Character Identification as a Moderator of the Relationship Between Social Norms and Sexual Risk-Reduction Intentions and Behavior: Findings from an eHealth Entertainment-Education Intervention Targeting Men who Have Sex with Men

Abstract

Normative based research found that norms are significant predictors of safe sex behavioral intentions and behavior. Research shows that group identity moderates the relationship between norms and intentions/behavior. The present study used the theory of normative social behavior to evaluate whether identification with characters in an HIV-prevention interactive video moderated the relationship between sexual risk-taking norms and risk reduction intentions/behavior. Participants included 540 men between the ages 18 and 30 enrolled in a randomized controlled trial with a 3-month follow-up. We found support for the hypothesized interaction. At low levels of character identification, the negative relationship between sexual risk-taking norms and risk reduction intentions was strong. However, as character identification increased, the link between norms and intentions became weaker. The mean intentions score of high risk norm participants was elevated to the level reported by low risk norm participants, demonstrating the favorable effect of character identification on narrative persuasion in entertainment-education. The utility of a social norms approach to health behavior change will be discussed in the context of eHealth interventions.

The HIV epidemic continues to disproportionately affect men who have sex with men (MSM) in the U.S. (Center for Disease Control and Prevention [CDC], 2017). Between 2008 and 2014 the number of new HIV infections declined by 18% among white MSM, stabilized among black MSM, and increased by 20% among Latino MSM (CDC, 2017). Although most existing HIV prevention interventions attempt to reduce unsafe sexual practices, HIV prevention interventions targeting MSM have been successful in changing social norms via peer groups (Duan, Zhang, Wang, Wei, Yu, & She, 2013; Kegeles, Hays, & Coates, 1996).

The present study contributes to normative research by applying the theory of normative social behavior (TNSB) to an interactive, video-based safe sex intervention. Interactive safe sex videos, attempt to communicate health risks and change behavior by providing people with an opportunity to learn about HIV/AIDS and practice healthy decision-making in a safe, virtual environment (Noar, 2011). The study informs health agencies and educational institutions about ways in which new media technologies, like video-, smartphone-, and virtual reality-based interventions can promote norms and behavior change among MSM.

The Role of Social Norms in Influencing Behavioral Outcomes

Research on social norms demonstrated that college students who perceived that the majority of students on campus engaged in specific health behaviors (i.e. drinking, smoking, and condom use), were more likely to engage in similar behaviors (DiGuiseppi, Meisel, Balestrieri, Ott, Clark, & Barnett, 2018; Wright, Tokunaga, & Kraus, 2016). Studies have shown that social norms were significant predictors of HIV risk-reduction activities as well (Miner, Peterson, Welles, Jacoby, & Rosser, 2009; Zhou et al., 2017).

The theory of normative social behavior posits that injunctive norms, outcome expectations, and group identity moderate the relationship between descriptive norms and related behaviors (Rimal & Real, 2005; Rimal & Lapinski, 2015). The present study extends this theoretical framework by exploring various types of group identity as moderators of the relationship between individuals’ perceived sexual risk-taking descriptive norms and risk-reduction intentions and behavior.

Group identity was conceptualized as perceived similarity to others and aspirations to be like the referent other (Rimal & Real, 2005). According to the TNSB, as individuals’ identification with the group increases and they perceive that others in their reference group engage in a behavior, they are more likely to engage in the behavior themselves (Carcioppolo, Dunleavy, & Qinghua, 2017; Rimal, 2008).

The development of media-based e-Health interventions provides a very interesting opportunity for participants to identify with characters depicted in on-screen narratives. Slater and Rouner (2002) defined identification with characters as an experience “in which an individual perceives another person as similar or at least as a person with whom they might have a social relationship” (p. 178), and hypothesized that identification with characters and engagement with the storyline is linked to narrative persuasion.

In the current work, we present secondary analyses of a randomized controlled trial funded by the National Institute of Allergy and Infectious Diseases (NIAID). This longitudinal trial examined the effectiveness of Socially Optimized Learning in Virtual Environments (SOLVE) – an interactive, video-based HIV-prevention intervention for young adult MSM. Miller and her colleagues have developed and tested several HIV-prevention eHealth interventions using the SOLVE approach; these interventions have been shown to successfully reduce HIV risk behaviors (Christensen et al., 2013; Miller et al., 2012).

These interventions build upon a neuroscience-based model of decision-making (Bechara, Damasio, Tranel, & Damasio, 1997; Damasio, 2000) that takes into account both deliberative and non-conscious decision-making processes. SOLVE interventions are narrative-based eHealth applications (e.g., interactive movies; video games) that place participants in a virtual world simulating the risky situations young MSM typically encounter on first dates or “hook-ups.” The participant determines how the narrative proceeds by making self-regulatory decisions that affect the two “model characters” (i.e., the characters that represent the participant and his virtual sex partner). The participant also encounters two “guide characters” that provide health information, advice, social support, and other assistance throughout the narrative.

SOLVE incorporates many theory-based behavior change techniques commonly used in health communication interventions (see taxonomy described by Michie, Ashford, Sniehotta, Dombrowski, Bishop, & French, 2011). For example, SOLVE (1) provides information regarding general and individual behavioral consequences of risky sex, (2) promotes goal setting and action planning, (3) helps participants identify and overcome barriers to safe sex, (4) prompts self-monitoring of sexual behavior, (5) provides opportunities to “virtually” practice risk-reduction strategies, (6) stimulates anticipation of future rewards, and (7) helps participants manage stress and emotional states – such as shame – that may precipitate risky behavior.

The SOLVE intervention tested in the current trial consisted of a 20 to 30 minute interactive video delivered via DVD technology. At the start of the video, two “guide characters” appear on-screen and introduced themselves. The older, more authoritative guide sets up the narrative by saying “In this video you’re going to meet two guys hooking up for the first time and you get to be one of the characters.” The guides mention that they will be “checking in on things when it counts…when things start getting hot.” They then introduce the two main “model characters” – one represents the participant, the other is the sexual partner he will meet later on as the story progresses. The next scene takes place in the self-character’s apartment as he is getting ready for the evening. Normative beliefs are elicited as this character models risk-reduction behavior such as checking the expiration date on his condom packages. After this opening sequence, the participant encounters his first choice point and must decide whether he would like to meet his partner at a dance club or over the Internet. The two characters then meet and engage in a flirtatious conversation. It soon becomes clear that they are both attracted to each other and are interested in pursuing a sexual encounter. As the storyline continues, the participant encounters more choice points such as whether to drink alcohol and take methamphetamine.

Back at his partner’s apartment, the participant must decide when and how to initiate a conversation about safe sex. In the bedroom, he makes decisions regarding which sexual activities he would like to engage in and whether to stay safe or take risks. Whenever the participant makes a risky decision, the narrative pauses and the guide characters appear on screen to provide persuasive health messages. These messages involve an ICAP process (Interrupt and Challenge risky decisions, Acknowledge desires, and Provide a means to achieve desires while avoiding risk). After the guides intervene, the participant is given a “second chance” to make a safer choice or not. At the end of the intervention, the guide characters appear once again to review each choice made throughout the narrative. Participants are presented with hypothetical scenarios in which the main characters model risk-reduction behaviors, helping the participant concretely visualize what he can do to stay safe in the real-world. A primary goal of the intervention is to enhance normative beliefs regarding condom use and related communication (i.e., initiating a conversation about condoms, negotiating condom use with an unwilling partner, and refusing sex if no condoms are available). For more implementation details see Appleby, Godoy, Miller, and Read (2008), and Miller, Christensen, Godoy, Appleby, Corsbie-Massay, and Read (2009).

The trial examined the effectiveness of SOLVE compared to a waitlist control group using a pre-post design with 3-month follow-up. In the treatment condition, participants were exposed to the interactive version of our video intervention. In the waitlist control condition, participants were not exposed to the intervention during the trial, but had the option of viewing the video after the trial had completed. The trial design included two additional control groups but data from these conditions are not included in the present analyses.

Based on social norms research the following hypotheses are proposed: (H1) at baseline, the more participants perceived sexual risk taking as normative, the more they engaged in unprotected anal intercourse (UAI) during the previous three months, and (H2) at baseline, the more participants perceived sexual risk taking as normative, the less likely they were to endorse risk-reduction behavioral intentions. Furthermore, (H3) perceived sexual risk-taking norms will be lower at 3-month follow-up compared to baseline, and this reduction will be greater among participants in the treatment condition compared to the waitlist control condition.

The final set of hypotheses pertains only to participants randomly assigned to the eHealth intervention condition. According to the TNSB, character identification should attenuate (or repair) the unhealthy, negative association between sexual risk-taking norms and risk-reduction intentions/behavior. The following hypotheses are proposed: (H4) at follow-up, perceived sexual risk-taking norms will be negatively related to risk-reduction behavioral intentions and the magnitude of this association will become weaker as identification with the (a) model characters and (b) guide characters increases, and (H5) perceived sexual risk-taking norms will be negatively related to sexual risk-taking behavior and the magnitude of this association will become weaker as identification with the (a) model characters and (b) guide characters increases.

The eHealth intervention includes two types of characters, guide characters and model characters. Therefore, hypothesis 4 (i.e., intentions) and hypothesis 5 (i.e., behavior) will be tested twice: once with guide character identification as the moderator and once with model character identification as the moderator. We make no a priori hypotheses regarding differential effects of these two types of character identification.

Method

Recruitment and Procedure

Participants were recruited throughout Los Angeles, California via: (1) print and online media advertisements and (2) in-person street intercepts and bar/club venues during the hours of 10:00pm to 2:00am daily with occasional afternoon recruitment at special events such as gay pride festivals. Potential participants received $3.00 for completing a screening survey at in-person recruiting sites; no compensation was offered to those who completed the screening survey online or over the telephone. The men who were eligible to participate were presented with an information sheet describing the study protocol. If they were interested in enrolling, they chose an anonymous codename and scheduled their baseline interview, which had to occur within two weeks after determining eligibility.

Participants were asked to complete two sessions, each lasting approximately two hours. These sessions were three months apart and took place at an off-campus, university-affiliated office. At the beginning of the first session, a recruitment/retention coordinator greeted the participant, led him to a private interview room, and completed the informed consent process. Participants were then assigned to condition, stratified by age (18 to 24 versus 25 to 30 years old) with a balanced 1:1 allocation. The randomization sequence was created using a computer-generated list of random numbers. After assignment to condition, the recruitment/retention coordinator introduced the participant to one of several trained interviewers, selected to conduct the interview based on their current availability. If the participant was assigned to the treatment group, they later met with a separate interventionist who set up the DVD-based interactive video and provided detailed instructions on how to use the remote control to make choices. The original interviewer returned to administer a post-intervention survey immediately following exposure to the intervention. At the end of this first session, the recruitment/retention coordinator met with the participant to schedule a follow-up session to take place three months later. At this second session, participants reported follow-up data to the same person who conducted their first interview when possible. At the end of this follow-up session, the participants originally assigned to the waitlist control condition were given the option to view the interactive video intervention. Participants were offered $50.00 in compensation per session. The study protocol was approved by the University of Southern California’s institutional review board.

Participants

To be eligible for this trial, participants had to be a male between the ages of 18 and 30 years of age who self-identified as Black/African-American, Latino/Hispanic, or White/Caucasian. They also had to report at least one instance of UAI with another man in the three months prior to enrollment in the study. Exclusion criteria included HIV seropositivity and prior use of non-prescription injection drugs. A total of 540 participants were assigned to the treatment (n=276) and waitlist control conditions (n=264) with 172 (32%) lost to follow-up. Participants were about 24 years old on average (SD = 3.4), most identified as gay or homosexual (86%), and 56% had an educational attainment level of “some college” or higher. Approximately 41% were White/Caucasian, 40% were Latino/Hispanic, and 19% were Black/African-American.

Description of the Intervention Development and Production Process

We produced three separate versions of the interactive video intervention in a conscious effort to maximize participants’ identification with the “guide” and “model” characters. Each version of the intervention was culturally-tailored to one of the three racial groups with the highest prevalence of HIV: African American, Latino, and Caucasian. In addition to race, the characters in the story were designed to match other demographic features of the study participants – including sexual identity, average educational attainment, and average age. In an effort to further enhance identification, we conducted a series of focus groups and one-on-one interviews with members of the target population. This formative research informed many aspects of the design and content of the intervention such as the setting, set decorations, physical attributes of the guide and model characters, and communication style preferences. Another pilot study helped us identify the narrative structures of sexual scenarios typically experienced by our participants. This allowed us to identify key features of the social interactions that lead up to risky sexual behavior (e.g., substance use). Data from these qualitative and quantitative pilot studies revealed that sexual encounters and the narratives that precede them are quite similar for African American, Latino, and Caucasian MSM. For example, the two most frequently reported ways in which men in this population met a sexual partner were through the Internet and at a bar/club and so those were the two choices we included in the intervention. In addition to preferred personality and physical features of sexual partners, the data also revealed the most frequently reported sexual acts and ways to initiate conversations about safe sex.

All of the formative research mentioned above served as the foundation for the 3 scripts we developed. The basic narrative storyline was then culturally tailored by one of three professional screenwriters who matched the demographic characteristics of the audience. For example, we hired a screenwriter identifying as a Latino MSM to tailor the script for the video that would ultimately be shown to Latino MSM participants enrolled in the study. Character identification was also enhanced through the use of Community Advisory Boards (CABs) who reviewed the scripts and provided feedback regarding the realism of the intervention and its ability to resonate with the target audience. There were two types of advisory boards: leadership CABs (i.e., representatives from community-based organizations, the entertainment industry, and local businesses frequented by MSM) and youth CABs (i.e., young MSM, all of whom met the study’s eligibility criteria). During video production, the director, producers, and many other key personnel were members of the LGBT community and represented diverse cultural backgrounds. These measures were taken to enhance realism and cultural sensitivity throughout the process so that character identification was maximized. We strongly believe our efforts resulted in a realistic intervention that provided ample background information about the guide and model characters so that participants could make accurate identification judgements.

Measures

Perceived sexual risk-taking norms

Participants were asked to respond to a measure assessing perceived normative acceptance of sexual risk-taking behaviors, operationalized as descriptive norms. The interviewer provided the following instructions: “Please tell me the percentage of men from 0% (none) to 100% (all) that reflects how many gay or bisexual men of your age and ethnic group you think have engaged in each of the following behaviors in the past six months.” These instructions are in line with theory of normative social behavior (Rimal, 2008; Rimal & Real, 2005), which conceptualizes norms as perceptions of acceptance by one’s peer group. The measure addressed five specific risk behaviors related to having unprotected sex with another man. Specifically, participants were asked what percentage of their peers believe it is okay to (1) occasionally slip up and not use a condom when they have anal sex, (2) stop using condoms if they are really in love with someone, even if they are not sexually exclusive with him, (3) have anal sex without using a condom when the partners do not know each other’s HIV status, (4) have anal sex without using a condom if they have been safe for a while, and (5) bareback – have anal sex without a condom – regularly. This scale is based on similar measures of safe sex norms used by Kelly et al. (1990), Kelly et al. (1992), and Kalichman, Roffman, Picciano, and Bolan (1998). A principal components factor analysis with varimax rotation yielded a single factor (Eigenvalue = 2.73) explaining 54.52% of the variance (KMO = .82; Bartlett’s X² (10) = 706.98, p < .001). Responses to the five items were summed to form a composite. Internal consistency was high at baseline (Cronbach’s alpha = .79; M = 46, SD = 19) and at follow-up (Cronbach’s alpha = .83; M = 43, SD = 19).

Sexual risk-taking behavior

At baseline, participants reported the number of times they had engaged in receptive and insertive UAI during the 3-month period prior to enrolling in the trial. As mentioned above, a requirement for eligibility was that the participant had to have recently engaged in at least one instance of UAI. This instance of UAI could have occurred with a non-primary partner or with a primary partner, who we defined as someone the participant was in a romantic relationship with for at least three months and shared a special emotional bond. Sexual risk-taking has been operationalized as the number of times participants took sexual risks with non-primary partners – outside of a romantic relationship. Therefore, some participants – those who only had UAI within their relationships – have values of zero on our measure of risk (even though all participants in the study had recently engaged in some form of UAI). Counts of sexual risk-taking with non-primary partners were reported at the beginning of the baseline session (M = 2.68, SD = 7.05; Median = 1; Min = 0, Max = 83) and again three months later at the beginning of the follow-up session (M = 2.40, SD = 10.27; Median = 0; Min = 0, Max = 84).

Risk reduction behavioral intentions

Participants reported their intentions to engage in a variety of HIV risk-reduction strategies by responding to an 18-item scale that ranged from 1 (definitely will not do) to 10 (definitely will do). This measure was adapted from a scale used by Kalichman and Rompa (1995), designed to reflect common strategies utilized by HIV-prevention risk-reduction interventions. A principal components factor analysis with varimax rotation yielded two factors (KMO = .95; Bartlett’s X² (153) = 6416.77, p < .001). The first factor (Eigenvalue = 9.27) explained 51.52% of the variance and comprised 11 items reflecting “sexual script self-regulatory intentions.” Sample risk-reduction strategies in this category include (1) refusing to have sex if your partner insists on having risky sex, (2) telling your partner that you need to practice safer sex, (3) switching from the things that are getting risky to something safer, and (4) telling yourself that safe sex is as good as intercourse without a condom. The second factor (Eigenvalue = 1.23) explained 7.04% of the variance and comprised 4 items reflecting “condom preparedness intentions.” These risk-reduction strategies were (1) keeping condoms nearby, (2) always making sure you have condoms, (3) making sure a condom will be available if it looks like you are going to have sex, and (4) making sure that your condom supply is in good condition – not too old, etc. Composites for both factors were created by averaging the items. The 11-item measure of sexual script self-regulatory intentions was reliable at baseline (Cronbach’s alpha = .92; M = 6.49, SD = 1.19) and follow-up (Cronbach’s alpha = .93; M = 6.75, SD = 1.06). The 4-item measure of condom preparedness intentions was also reliable at both baseline (Cronbach’s alpha = .91; M = 6.68, SD = 1.37) and at follow-up (Cronbach’s alpha = .92; M = 6.98, SD = 1.17). The remaining 3 items were excluded due to weak loadings.

Identification with the characters

Participants randomly assigned to the treatment condition used a 4-item scale ranging from 1 (not at all) to 10 (very much) to report how much they “could identify with” the four main characters portrayed in the interactive video intervention – the two model characters and the two guide characters. These four questions were asked immediately following exposure to the intervention; each referring to a different character. The interviewer provided the participant with a flipbook containing photographs (i.e., screenshots) of the four characters, each on a separate page. The purpose of providing these photos was to ensure the participant knew which character the interviewer was referring to during each question. We averaged the items together, creating two separate composites; one representing identification with the model characters (Cronbach’s alpha = .89; M = 5.93, SD = 1.87) and another representing identification with the guide characters (Cronbach’s alpha = .88; M = 5.62, SD = 2.18).

Attrition Analyses

We conducted a series of attrition analyses to determine whether participants who returned for the follow-up session differed from those who dropped out of study on any of the baseline variables. The results of a MANOVA indicated that there were two differences between these groups, F(5, 529) = 3.43, p < .01; Wilk’s Λ = .97. Participants who dropped out of the study tended to be younger, F(1, 533) = 7.84, p < .01, and less educated, F(1, 533) = 15.53, p < .001. However, attrition was not related to self-regulatory intentions, F(1, 533) = 0.09, p = .77, condom preparedness intentions, F(1, 533) = .08, p = .77, or risk-taking norms, F(1, 533) = .77, p = .38. Chi-square analyses revealed that the likelihood of drop out was not associated with assignment to condition, X² = .33, p = .57, racial identification, X² = .39, p = .82, or sexual identification, X² = .01, p = .92. A Mann-Whitney test indicated that drop out was not related to baseline sexual risk-taking behavior, U = 31,141, p = .75. Finally, t-tests conducted on data from the treatment group revealed that drop out within that condition was not related to identification with the guide characters, t(274) = −.78, p = .44, or model characters, t(274) = −.77, p = .44. The Appendix describes the rationale for using complete case analyses.

Primary Analyses

Sample characteristics and baseline data are presented in Table 1. Preliminary analyses have shown that educational attainment (r_s = −.14, p<.01) and sexual identification (r_s = .09, p < .05) were related to baseline UAI. Therefore, educational attainment and sexual identification were included as covariates in all models.

Table 1.

Baseline Data Split by Condition (N=540)

	Combined (N = 540)	Intervention (N = 276	Control (N = 264)
Race/Ethnicity, % (n)
White	40.6 (219)	40.2 (111)	40.9 (108)
Latino	40.0 (216)	40.2 (111)	39.8 (105)
Black	19.4 (105)	19.6 (54)	19.3 (51)
Sexual Identification, % (n)
Gay Identified	86.3 (466)	87.0 (240)	85.6 (226)
Non-Gay Identified	13.7 (74)	13.0 (36)	14.4 (38)
Educational Attainment, % (n)
Less Than High School	1.5 (8)	1.8 (5)	1.1 (3)
High School or GED	13.9 (75)	14.5 (40)	13.3 (35)
Some College	40.4 (218)	39.9 (110)	40.9 (108)
Associate Degree	9.6 (52)	12.0 (33)	7.2 (19)
Bachelor’s Degree	29.1 (157)	26.1 (72)	32.2 (85)
Master’s Degree	5.2 (28)	5.8 (16)	4.5 (12)
Doctoral Degree	.4 (2)	0.0 (0)	.8 (2)
Age, M (SD)	24.1 (3.4)	24.1 (3.4)	24.2 (3.4)
Receptive NPUAI, M (SD)	.75 (2.7)	.89 (3.5)	.60 (1.3)
Insertive NPUAI, M (SD)	.95 (3.8)	.89 (1.7)	1.0 (5.1)

Note. No significant differences between conditions; NPUAI refers to the frequency of unprotected anal intercourse with a non-primary partner during the past three months

Negative binomial regression analysis was used to test the first hypothesis, that the more participants perceived sexual risk taking as normative, the more they engaged in UAI during the previous three months. This analytic approach was chosen based on preliminary analyses that revealed a non-normal distribution with extreme negative skew for the dependent variable (as is typical of count variables such as UAI). Hypothesis 1 was supported. As predicted, sexual risk-taking norms predicted the number of times participants engaged in UAI (β = .190, SE = .08, p < .05). The incidence rate ratio was IRR = 1.21 (95% CI: 1.03, 1.41) and represents the proportional increase in UAI for each unit increase in norms.

Hierarchical multiple regressions were used to test the second hypothesis, that the more participants perceived sexual risk taking as normative, the less likely they were to endorse risk-reduction behavioral intentions. As mentioned before, a factor analysis yielded two factors for the risk-reduction behavioral intentions scale; sexual script self-regulatory intentions and condom preparedness intentions. The results provided support for hypothesis 2. Norms negatively predicted sexual script self-regulatory intentions (β = −.15, p<.01), explaining 2% of the variance (R² = .02, F(3, 531) = 4.56, p < .01). Norms also negatively predicted sexual condom preparedness intentions (β = −.12, p < .01), explaining 1% of the variance (R² = .01, F(3, 531) = 2.65, p < .05).

Repeated measures ANOVA was conducted to test hypothesis 3, that perceived sexual risk-taking norms will be lower at 3-month follow-up compared to baseline, with the reduction being greater among participants in the treatment condition compared to the waitlist control condition. Means and standard deviations are presented in Table 2. The main effect of perceived sexual risk-taking norms was found to be significant, F(1, 363) = 5.57, p < .01, partial η ² = .015. This provides evidence that norms changed from baseline to follow-up, overall. The interaction between sexual risk-taking norms and condition was not statistically significant, indicating that the reduction in norms reported by participants in the experimental group was not greater than the reduction reported by participants in the control group, as we had expected. In the treatment condition, sexual risk-taking norms were significantly lower at 3-month follow-up (M = 42.86, SD = 17.58) compared to baseline (M = 45.34, SD = 18.18). Surprisingly, norm change did not differ by condition; the same pattern was observed in the control group: Sexual risk-taking norms were significantly lower at 3-month follow-up (M = 44.90, SD = 19.60) compared to baseline (M = 46.54, SD = 19.64). Therefore, hypothesis 3 was only partially supported.

Table 2.

Means and standard deviations of perceived sexual risk-taking norms across time

	Baseline		3 month follow up
	Experimental	Control	Experimental	Control
	M (SD)	M (SD)	M (SD)	M (SD)
Perceived sexual risk-taking norms	45.34 (18.18)	46.54 (19.64)	42.86 (17.58)	44.90 (19.60)

The next set of analyses pertain only to people exposed to the intervention. Hypothesis 4 predicted that character identification would moderate the relationship between norms and intentions such that the magnitude of the norms/intentions relationship should become weaker as identification increases. The first regression tested the moderating effect of identification with model characters on the relationship between norms and sexual script self-regulatory intentions. The main effect was significant: Namely, the more participants perceived sexual risk taking as normative, the less likely they were to endorse sexual script self-regulatory intentions (β = −.24, p < .01). The interaction between norms and model identification was significant (β = .19, p < .05). The full model accounted for 11% of the variance in intentions (R² = .11, F(5,176) = 4.25, p < .01), with 3% attributed to the interaction. The Johnson-Neyman technique was used to probe the interaction. This procedure identifies the specific values of the moderator for which the relationship between the predictor and outcome are significant (Field, 2013; Hayes, 2013; Hayes & Matthes, 2009). When model identification was 6.36 or below, the effect of perceived norms on intentions was significant and negative (p < .05). By contrast, when model identification was 6.40 and above, no significant effect was observed. A plot of the means reveals that the intentions scores of high risk norm participants were elevated to the level reported by low risk norm participants. See Figure 1. Revealing a similar pattern of results, a second regression tested the moderating effect of identification with model characters on the relationship between norms and the second type of intentions – condom preparedness intentions. The main effect was significant (β = −.21, p < .01) and the interaction was marginally significant (β = .16, p = .07). The model accounted for 6% of the variance (R² = .06, F(5,178) = 2.34, p < .05), with 2% contributed by the interaction. When model identification was 6.38 or lower, the effect was significant and negative (p < .05). At values of 6.40 and above, no significant effect was observed. See Figure 2.

Figure 1.

Identification with model characters as a moderator of perceived sexual risk-taking norms and sexual script self-regulatory intentions.

Figure 2.

Identification with model characters as a moderator of perceived sexual risk-taking norms and condom preparedness intentions.

In the next two analyses, model identification was replaced with guide identification. When the outcome was sexual script self-regulatory intentions, the main effect of norms was significant and in the expected direction: (β = −.19, p < .05) and the interaction between norms and guide identification was significant (β = .17, p < .05). The full model accounted for 13% of the variance (R² = .13, F(5,176) = 5.33, p < .001), with 3% attributed to the interaction. When guide identification was 5.70 or below, the effect was significant and negative (p < .05). When it was 5.95 and above, no effect was observed. See Figure 3. A similar pattern emerged when testing the moderating effect of identification with guide characters on the relationship between norms and the second type of intentions – condom preparedness intentions. The main effect was significant (β = −.17, p < .05) and so was the interaction term (β = .15, p < .05). The full model accounted for 8% of the variance in condom preparedness intentions (R² = .08, F(5,178) = 3.16, p < .01), with 2% due to the interaction. When identification was 5.68 or lower, the effect was significant and negative (p < .05), while at values of 5.95 and above, no effect was observed. See Figure 4. Taken together, these analyses provide support for hypothesis 4, that the magnitude of the norms/intentions relationship should become weaker as character identification increases.

Figure 3.

Identification with guide characters as a moderator of perceived sexual risk-taking norms and sexual script self-regulatory intentions.

Figure 4.

Identification with guide characters as a moderator of perceived sexual risk-taking norms and condom preparedness intentions.

Finally, two negative binomial regression analyses were conducted to test hypothesis 5, that character identification would moderate the relationship between norms and behavior (UAI at three-month follow-up) such that people with high character identification will report lower rates of follow-up UAI regardless of their normative beliefs. In addition to educational attainment and sexual identification, baseline UAI was included in both models as a covariate. Across the two analyses, we found no support for hypothesis 5.

Discussion

This study contributes to previous work on normative research by assessing the influence of perceived descriptive sexual risk-taking norms on risk-reduction intentions and sexual risk-taking behavior. The present findings lend support to one of the central hypotheses of the study, that identification with model characters and guide characters from the interactive intervention moderated the relationship between perceived sexual risk-taking norms and risk-reduction behavioral intentions. In other words, character identification seems to fix or repair the undesirable negative relationship between risk-taking norms and safe sex intentions. It is important to mention that, in all instances, the interactions were significant at low and moderate-low identification, but not at high identification. Even though the hypothesis was supported, the effect sizes for the interactions were small.

These findings are consistent with past work in normative research literature (Neighbors et al., 2010). While these findings are in line with the tenets of the TNSB (Byron et al., 2016), more research is needed to explore these relationships in e-Health intervention application. Interestingly, the interaction between norms and identification with guide characters explained more variance in the dependent variables compared to the interaction between norms and identification with model characters. One explanation might be that participants saw the guide characters as their reference group, supporting the tenets of the TNSB (Rimal, 2008). While the model characters modeled positive and negative behaviors, depending on the participants’ ingame choices, the guide characters were portrayed as peer counselors that consistently modeled positive behaviors, such as negotiating condom use and safer sex practices (Miller et al., 2012).

Norms significantly predicted sexual script self-regulatory intentions, both at baseline and 3-month follow-up (Carcioppolo et al., 2017). The more participants estimated that others engage in sexual risk-taking behaviors, the less they intended to engage in strategies that would reduce their risk for contracting HIV. Similarly, norms predicted condom preparedness intentions at baseline and 3-month follow up. The more participants estimated that others engage in sexual risk-taking behaviors, the less they intended to take precautions that would ensure a condom would be available in the heat of the moment. These findings are in line with previous research (Byron et al., 2016; Carcioppolo et al., 2017).

The pre-post experimental design allowed for exploring norm change as a consequence of participating in the intervention. The findings suggested that perceived sexual risk-taking norms decreased from baseline to 3-month follow-up for participants in the experimental group as well as participants in the control group. Participants in both groups reduced their estimates of the percentage of gay or bisexual men of their age and ethnic group that engage in various HIV risk behaviors. The mean reduction was greater in the experimental group, however, the difference between groups was not statistically significant. It is plausible that the pretest survey served as an “intervention” in and of itself, making safe sex norms salient among participants in both groups. All participants chose to enroll in a trial that was clearly identified as HIV-prevention research and spent approximately one hour responding to numerous baseline survey questions related to safe sex. Similar “control group improvement” effects are common and have been described in the health communication literature (see Waters, Reeves, Fjeldsoe, & Eakin, 2012). It should be noted that the randomized controlled trial of this intervention provided evidence that, compared to control participants, those exposed to the intervention showed statistically significant improvement in HIV/AIDS knowledge, safe sex self-efficacy, risk perceptions, shame reduction, safe sex intentions, and condom usage (Christensen et al., 2013).

Turning to the norms-behavior hypotheses, perceived sexual risk-taking norms predicted sexual risk-taking behavior at baseline, but the effect was not significant at 3-month follow-up. One explanation for these mixed findings might be the fact that norms seem to be inconsistently related to behavioral outcomes (Carcioppolo & Jensen, 2012). Several studies on college students’ alcohol usage found that drinking behavior was predicted by subjective norms, rather than descriptive norms (Campo, Brossard, Frazer, Marchell, Lewis, & Talbot, 2003). Also, research on social norms and HIV-prevention interventions found that the effect of social norms on unsafe sex is mediated by condom use self-efficacy and safer sex intentions (Miner et al., 2009). Therefore, more research is needed to understand the relationship between descriptive and injunctive norms, and behavior, not only in HIV-prevention, but in regard to other negative health behaviors as well. The other main hypothesis of the study was that identification with model and guide characters would moderate the relationship between perceived sexual risk-taking norms and sexual risk-taking behavior at 3-month follow-up. However, there was no main effect for perceived risk-taking norms on risky sexual behavior and the interaction was not statistically significant. And so the question remains as to whether the observed intention effects would translate to meaningful differences in sexual practices. Perhaps additional exposures to the intervention or a more tailored messaging strategy would help overcome the intention-behavior gap. It is an important next step to conduct systematic studies that attempt to identify the best ways to translate our effects to meaningful differences in actual behavior.

The present study lends support for the applicability of the TNSB to e-health interventions, by highlighting group identity as a moderator of the relationship between descriptive norms and behavioral intentions. However, a new operational definition for the concept of group identity in the context of e-health interventions is needed. In the present study it appeared that the participants saw the guide characters as a reference group, and not the model characters. Future e-health interventions should focus on identifying participants’ reference group and ensure that the health behaviors modeled by this group are positive. In line with previous research, the present study supported the descriptive norms/behavioral intentions relationship (Halim, Hasking, & Allen, 2012). Moreover, the present study did not find support for the descriptive norms/behavior relationship, adding to research that reported mixed results on this issue (Carcioppolo & Jensen, 2012). Future research should attempt to parse out the relationships between different types of norms and behavior in the context of e-health interventions as well.

Limitations

There are several limitations to the present study that should be addressed. First of all, the identification scale included only 2 items, character identification not being one of the main focus of this NIAID-funded trial. However, at 3-month follow up there was a statistically significant interaction between identification with characters and perceived sexual risk-taking norms and so the 2-item scale was indeed able to detect the hypothesized effect. Second, this study is limited in that the perceived sexual risk-taking norms scale uses a “past 6-months” time frame; however, the longitudinal trial lasted 3 months, not 6 months. Therefore, our findings must be interpreted with this limitation in mind. The overlap in time frame might obscure any existing associations, resulting in null effect.

Another limitation is related to the age of the dataset. Development of the intervention began in 2003 with data collection occurring from 2005 to 2008. Given the age of the dataset, it is important to highlight the significance of the present study in the context of rapidly-evolving digital innovations that are common to computer technology-based interventions (CBIs) such as the one tested here. The advantages/affordances of CBIs include low cost of implementation after intervention development, intervention fidelity, tailoring, interactivity, multimedia capabilities, and flexibility in terms of dissemination (Noar, 2011). While technologies become more sophisticated, CBIs affordances oftentimes stay the same. The DVD-based intervention used in this study leveraged basic affordances such as interactivity and message framing in the same way a modern smartphone-based intervention might today. Many contemporary digital media scholars argue that the affordances are what drive media effects, not the technology itself. We certainly understand that there may be aspects of modern technology that could possibly affect the intervention’s efficacy, such as screen resolution, audio fidelity, and computing speed. However, the concepts investigated are rooted in basic psychological principles; therefore, we believe findings from this study are generalizable to future technologies, such as smartphone applications and virtual reality.

On a related note, the intervention focused on condom usage as a primary strategy for HIV risk-reduction, which was appropriate for the time period. More modern interventions are likely to incorporate PrEP as a risk-reduction strategy. We believe the concepts of norms, character identification, and behavioral intentions are universal enough to be easily adapted to a variety of prevention strategies, including the use of PrEP. While, our intervention focused on condom preparedness and self-regulatory skills such as avoiding alcohol and drugs prior to sexual encounters, these specific risk-reduction strategies are not necessarily a great fit for an intervention aimed at promoting the uptake and sustained usage of PrEP. Therefore, our study is limited in that we do not know whether the findings would be replicated in the context of PrEP education.

Similarly, there are other variables we could have included to assess risk. Participants might choose to avoid anal sex all together and engage in relatively safer acts such as oral sex or mutual masturbation. It is also possible that participants started the sexual encounter without a condom, but then chose to use one at some point during the encounter. Other interesting outcomes include verbal and non-verbal communication, condom negotiation attempts, and serosorting behaviors. Alternative outcomes of interest should be researched when possible.

Another limitation is related to sample characteristics. The sample was limited to young MSM, living in a relatively liberal community; most self-identified as gay. Future research is necessary to replicate these findings with older participants, as well as heterosexual participants, and participants who are less “out”.

This also relates to response bias and social desirability, which are possible concerns. Prior to enrollment, all prospective participants were told that this was a study about HIV and sexual health. However, we recognize that high-risk men are likely to be underrepresented in our sample. One explanation is that they might have feared to be judged for prior unsafe sex practices. Therefore, we can make no assumptions about whether our findings might generalize to this vulnerable population. Social desirability might be an issue as well. The study required participants to answer survey questions that were asked in-person by a trained interventionist. This may have resulted in under-reporting of risky behavior. However, we would expect an under-reporting effect to occur uniformly across the two conditions, and so any effects of social desirability should not interfere with or bias the general, overall patterns observed in the data. There may, however, be an attenuation of effect size.

Regarding attrition, it was revealed that drop-out rates were higher among younger and less educated participants. It has become clear that this is a limitation suffered across many different health intervention contexts. It is possible that the composition of the final sample was biased towards individuals who have successfully learned self-regulatory skills. As a result, our confidence in the generalizability of these findings to younger, less educated individuals is compromised.

Two additional limitations should be acknowledged. Participants were only exposed to the intervention on one occasion and the duration was relatively short (i.e., 30 minutes). However, it is important to highlight that the relatively short exposure produced significant effects at 3-month follow-up. Future research should replicate these findings following multiple exposures of the story-based intervention.

Conclusions

In sum, the present study assessed whether identification with characters, defined as perceiving another person as similar to oneself or as a person with whom one might have a social relationship with (Slater & Rouner, 2002), moderated the relationship between perceived sexual risk-taking norms and risk reduction intentions/behavior. Findings demonstrated that the more participants identified with characters, the weaker became the negative relationship between perceived sexual risk-taking norms and risk reduction behavioral intentions. Although identification with characters moderated the effect of perceived sexual risk-taking norms on behavioral intentions, identification with characters did not moderate the effect of perceived sexual risk-taking norms on sexual risk-taking behavior. Future studies should explore in more detail the impact of identification in interactive interventions in an attempt to expand the theory of normative social behavior and advance prevention intervention communication science.

Appendix

Rationale for Using Complete Case Analyses

Our justification for using “complete case analysis” (rather than multiple imputation) was guided by several statistical resources – and the rationale varied depending on the specific hypothesis in question.

For the first two hypotheses, we followed recommendations outlined by Jakobsen and colleagues (2017) in an article titled “When and how should multiple imputation be used for handling missing data in randomized clinical trials – a practical guide with flowcharts”. In this article, the authors discuss various methods for analyzing clinical trial data like ours, where it is common to have a large proportion of participants lost to follow-up. The authors recommend complete case analysis if only the dependent variable has missing values – because according to them, “No additional information will be obtained by, for example, using multiple imputation but the standard errors may increase due to the uncertainty introduced by the multiple imputation” (p. 4). This situation applies to hypotheses 1 and 2 because they only rely on the baseline data (i.e., no follow-up data used in these analyses). So multiple imputation of follow-up data would not be needed here because no follow-up data is needed to test these hypotheses.

For hypothesis 3, repeated measures ANOVA was used. We are aware of recently developed procedures for analysis of variance of multiply imputed data [see van Ginkel, J. R., & Kroonenberg, P. M. (2014). Analysis of variance of multiply imputed data. Multivariate behavioral research, 49(1), 78–91]. However, we viewed this procedure as problematic because “It involves both reformulation of the ANOVA model as a regression model using effect coding of the predictors and applying already existing combination rules for regression models” (p. 78). We did not view this as an appropriate option given the continuous nature of one of the independent variables included in the moderation analysis. Also, we thought it would be best to retain the original format and conduct a complete case analysis so that the analysis method would be consistent with the way in which the other hypotheses were analyzed.

For hypothesis 4, we used Hayes’ PROCESS macro for SPSS to analyze the data. We chose this statistical tool because of its ability to handle complex interaction effects while providing Johnson-Neyman estimations, which allowed us to take a more fine-grained look at the observed moderation effect – a primary aim of this project. According to Hayes’ website, “PROCESS requires complete data. It has no internal procedure for dealing with missing data other than listwise deletion. PROCESS does not integrate with the multiple imputation routines built into SPSS or SAS.”

For hypothesis 5, we used a negative binomial regression because the dependent variable was a “count” variable. Negative binomial regression analysis is recommended for count outcomes because the analysis accounts for the overdispersion common in these types of variables. Yes, SPSS allows for missing data analysis. However, to our knowledge, SPSS does not currently provide a reliable solution for multiple imputation when using the negative binomial (or poisson) regression techniques that are needed to accurately estimate parameters when dealing with overdispersed count variables (such as the number of times a participant engaged in risky sexual behavior during the prior three months). These limitations are discussed by Kleinke and colleagues [Kleinke, K., Stemmler, M., Reinecke, J., & Lösel, F. (2011). Efficient ways to impute incomplete panel data. AStA Advances in Statistical Analysis, 95(4), 351–373]. Weighing all of our options and taking these specific issues into account, our research team came to the conclusion that testing our hypotheses with complete case analysis would be the best option for presenting the most coherent report of our findings.