Internet-delivered interventions can effectively change health risk behaviors and their determinants [1,2], but the actual use of these interventions by the target group once they access the website is very low [3,4]. For example, server statistics of a web-based intervention promoting heart-healthy behaviors showed that 285,146 visitors from unique IP addresses accessed the home page over a 36-month period, but 56.3% of visitors left the intervention website within 30 sec [5]. This finding touches upon the critical issue in Internet-delivered interventions: How can these interventions have a public health impact if people use the actual intervention so briefly? Therefore, it is relevant and necessary to focus on factors related to use of an intervention once people arrive at the intervention website (ie, website use) [6]. These factors relate to the visitor (eg, people’s motivation to be healthy [7,8]) as well as the intervention website (eg, offering tailored information [9-11]).

The content of the website is important [12], but the specific characteristics of the website itself are also important. A previous study stressed the need for future research to systematically manipulate website characteristics and, subsequently, to link these manipulations to website use [13]. The current study follows this recommendation and focuses on user control as a website characteristic.

User control covers the voluntary and instrumental actions of the website visitor [14,15]. This is an important characteristic of a website that shapes the two-way online interaction and the exchange of information with website visitors resulting in a user experience [16,17]. User experience refers to what a person thinks and feels during and after exposure to a website [18]. The main idea is that a positive user experience increases website use. User experience consists of cognitive perceptions and affective perceptions [19]. Cognitive perceptions are rational in nature and induced by utilitarian or cognitive motives. Affective perceptions are emotional in nature and induced by hedonic or affective motives [20].

The key user perceptions are efficiency, effectiveness, enjoyment, and active trust [19]. These terms are derived from other fields, such as e-commerce. Although they can have a different meaning within public health, we chose to use the same terminology as in previous studies for consistency and to avoid confusion. Efficiency refers to easy search and access of information provided; effectiveness refers to the quality of that information (eg, in terms of relevance) [21]. These cognitive perceptions have parallels with perceived ease of use and perceived usefulness in the technology acceptance model, but are applicable in a broader context [22]. The idea that a positive user experience increases website use does not only apply to cognitive perceptions, but also to affective perceptions [23]. These affective perceptions are often referred to as enjoyment [24]. Active trust refers to the confidence in acting on the provided information on a website, which can result in increased website use [25]. The previous study that served as the basis for the present one consistently demonstrated that effectiveness and enjoyment both had a positive effect on intention to use, which was mediated by active trust [13]. Efficiency did not have an effect, but this could be explained by the goal of the websites being used in that study, which were aimed purely at behavioral change instead of providing information only (such as the websites used in the present study). Therefore, in this study, efficiency, effectiveness, and enjoyment were expected to increase website use (Hypothesis 1), which was expected to be mediated by active trust (Hypothesis 2).

To gain more insight into how user control can help to increase website use, it is necessary to study the effect of user control on user perceptions. Since user control provides the ability to explore and to understand the structure of a website [26,27], it allows visitors to be involved in the cognitive processing of information. This is closely related to the concept of efficiency (ie, “easy search and access of information”) and indicates a certain level of user involvement. Previous research revealed that the positive effect of freedom of choice (ie, high user control) on preference regarding websites was mediated by efficiency [28].

One of the most common issues for websites is the lack of user control [29], which might lead to a reactance effect [30]: A constrained freedom of choice results in a negative effect on preference for that website [28]. The visitor, however, can still decide whether or not to keep using the website. When a visitor has the freedom to decide whether to use the website, but at the same time has less influence on how to use it, this can be interpreted as a form of libertarian paternalism. To elaborate, this is a weak form of paternalism that guides people (eg, a tunneled version of a website with less user control) without necessarily restricting their choices (eg, the decision to keep using a website) [31]. The crux of libertarian paternalism is that by allowing choice, but designing it in such a way that skews outcomes toward particular directions (eg, increased website use), the visitor experiences a degree of free will [32]. Hence, user control was hypothesized to increase efficiency [33,34] (Hypothesis 3), but to decrease actual website use, because it is the opposite of libertarian paternalism (Hypothesis 4).

Hypothesis 4 may seem in opposition to Hypothesis 3: Although user control was expected to decrease website use, it was expected to increase efficiency. To answer the question whether user control can help to increase website use, we explored whether the possible direct effect of user control surpasses the possible effect mediated by efficiency. Figure 1 provides an overview of the conceptual model used in this study.

The experimental condition for this study was a website about hepatitis A, B, and C virus (HAV, HBV, and HCV) infections. These infections all affect the liver, but each of the HAV, HBV, and HCV infections differs in terms of mode of transmission, consequences, and prevention. The website consisted of a home page plus four pages of information per hepatitis virus. The first page introduced the virus briefly and the other three pages outlined information about transmission, consequences, and prevention, respectively. This resulted in a total of twelve pages of website content (including all virus types) plus the home page. The content for these pages was based on information from the Dutch National Hepatitis Centre (eg, information leaflets) and was limited to mode of transmission, consequences, and prevention of HAV, HBV, and HCV infections. The content was text-based, purely informative, non-tailored, and very brief (ie, 5–10 lines of text per page).

Perceived user control is higher (5.2 vs 3.9 on a 7-point Likert scale) if participants had freedom of choice (eg, could skip pages) instead of a tunneled version of the website (F _1,452 = 134.32, P < .001), indicating that the manipulation of user control was successful. Table 2 shows hepatitis knowledge per group, both at the pre-test and follow-up. There are group differences regarding hepatitis knowledge at follow-up, after controlling for hepatitis knowledge at the pre-test (F _2,567 = 47.24, P < .001). All pairwise comparisons are significant (P < .001) indicating that participants in the tunneled group score higher on hepatitis knowledge compared with the freedom of choice group. Both experimental groups score higher on hepatitis knowledge in comparison with the control group, indicating that participants do not only click through the website, but actually process and learned the content.

Table 3 shows that having a lesser degree of user control has a negative effect on efficiency, but a positive effect on number of pages visited (confirming Hypotheses 3 and 4). Participants do not simply click from one page to the next, since the time on the website is also longer in the tunneled group (3:50 min) compared with the freedom of choice group (2:38 min) (F _1,452 = 6.32, P = .01).

Figure 3 illustrates the final structural equation model. User control has a positive effect on efficiency, but a negative effect on website use. The direct effect appears to surpass the effect mediated by efficiency because website use is higher in the tunneled group compared to the freedom of choice group. These findings also support Hypotheses 3 and 4. Efficiency has a positive effect on website use, but effectiveness and enjoyment do not have a direct effect on website use (only partly confirming Hypothesis 1). Therefore, these paths were removed from the final model. Active trust, however, mediates the relationship between efficiency, effectiveness, enjoyment, and active trust (confirming Hypothesis 2). Based on modification indices, no paths were added to the conceptual model, implying that user control is only related to efficiency and website use. The CFI and TLI are .97 and .96, respectively; RMSEA and SRMR are .08 and .03, respectively. All of these fit indices indicate a close fit for the final model.

A key finding of this study is that user control does not help to increase website use, which leads to a smaller effect on knowledge gain in comparison with the tunneled version of the website. Although visitors thought that having control made it easier to search and access information, this was negated by the direct negative effect that user control had on website use. In short, user control decreases website use. The increase in perception of efficiency, however, is in-line with the idea of freedom of choice as more important than its actual existence [46]. These findings indicate that, for future intervention websites, visitors should be carefully guided through the intervention (ie, less user control).

As an element of interaction design (ie, options involved in performing and completing tasks), it is proposed user control should fit the task of the user and the objectives of website accessed [18]. For example, if a goal for a website is to encourage visitors to review all pages about self-management, then it would be useful for visitors to be guided through these pages. However, if the website is to serve as a data bank or encyclopedia that visitors may consult, then it would be more useful to add a search function and an effective menu structure [47].

Most of our hypotheses regarding user perceptions are confirmed and are in-line with the previous study [13]. Active trust mediates the positive effects of efficiency, effectiveness, and enjoyment on website use. The lack of a direct effect of enjoyment on website use was unanticipated. Based on previous research, an explanation could be that enjoyment is related to cognitive perceptions [48], and website use is fully mediated by active trust. Thus, cognitive perceptions might be a catalyst for the positive effect of enjoyment on website use. The lack of a direct effect of efficiency on website use is puzzling. Since previous research demonstrated that active trust is a mediating variable associated with intention to use a website [49,50], it might be that active trust reduces the possible direct impact of efficiency since the explained variance of active trust overlaps with the explained variance of efficiency. Future research is needed to investigate the plausibility of this explanation.

Another avenue for future research is to examine factors related to the visitor given that the direct effect resulting from the manipulation of a website characteristic (ie, user control) surpassed the effect mediated by user perceptions (ie, efficiency). Hence, it is worthwhile to investigate whether the impact of website characteristics is greater for certain visitors than for others. Knowing something about the personalities of those who favor certain website characteristics will provide better insight into factors behind website use [51]. Ross and colleagues [52] took a first step in this direction by linking personality factors to the use of Facebook features.

Contrary to previous research [13,19,33], a positive characteristic of this study is that actual website use was tracked and measured instead of using self-reported data only. More specifically, website use was tracked by means of server registrations, which in contrast to self-reports, is independent of visitors’ memory, interpretation, or social desirability [53,54]. Moreover, within the setting of a research panel in which participants could complete the study at their own convenience (eg, at their own home), a real-life situation has been mimicked. Participants could freely explore the website, without the limitations of a laboratory setting (eg, standardized environment, forced exposure), which enhanced the validity of the study (ie, in vivo versus in vitro testing). Finally, there was a relatively small dropout rate between pre-test and follow-up in this study, which was neither selective dropout nor differed between groups. Hence, there is still good variation in gender, age, and level of education of the participants to warrant generalizability of the findings to the Dutch Internet population above 18 years.

Finally, there are two additional points that evolved from this investigation. First, knowledge regarding hepatitis increased in both experimental conditions compared to the control group. This implies the hepatitis knowledge questionnaire is appropriate to assess whether participants processed and learned content of the website. This is the case even though participants were not necessarily looking for information on hepatitis, which is essential in primary prevention websites aimed at the general public. Nevertheless, it could be that participants were highly interested in hepatitis, since they agreed to participate in a study about this topic. The low scores regarding hepatitis knowledge at pre-test, however, do not support this possible explanation. Second, knowledge increased more in the tunneled group in which website use was higher. Since user control is not directly related to retention [55], this suggests website use increases the likelihood of changes in determinants of health risk behaviors—which is important from a public health point of view. Thus, as an element of interaction design, user control should be carefully considered during the development of Internet-delivered interventions. Our findings indicate future interventions should carefully guide visitors through the website (ie, less user control) to increase website use and subsequently elevate the public health impact of these interventions.