Research has shown that visual aesthetics is a key determinant of the perceived credibility of human-computer-interaction systems. However, there is limited research on which of the two dimensions of visual aesthetics (classical and expressive) has a stronger impact on the perceived credibility of persuasive health apps. Consequently, we conducted an empirical study among 669 subjects to investigate: (1) which of the two dimensions has a stronger influence on the perceived credibility of fitness apps modeling exercise behavior; and (2) the moderating effect of gender-based personalization. Our results show that, regardless of gender-based personalization, it is the perception of classical aesthetics that determines the judgment of the credibility of a persuasive health app. Expressive aesthetics has no significant influence on perceived credibility. Our findings underscore the need for designers of persuasive systems in the health domain to focus more on the classical dimension of aesthetics (orderliness, clarity and simplicity) when designing health applications for behavior change in order to enhance their perceived credibility.
• Human-centered computing → Human computer Interaction → HCI design and evaluation methods → User models Classical aesthetics; expressive aesthetics, credibility; fitness app; persuasive technology; gender, personalization; path model
HealthRecSys’18, October 6, 2018, Vancouver, BC, Canada. © 2018 Copyright for the individual papers remains with the authors. Copying permitted for private and academic purposes. This volume is published and copyrighted by its editors.
Credibility is an important design attribute that determines the
adoption and usage of persuasive systems that are designed to
promote behavior change [
Consequently, we carried out an empirical study in the fitness domain among 669 participants, using exercise behavior modeling as a case study. Behavior modeling is a persuasive strategy used in persuasive systems (such as fitness apps) to motivate behavior change. In behavior modeling, “an expert shows [a] person how to correctly perform a behavior, for example, in class or on video” (p. 382) [8]. The expert, also known as a social agent or coach, can be real or virtual. However, in computerized systems, for users to imitate the modeled behavior demonstrated by the virtual coach, there is need for the users to have trust in its visual design. This has become important due to the fact that a badly modeled exercise behavior can lead to a body injury when the user tries to replicate it. Thus, the behavior models have to be designed in a such a way that enhances their perceived credibility.
Based on our structural equation model (SEM), our results show that it is solely the perception of the classical dimension of aesthetics that determines the perceived credibility of the health app. Expressive aesthetics has no significant influence on perceived credibility. Our finding underscores the need for designers of persuasive health systems, such as behavior models in fitness apps, to focus more on classical aesthetics (cleanness, orderliness, clarity and pleasantness) than expressive aesthetics as a means to increase the perceived credibility of their persuasive applications.
The rest of this paper is organized as follows. Section 2 focuses on background and related work. Section 3 focuses on the research method. Section 4 focuses on the result. Section 5 focuses on the discussion. Finally, Section 6 focuses on the conclusion.
This section provides a brief overview of the concepts of aesthetics and credibility, their relationships, and a review of related work. 2.1
Aesthetics refers to sensory pleasure and delight derived from the perception of an object or artifact. It can be either a property of an object or perceived, or both [9]. Specifically, in this paper, we are concerned with the subjective judgment of aesthetics, which is in the eye of the beholder (i.e., the user). In HCI, visual aesthetics— defined as the pleasing appearance of HCI artifacts [10], user interfaces and systems—is regarded as a multi-dimensional construct composed of classical aesthetics and expressive aesthetics [11]. Classical aesthetics entails the historic notion of beauty: “simplicity,” “orderliness,” proportion, “symmetry,” etc. These attributes of classical aesthetics highly correlate with perceived usability [11], [12]. Thus, it is described by words such as “well-organized,” “clear,” “clean,” etc. On the other hand, expressive aesthetics has to do with the expressive power of the HCI system designer, which reflects his/her creativity and originality. Thus, it is described by words such as “original,” “creative,” “fascinating,” “sophisticated,” etc. [11]. 2.2
Credibility refers to the believability of a HCI system [
A substantial amount of research has been carried out with respect to the relationship between perceived aesthetics and perceived credibility in the evaluation of HCI systems. Robins and Holmes [7] examined the link between the perceived aesthetics and the perceived credibility of websites. They found that a high-aesthetic website is more likely to be perceived credible than a lowaesthetic website. Alsudani and Casey [14] conducted a study on the judgment of the credibility of recruitment agency websites. They found that aesthetics characteristics such as unity in design, balance and harmony significantly influence perceived credibility. Fogg et al. [15] conducted a large scale study on the judgment of web credibility. They found that about 46% of participants’ comments on the credibility of various websites cutting across different domains were about design look—which is related to visual aesthetics [4]. 1 In this paper, we are specifically investigating the moderating effect of the gender of the behavior model as well as the gender of the observer.
Oyibo and Vassileva [4], [5] investigated which of perceived aesthetics and perceived usability has a stronger effect on perceived credibility in the tourism domain. They found that perceived aesthetics has a stronger effect than perceived usability on perceived credibility. Furthermore, Oyibo et al. [6] investigated which of the two dimensions of perceived aesthetics has a stronger effect on perceived credibility in the same domain. They found that, irrespective of age and gender, the classical dimension has a stronger effect than the expressive dimension on perceived credibility.
However, they are yet to be studies in the health domain focused on uncovering the more influential dimension of visual aesthetics on the perceived credibility of persuasive systems. The reviewed studies in the first paragraph did not address this research question. Moreover, Oyibo et al.’s [6] findings in the second paragraph are in the tourism domain, which may not generalize to the health domain. Thus, the main thrust of this paper is to investigate which of the two dimensions of visual aesthetics has a stronger influence on perceived credibility. The secondary objective of this paper is to investigate whether the previous finding in the tourism domain (classical aesthetics has a stronger influence than expressive aesthetics on perceived credibility) generalizes to the health domain. 3
This section focuses on our research question, research model, measurement scales and participants’ demographics. 3.1
The aim of this study is to answer the following research questions on the relationship between the two aesthetics dimensions and perceived credibility in the health domain: 1. 2. 3.
What drives the perceived credibility of fitness apps featuring exercise behavior models: classical aesthetics or expressive aesthetics, or both? Are the interrelationships among all three design constructs moderated by the gender of the behavior model and the gender of the observer? Does the finding that “classical aesthetics has a stronger influence than expressive aesthetics on perceived credibility in the tourism domain” [6] generalize to the health domain? 3.2
To address our research questions, we designed a fitness app prototype featuring exercise behavior models. We considered race (black/white), gender (male/female) and exercise-type (pushup/squat) in the design of the behavior models (Figure 1).1 This resulted in eight versions. Push-up and squat were chosen because they are among the most commonly featured exercise-types in the fitness apps market. We randomly presented one of the eight designs to each participant and asked them to answer questions on classical and expressive aesthetics, and perceived credibility.
To measure the three constructs of interest, we used existing validated instruments. For classical and expressive aesthetics, we used the scale proposed by Lavie and Tractinsky [11], as adapted by van Schaik and Ling [17]. For credibility, we used a single-scale item scale, which research has shown is as reliable as multi-item scales [18]. All three scales ranged from “Strongly Disagree (1)” to “Strongly Agree (7).” Table 1 shows their respective items. In the questionnaire, the overarching question preceding all of the items (randomized) is, “Please rate the visual design above on the following criteria based on your first impression.” Based on prior literature, we formulated three hypotheses, which are represented in the research model shown in Figure 2. All three hypotheses (H1, H2 and H3) are based on the findings by Oyibo et al. [6] in the tourism domain. Specifically, they found that classical aesthetics has a stronger influence than expressive aesthetics on perceived credibility. However, in the health domain, we adopted an exploratory approach to investigate which of the two aesthetics dimensions has a stronger influence on perceived credibility. We also used an exploratory approach to investigate whether the three interrelationships in Figure 2 are moderated by the gender of the behavior model and the gender of the observer. The survey was approved by the Behavioral Research Ethics Board of the University of Saskatchewan. Thereafter, it was posted on Amazon Mechanical Turk to recruit participants resident in Canada and United States. Each of the participants who completed the survey was compensated with $0.6 for their time. Table 2 shows the demographics of the valid participants after cleaning: 48.9% females and 51.1% males.
In this section, we present our measurement models, structural models, and multigroup analysis results obtained from the SEM analysis using “PLSPM” package in R [19]. 4.1
We evaluated the inner models to ensure the reliability of our structural models [20]. Specifically, the indicator reliability metric (outer loadings) and internal consistency reliability metric (DG.rho) were greater than 0.7. Similarly, the convergent validity (Average Variance Extracted) was greater than 0.5. Finally, with respect to discriminant validity, no construct’s indicator loaded higher on any other construct than itself. 4.2
Figure 3 shows the global models for the male and female observers. The goodness of fit (GOF) represents how well the model fits the data, which is moderate: 56% for males and 66% for females. The coefficient of determination (R2) represents the amount of variance of perceived credibility explained by the model: 43% for males and 56% for females—also moderate. The path coefficient (β) represents the direct effect of one construct on another. Overall, the direct effect of classical aesthetics on perceived credibility is significant for males (β = 0.58, p < 0.001) and for females (β = 0.65, p < 0.001). However, the direct effect of expressive aesthetics on perceived credibility is not significant for both males (β = 0.12, p = n.s) and females (β = 0.13, p = n.s). Moreover, we built subgroup models based on the following criteria (dyads): 1. 2. 3. 4.
Males evaluating male behavior models (MM) Males evaluating female behavior models (MF) Females evaluating female behavior models (FF)
Females evaluating male behavior models (FM) MM and FF represent gender-based personalization, while MF and FM represent contra-tailoring. As shown in Figure 4, the βs, GOFs and R2s, to a large extent, confirm the findings at the global level. For example, in the global model, we see that the relationship between classical aesthetics and perceived credibility is significant (p < 0.001), while that between expressive aesthetics and perceived credibility is not significant. This is perfectly confirmed at the subgroup level, with the relationship between classical aesthetics and perceived credibility for FM being the strongest (β = 0.82, p < 0.001) and that for MM being the weakest (β = 0.46, p < 0.001).
p n.s n.s n.s p .05 n.s n.s p n.s n.s n.s p n.s n.s n.s
Classical Aesthetics→Credibility Expressive Aesthetics→Credibility Classical Aesthetics→Expressive Aesthetics The MGA for male vs. female observers shows that that there is no significant difference between both genders of observers in their evaluation of the behavior models. However, at the level of personalization, the MGA for MM vs. MF shows that there is a significant difference between both groups in the relationship between classical aesthetics and perceived credibility (p < 0.05). Moreover, there is a significant difference between MF and FM with respect to the relationship between classical aesthetics and expressive aesthetics (p < 0.05).
We presented a path model to investigate: (1) which of the two dimensions of visual aesthetics has a stronger influence on the perceived credibility of the visual design of behavior models aimed at changing behavior; and (2) the moderating effect of gender-based personalization. In this section, we discuss the validations of two of our main hypotheses and their implications, our contributions and the limitations of our study.
Our path analysis (Figures 3 and 4) shows that, regardless of the gender of the observer, two of our three hypotheses are supported: classical aesthetics has a significant influence on perceived credibility (H1); (2) classical aesthetics has a significant influence on expressive aesthetics (H3); but (3) expressive aesthetics has no significant influence on perceived credibility (H2). The validation of H1 suggests that users rely on the classical dimension—and not the expressive dimension—of visual aesthetics in the evaluation of the perceived credibility of behavior model design. This finding replicates that of Oyibo et al. [6] in the tourism domain, in which the authors found that the classical dimension of aesthetics has a stronger influence than the expressive dimension on perceived credibility. The confirmation of H1 in the health domain suggests that visual aesthetics attributes, such as cleanness, clarity and pleasantness (encompassed in simplicity and orderliness), have more impact on the perceived credibility of the behavior model design than complex attributes (e.g., fascination, sophistication, creativity, etc.). This is well reflected in the following participant’s comment: “It's really clear, it is very credible because it's an actual exercise and it shows which muscles are working. It's not creative because it just shows a woman doing squats” [P116, squat]. Despite commenting that the squat behavior model design is not creative (low in expressive aesthetics), participant 116 still considered the design “very credible” owing to its perceived clarity (classical aesthetics). Another comment that reflects participants’ judgment of credibility based on mainly classical aesthetics is: “The squat video is self explanatory and clean. Perfect for my needs” [P96, squat].
A possible explanation for why classical aesthetics has a
stronger influence on perceived credibility than expressive
aesthetics does is that fitness apps modeling behavior are more of
utilitarian than hedonic systems. Utilitarian systems provide
instrumental value to the user. They “address tasks and activities where
user motivation in using the system is driven by the expectation of a
reward or benefit external to his interaction with the system” (p.
620) [
With respect to the personalization of the gender of the behavior models to the user (Table 3), the significant difference we found between the group that evaluated behavior models of the same gender and the group that evaluated behavior models of the opposite gender is with respect to MM vs. MF. The effect of classical aesthetics on perceived credibility is significantly stronger for the opposite-gender MF dyad (β = 0.67, p < 0.001) than for the samegender MM dyad (β = 0.46, p < 0.001). This finding seems to be replicated in the MGA for FF (β = 0.56, p < 0.001) vs. FM (β = 0.82, p < 0.001)—only that the numerical difference between both groups is not significant. Despite the difference between FF and FM is not statistically significant, the significant difference between MM and FM seems to suggest that users rely on classical aesthetics more in the judgment of credibility of the behavior model design when the behavior models are contra-tailored. In contrast, users seem to rely on expressive aesthetics more in the judgment of credibility of the behavior model design when the behavior models are personalized. For example, with respect to MM vs MF, we see that the effect of classical aesthetics on perceived credibility is significantly stronger for MF (β = 0.67, p < 0.001) than for MM (β = 0.46, p < 0.001), while the effect of expressive aesthetics on perceived credibility is non-significantly stronger for MM (β = 0.15, p = n.s) than for MF (β = 0.12, p = n.s). Similarly, we see that the effect of classical aesthetics on perceived credibility is non-significantly stronger for FM (β = 0.82, p < 0.001) than FF (β = 0.56, p < 0.001), while the effect of expressive aesthetics on perceived credibility is non-significantly stronger for FF (β = 0.20, p = n.s) than MF (β = 0.00, p = n.s). However, given the nonreplication of the significant difference in the classical aestheticsperceived credibility relationship between MM and MF with that between FF and FM, there is a need for further research into the above highlighted proposition. 5.3
Based on the significant relationship between classical aesthetics and perceived credibility (p < 0.001), we recommend the following general design guideline:
When designing persuasive systems with utilitarian value or benefit, such as fitness applications featuring behavior models (virtual coach) aimed at changing behavior, HCI designers should focus on enhancing classical aesthetics (cleanness, orderliness, clarity and simplicity) to enhance the perceived credibility of the fitness applications.
Our main contribution to the body of knowledge is that we replicated the above prior finding in the tourism domain [6] in the health domain, indicating that the hypothesis—classical aesthetics has a stronger effect than expressive aesthetics on the perceived credibility of utilitarian systems—is likely to cut across domains. Aside from the relationship between classical aesthetics and perceived credibility, we showed that there is a significant relationship between the two dimensions of visual aesthetics, which confirms our third hypothesis (H3). This suggests that the higher users’ perception of a fitness app to be classically aesthetic, the higher will be their perception of the app as expressively aesthetic. However, a high perception of the app as expressively aesthetic does not result in a high perception of the app as credible, indicating that expressive aesthetics does not play a mediating role between the effect of classical aesthetics on perceived credibility. 5.4
Our study has some limitations. The main limitation is that our findings are based on users’ perception (i.e., self-report). As such, we cannot guarantee that they will generalize to an actual context in which the users have to use the fitness app and answer the posed questions thereafter. The second limitation of our study is that most of the participants are from Canada and the United States. This may threaten the generalization of our findings to other demographics. Thus, in future work, we look forward to investigating our findings among other demographics as well as by using an actual fitness app. [19] [20]