adolescence, 39, 53–64.
(2002). Alcohol use and risky sexual behavior among college students and youth: Evaluating the evidence. Journal of Studies on Alcohol, s14, 101–117. doi: 10.15288/jsas.2002.s14.101
(2016). Violent video games and violent crime. Southern Economic Journal, 82, 1247–1265. doi: 10.1002/soej.12139
(2014). Press CRTT to measure aggressive behavior: The unstandardized use of the competitive reaction time task in aggression research. Psychological Assessment, 26, 419–432. doi: 10.1037/a0035569
(2014). Video game violence use among “vulnerable” populations: The impact of violent games on delinquency and bullying among children with clinically elevated depression or attention deficit symptoms. Journal of Youth and Adolescence, 43, 127–136. doi: 10.1007/s10964-013-9986-5
(2011). The effects of risk-glorifying media exposure on risk-positive cognitions, emotions, and behaviors: A meta-analytic review. Psychological Bulletin, 137, 367–390. doi: 10.1037/a0022267
(2012). Reassessing media violence effects using a risk and resilience approach to understanding aggression. Psychology of Popular Media Culture, 1, 138–151. doi: 10.1037/a0028481
(2011). Pathological video game use among youths: A two-year longitudinal study. Pediatrics, 127, e319–e329. doi: 10.1542/peds.2010-1353
(2016). Working towards an international consensus on criteria for assessing Internet gaming disorder: A critical commentary on Petry et al. (2014). Addiction, 111, 167–175. doi: 10.1111/add.13057
(2014). A longitudinal study of risk-glorifying video games and behavioral deviance. Journal of Personality and Social Psychology, 107, 300–325. doi: 10.1037/a0036058
(2015). Manufacturing consensus in a diverse field of scholarly opinions: A comment on Bushman, Gollwitzer, and Cruz (2015). Psychology of Popular Media Culture, 4, 222–229. doi: 10.1037/ppm0000056
(2003). Let the games begin: Gaming technology and entertainment among college students. Pew Internet and American Life Project. Retrieved from http://www.pewinternet.org/files/old-media//Files/Reports/2003/PIP_College_Gaming_Reporta.pdf.pdf
(2016). The demographic and political composition of Mechanical Turk samples. SAGE Open, 6(1), 1–17. doi: 10.1177/2158244016636433
(2016). Sexy, strong, and secondary: A content analysis of female characters in video games across 31 years. Journal of Communication, 66, 564–584. doi: 10.1111/jcom.12237
. (2016). Fast facts. Retrieved from http://nces.ed.gov/fastfacts/display.asp?id=98
(2007). Differences in health risk behaviors between college freshmen living in special interest housing and traditional housing, (Unpublished doctoral dissertation). Retrieved from http://hdl.handle.net/10919/28081
(2014). An international consensus for assessing Internet gaming disorder using the new DSM-5 approach. Addiction, 109, 1399–1406. doi: 10.1111/add.12457
(2015). Digital games research: A survey study on an emerging field and its prevalent debates. Journal of Communication, 65, 975–996. doi: 10.1111/jcom.12182
(2010).
Who are the crowdworkers? Shifting demographics in Mechanical Turk . In CHI ‘10Eds., Extended Abstracts on Human Factors in Computing Systems (CHI EA ‘10) (pp. 2863–2872). New York, NY: ACM. doi: 10.1145/1753846.1753873(2006). Prevention of eating disorders in at-risk college-age women. Archives of General Psychiatry, 63, 881–888. doi: 10.1001/archpsyc.63.8.881
(2003). Taking up binge drinking in college: The influences of person, social group, and environment. Journal of Adolescent Health, 32, 26–35. doi: 10.1016/S1054-139X(02)00457-3
