This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Serious Games, is properly cited. The complete bibliographic information, a link to the original publication on https://games.jmir.org, as well as this copyright and license information must be included.
Serious games for the training of prevention behaviors have been widely recognized as potentially valuable tools for adolescents and young adults across a variety of risk behaviors. However, the role of agency as a distinguishing factor from traditional health interventions has seldom been isolated and grounded in the persuasive health communication theory. Fear appeals have different effects on intentions to perform prevention behaviors depending on the immediacy of the consequences. Looking into how to increase self-efficacy beliefs for health behavior with distant consequences is the first step toward improving game-based interventions for adverse health outcomes.
This study aimed to investigate the effect of agency on self-efficacy and the intention to drink less alcohol in an interactive digital narrative fear appeal. Furthermore, the communicated immediacy of threat outcomes was evaluated as a potential moderator of the effect of agency on self-efficacy.
A web-based experimental study was conducted with university students (N=178). The participants were presented with a fear appeal outlining the consequences of excessive alcohol use in a fully automated web-based interactive narrative. Participants either had perceived control over the outcome of the narrative scenario (high agency) or no control over the outcome (low agency). The threat was either framed as a short-term (high immediacy) or long-term (low immediacy) negative health outcome resulting from the execution of the risk behavior (drinking too much alcohol).
A total of 123 valid cases were analyzed. Self-efficacy and intention to limit alcohol intake were not influenced by the agency manipulation. Self-efficacy was shown to be a significant predictor of behavioral intention. The immediacy of the threat did not moderate the relationship between agency and self-efficacy.
Although agency manipulation was successful, we could not find evidence of an effect of agency or threat immediacy on self-efficacy. The implications for different operationalizations of different agency concepts, as well as the malleability of self-efficacy beliefs for long-term threats, are discussed. The use of repeated versus single interventions and different threat types (eg, health and social threats) should be tested empirically to establish a way forward for diversifying intervention approaches.
ClinicalTrials.gov NCT05321238; https://www.clinicaltrials.gov/ct2/show/NCT05321238
Excessive alcohol abuse is an alarmingly large cause of death in European Union member states, as well as in Norway, Switzerland, and the United Kingdom [
We argue that the use of interactive digital narratives (IDNs) could be a promising tool to enable the training of preventive behavior through direct interaction with the narrative as a protagonist. In contrast to traditional passive narratives, an IDN involves computer-based interactive storytelling that allows the user to intentionally influence a nonlinear narrative. This means that the interactor is not only experiencing fictional reality but can also be part of it by being able to make meaningful decisions that affect narrative outcomes [
Fear appeals have been widely assessed and used by many health communication researchers for interventions that focus on behavioral changes. Fear appeal communication aims to stimulate behavior change through the provision of a threat that causes a fear response. The message receiver is assumed to be motivated to resolve the arousal caused by the fear response and is presented with a solution to avert the threat. If the appeal is successful, the message perceiver adopts the prevention behavior. The term
Fear appeals are commonly used in many public health campaigns, making use of mostly text-based or pictorial stimuli to elicit fear responses. One of the most widely applied, as well as well-researched, theories on the underlying process of fear appeal communication is the extended parallel process model (EPPM) [
Witte [
Fear appeals used in public campaigns often make use of simple narratives through pictures and textual information that are aimed at persuading the message receiver by providing a way of averting the threat (ie, increasing efficacy beliefs). These simple narratives (eg, a picture of a man losing his leg because of smoking) heighten the relevant affective responses for persuasion (ie, fear and compassion), leading to more persuasive fear appeals [
In their review of narrative interventions, De Graaf et al [
The term
Many investigations have provided interactive elements in their message design (see the studies by Winskell et al [
Although seldom investigated experimentally, operationalizing agency through meaningful interaction seems promising in how this could relate to self-efficacy for threat prevention behaviors. Through the experience of agency, prevention behaviors can potentially be trained, and self-efficacy beliefs can be strengthened. Agency itself is only concerned with the experience of meaningful interactions within the narrative, whereas self-efficacy is concerned with the transfer of beliefs about behaviors in the real world. Self-efficacy is defined as the belief of an individual that they can execute a behavior recommended by a fear appeal message [
Real-life self-efficacy beliefs are amenable through experimental manipulation in persuasive communication [
In line with this, in this study, we expect that self-efficacy beliefs about real-life prevention behaviors can be affected by the provision of agency in an interactive narrative fear appeal. Although it is only a stepping stone in examining the process of persuasive IDNs for behavior change, the role of interaction as a key differentiator in the experience between passive and active media has rarely been investigated in a highly controlled narrative environment.
Although fear appeals for immediate threats have been shown to be effective, there is less evidence showing EPPM’s effectiveness for temporally distant threats [
As discussed by Klimmt and Hartmann [
This study aimed to assess the effect of agency and threat immediacy on self-efficacy and, consequently, the intention to perform prevention behaviors. The following hypotheses were tested using an IDN in this study:
Hypothesis 1: Higher agency in the narrative progression of the IDN fear appeal results in higher perceived self-efficacy for prevention behavior.
Hypothesis 2: The effect of agency on self-efficacy is moderated by framing the threat as an adverse health outcome with either high or low immediacy.
Hypothesis 3: Higher perceived self-efficacy will lead to a higher intention to perform the target behavior.
Data were collected from November 26 to December 18, 2020. The hypotheses, study design, and planned analyses were preregistered using the preregistration platform provided by the Wharton Credibility Lab. The independent variables were agency (low or high) and threat immediacy (low or high). Self-efficacy and behavioral intention served as the dependent variables in this study.
This study was approved by the Research Ethics and Data Management Committee (REDC) of the Tilburg School of Humanities and Digital Sciences (TSHD) and was conducted as a 2×2 between-subjects experimental study (reference number: REDC 2020.141).
A total of 178 participants were recruited from the Human Subject Pool of Tilburg University. All participants were enrolled in a master’s or bachelor’s degree program and received 0.5 course credits for participation in the study. Participants were allocated equally across all 4 conditions, and the study was conducted entirely on the web. No personally identifiable data (including IP addresses) were collected to ensure the anonymity of all participants.
Data were collected in 2 phases. The first half of the participants (95/178, 53.4%) who enrolled in the study were allocated to the high-agency condition. After the completion of data collection for high-agency participants, the second half (83/178, 46.6%) of the participants were assigned to the low-agency condition. The low-agency narratives were matched in terms of narrative content to the narratives of the high-agency group who took part in the study before them. All participants were randomly assigned to either the high- or low-immediacy conditions. Given the short time frame of data collection and the absence of events that could have an influence on drinking-related behavior (eg, public events), we assumed the time gap between the participation of the first and second halves of the participants to not affect our measures. Sufficient computer literacy and English language capabilities were assumed among the student population used in this study.
Participants were included in the analysis if they (1) were aged >18 years, (2) spent >480 seconds completing the survey, (3) passed both attention checks relating to narrative content, (4) correctly identified their experimental condition in the manipulation check, (5) did not choose to
The final sample used for analysis comprised 123 participants, with 59 (48%) in the high-immediacy condition and 64 (52%) in the low-immediacy condition. The split between high and low agency was almost equal to 49.6% (61/123) of participants in the high-agency group and 50.4% (62/123) of participants in the low-agency group.
The sample was representative of a student sample with a mean age of 21.4 (SD 3.086) years. Most held either a high school (53/123, 43.1%) or bachelor’s diploma (55/123, 55.3%), with few (2/123, 1.6%) participants having completed their master’s degree. There was a large gender imbalance in the sample, with approximately 72% (88/123) of the sample being female.
The entirety of this study took place on the web. Participants signed up through the Tilburg University Human Subject Pool. They were first presented with an information letter and then informed consent before agreeing to participate in the study. Next, demographic data (age, gender, and education) were collected, and participants were instructed to
The narratives showed a fictional scenario in which the participant was presented with a fear appeal message, together with a pictorial fear stimulus (including severity, susceptibility, and response efficacy statements). Pictures were added throughout the story to (1) improve transportation into the story world, which has been shown to improve narrative persuasion [
The first part of the fear appeal message (
Excerpts from the narrative fear appeal comprising (1) severity and susceptibility, (2) response efficacy, (3) self-efficacy, and (4) threat avoidance messages in the high-immediacy condition.
The scenario of a house party was chosen as it is a familiar example to our target population and should, therefore, be easy to imagine and connect to real-life risk situations (to enable behavior transfer). Choosing organ failure as a threat made it possible to keep the threat congruent across different immediacy conditions while only changing the time frame. The threat needed to be severe to cause sufficiently high levels of perceived fear, whereas the pictorial stimuli needed to be congruent with the threat without invoking disgust. Disgust has been shown to potentially undermine appeal content when combined with fear stimuli [
A total of 4 different versions of the narrative fear appeal were created (1 per condition).
As shown in
Overview of narrative nodes and progression for the high-agency condition. SE: self-efficacy.
Every participant received only 1 self-efficacy message, as subsequent choices were no longer related to the prevention behavior. The ending of the story highlights the meaningfulness of turning down the drink by showing the adverse consequences that drinking could have had that night (
To manipulate outcome immediacy, the fear appeal at the beginning of the narrative was presented as organ failure because of excessive alcohol consumption over a short period (high immediacy) or continuous consumption of large amounts of alcohol over a period of months (low immediacy). Similarly, the threat aversion message (
Demographic data were collected to account for appropriate randomization between the conditions. Participants were asked to indicate their age, gender, and educational background.
Self-efficacy beliefs were measured by adapting the item stems from Shi and Smith [
Behavioral intention was measured using items adapted from Fisher et al [
Items for the fear measure were taken from the Witte [
Items for perceived susceptibility and severity were adapted from Shi and Smith [
Severity, susceptibility, and response efficacy items were measured with 3 items each on 7-point Likert scales from
The reliability analysis of severity (3 items; mean 6.333, SD 0.744; Cronbach α=.571), susceptibility (3 items; mean 2.832, SD 1.216; Cronbach α=.744), and response efficacy (3 items; mean 4.640, SD 1.308; Cronbach α=.792) showed moderate scale reliability.
To validate agency manipulation, a scale was included to measure perceived agency after the participants experienced the narrative. The items for perceived agency were adapted from Fendt et al [
To check whether participants perceived a difference in the immediacy of the presented threat, they were asked to recall the content of the threat message presented at the beginning of the narrative. Participants had to indicate whether the protagonist died because of
To account for potential individual differences between participants, disinhibition, drinking frequency, alcohol dependence, and perceived control over drinking behavior were measured. Behavioral inhibition, measured by a subscale of the Sensation Seeking Scale-V by Zuckerman [
The subscales for measuring frequency and perceived control regarding excessive alcohol consumption were taken from Carrera et al [
Frequency was measured by asking participants how often they
As alcohol dependence is difficult to measure accurately in student populations, the B-YAACQ was developed by Kahler et al [
The conceptual model shown in
The a priori power calculation assumed a medium effect size (
Moderated mediation model for the relationship between agency and behavioral intention. H:hypothesis.
For the main data analysis, a 2-way ANOVA was preregistered to test the effects of agency and immediacy on self-efficacy. However, we later realized that a more advanced model that also takes into account the possible moderating effect of immediacy provides a better fit for the data; hence, we report the analysis in the following sections. It should be noted that the preregistered ANOVA yielded essentially the same outcome.
A total of 123 cases were analyzed. We first examined our control variables to ensure that their scores were equally distributed across all 4 conditions.
Three 1-way ANOVAs were conducted to test whether group assignment had an effect on the perceived control over alcohol intake, inhibition, or age. We obtained no significant results for group membership on perceived control (
An independent-sample
The analysis of path a1 indicates that the different levels of agency (low or high) are not significant predictors of the self-efficacy measure (path a1:
Effects overview of the hypothesis tests for the moderated mediation model.
|
Estimate (SE) | 95% CI | |||||||
|
|||||||||
|
Agency→self-efficacy | −0.611 (0.517) | −1.182 (119) | .24 | −1.634 to 0.412 | ||||
|
Self-efficacy→behavioral intention | 0.576 (0.112) | 5.135 (120) | <.001 | 0.354 to 0.798 | ||||
|
|||||||||
|
Agency×immediacy→self-efficacy | 0.215 (0.352) | 0.610 (119) | .54 | −0.483 to 0.913 | ||||
|
|||||||||
|
Agency→behavioral intention | −0.062 (0.189) | −0.325 (120) | .75 | −0.437 to 0.314 |
Additional exploratory analyses were conducted to contextualize the null effects. A median split was conducted by analyzing the effect of agency on self-efficacy for participants scoring high on susceptibility to ascertain whether the appeal or the pre-existing characteristics of the sample caused a potential floor effect for the susceptibility measure. The adjusted sample (75/123, 61%) showed no significant relationship between the agency and self-efficacy measures (
Although the manipulation of agency was found to be successful, this study found no effect of agency on perceived self-efficacy, and this relationship was also not influenced by the framing of the threat as a short-term (high immediacy) or long-term (low immediacy) adverse health outcome. A significant effect was found for the relationship between self-efficacy and behavioral intention. To point forward for future studies, the null findings of this study need to be discussed with regard to their contribution by contrasting the different approaches taken for this study.
In this study, agency had no significant effect on the self-efficacy perceptions of the participants. Agency was operationalized as the one-time execution of an action with a meaningful impact on the outcome of the narrative. To achieve this, a foldback structure was used to ensure that participants would encounter the same narrative content and be presented with a complete fear appeal where the prevention behavior was executed, and a self-efficacy message could be displayed. Although the manipulation check showed a difference in perceived agency between high- and low-agency conditions, the perceived agency might still not have been impactful enough to cause a change in the self-efficacy measure. Owing to the nonexistence of related work on the effects of agency on self-efficacy, these results are difficult to compare with other operationalizations of agency in the field. However, looking at the different theoretical conceptualizations of agency, we can derive possible explanations.
The manipulation in this study could conceivably be seen as a low-agency condition if agency is defined as the balance of affordances and constraints provided by the system [
Heightening perceived completeness could be achieved, given the successful use of illusory agency, presenting options that are perceived to be more complete even if they fold back to the same outcomes, or by loosening authorial control in more open experiences such as a sandbox game that enables emergent storytelling. The question is whether the elicitation of self-efficacy through agency is reliant on control over the chronology of message components or whether heightened agency through emergent storytelling can make up for the loss in control by providing a closer link between the protagonist and the message receiver in terms of perceived completeness of the available actions. Taking this a step further, there is evidence to form the assumption that agency does have an effect on persuasion, even if the agency manipulation is completely decoupled from the persuasion attempt and simply induced as a state before the intervention (eg, see the study by Damen et al [
The wide range of used agency definitions and their heterogeneous operationalization make generalizations about agency effects in health interventions difficult. For future studies, the plural modality of agency should be taken into account [
In this study, we found no significant effect of threat immediacy on the relationship between agency and self-efficacy. Organ damage was chosen as the threat as it can be a result of either short-term or long-term risk behavior and was henceforth communicated as a consequence of a single night or multiple months of drinking. There is ample work evaluating the effect of fear appeals on one-time versus repeated risk prevention behaviors, showing that appeals advising the one-time execution of behaviors are more likely to succeed [
In most IDN research, as well as game studies, behavior transfer between digital and real-life behavior is often assumed to be high when the digital execution matches that of real-life analog behavior. The prevention behavior for the threat in this study does not constitute a one-time behavior, as it needs to be executed multiple times for the threat to be averted. With regard to immediacy in this study, there is, therefore, a mismatch between both our short- and long-term threats and our digital narrative in which the prevention behavior (rejecting a drink) is only executed once. Although a single appeal message can influence self-efficacy for repeated prevention behaviors (eg, see the study by Smith and Stutts [
A possible direction for future studies would be to make the long-term, as well as the short-term, impact of risk behavior more salient through the provision of repeated interventions. As discussed by Shi and Smith [
The null effects for agency and threat immediacy could also result from the low perceived susceptibility of the target population (young adults) to the threat used in this study (organ failure because of excessive drinking). The so-called
The issue of perceived threat susceptibility can be further explained by our exclusion criteria. The criteria specifically excluded participants who showed signs of alcohol dependence when they might actually have been participants for whom the threat would be most relevant and the preappeal efficacy would be low. However, as the aim is to increase the effectiveness of early intervention approaches, more effective ways of eliciting prevention behaviors for normal populations should still be investigated more thoroughly.
An alternative approach could be the use of appeals that do not focus on health threats to increase susceptibility as a prerequisite for successful narrative persuasion. Health threats themselves have been the main focus of health-related interventions and are effective for older cohorts; however, this might not always be the case for younger populations. As suggested by Pechmann et al [
This study did not use a pre- or postmeasures design to account for the potential influence of baseline beliefs already present in participants before the appeal message was presented. However, it has to be noted that disinhibition, drinking frequency, alcohol dependence, and perceived control over drinking behavior were measured to account for individual differences between participants. These were measured after the appeal message so as to minimize the influence on the responses to the IDN appeal. Although we believe that these constructs relate to either stable traits or general behavior and therefore should not have been influenced by the manipulation, we cannot fully discount the potential influence of the IDNs on these constructs.
Furthermore, it should be noted that the current sample was predominantly women (88/123, 72%). Previous studies have shown gender differences in the effectiveness of fear appeals, which might have influenced the results. In general, this sample might not be representative of the general population.
Finally, this study fell short of its goal of collecting a total of 140 valid samples, leading to a potential shortcoming in statistical power, given the complex study design. Furthermore, it could even be that the authors assumed a medium effect size for a single intervention transferring virtual behaviors to real-world beliefs and behavioral intentions.
Finally, the context of this study must be mentioned. The study was conducted during the COVID-19 pandemic, which limited the participants’ exposure to risk situations and could have skewed the general perception of the appeal as personally relevant. Although there was no active lockdown in place at the time of data collection, there was a limit to the number of people allowed to gather more generally. This made the otherwise common risk environments that facilitate drinking, such as parties, almost nonexistent. As social facilitation of drinking is one of the core determinants of risk behavior related to excessive alcohol use for college students [
In this preregistered study, no significant effects were found for agency on self-efficacy and behavioral intention. In addition, we observed no effect for the influence of threat immediacy on the relationship between agency and self-efficacy.
The multitude of conceptual distinctions of different forms of agency and the different ways of integrating them into IDNs provides a challenge that needs further evaluation to affect self-efficacy and, more generally, behavioral change. There is a need to empirically test the difference between conceptually different agency concepts to establish their impact on persuasiveness in narrative health interventions. Although there is evidence to assume the effects of interaction on antecedents of narrative persuasion (eg, identification and transportation), more work is needed to understand how different agency conceptualizations directly affect the processing of health messages. In addition, more work is needed that empirically contrasts passive and interactive interventions to isolate the effects. Much of the previous work done in this field use fully-fledged serious game experiences without passive controls, making it difficult to ascribe effects to single factors in such complex systems. With regard to the theoretical assumptions underlying narrative fear appeals, the personal relevance of the content should be ensured through pilot testing before the deployment of the intervention. Without creating the personal relevance of the narrative fear appeal, the intended effects could be undermined because of a lack of perceived susceptibility to the threat. Understanding how different kinds of threats are perceived by different target populations should be investigated more thoroughly to adapt interventions more effectively. Furthermore, the effect of intervention repetition for different kinds of prevention behaviors is poorly understood, making generalizations across different fear appeal threats difficult. A systematic approach contrasting one-time and repeated prevention behaviors for different kinds of threats could establish a clearer picture of the potential necessity of adaptations for threats that are the result of repeated risk behaviors.
In conclusion, despite the null findings for the effects of agency on self-efficacy, this study highlights the potential for further exploration of agency concepts and threat types in terms of their embedding into a narrative intervention.
CONSORT-EHEALTH checklist (V 1.6.1).
bias-corrected and accelerated
Brief Young Adult Alcohol Consequences Questionnaire
extended parallel process model
interactive digital narrative
HE designed the study and wrote the manuscript. LNvdL, RvE, and EK supervised the entire process and provided feedback on the study and the written manuscript. All authors have read and approved the final manuscript.
None declared.