Abstract
We propose the concept of directive driving automation that positively influences drivers’ intentions to achieve shared goals. As a step toward directive driving automation, this survey study explored how social norms can persuade drivers to continue using driving automation. We tested social norm messages using a 2x2x2x2 factorial within-subject design: norm type (descriptive vs. injunctive), explanation (absent vs. present), spatial scale (local vs. global), and outcome criticality (safety-critical vs. non-safety-critical). The results suggest that framing messages as descriptive norms and providing explanations relevant to the driver can encourage them to continue using automation, especially with safety-critical outcomes. This study highlights the importance of considering message characteristics in persuasive interventions to promote the safe use of driving automation.
Introduction
Driving automation has changed substantially over the past decade. Some production-level vehicles with Advanced Driver Assistance Systems can not only drive by themselves under the driver’s supervision but can also monitor the driver’s readiness to perform the driving task (e.g., eyes on the road, hands on the steering wheel, etc.). With further improvements in these advanced technologies, interacting with driving automation will soon become a more collaborative task, like driving with an intelligent co-driver (Lee et al., 2022). To support collaboration, automation observability and directablity—the ability to see automation’s behavior and intention and the ability to redirect automation behavior and intention—have been conceptualized to create an interdependent relationship (Christoffersen & Woods, 2002).
However, observable and directable automation may not fully represent a fundamental aspect of collaboration: a horizontal relationship, which refers to a relationship between individuals that are on the same level. The concept of observable and directable automation focuses solely on the human operator’s perspective and may consider automation as a subordinate system. We suggest the concepts of directive and observant automation, which are the complements of observable and directable automation. While directable automation allows a person to adjust the intentions of the automation, directive automation changes the intentions of the person (Chiou & Lee, 2021). In the context of driving, we can envision directive driving automation that alters the driver’s intention to achieve a shared goal. This could involve encouraging the use of driving automation for safer driving when automation is more capable of controlling the vehicle or for efficient travel when using driving automation can enhance traffic flow. Although the concept holds promise, implementing directive driving automation can pose challenges; how can driving automation persuade the human driver?
Persuasive Technology
Persuasion involves influencing someone’s beliefs, attitudes, behaviors, or decisions through communication (Gass & Seiter, 2011). Persuasion is a non-coercive method similar to nudging, which is a behavioral intervention strategy to guide individuals toward desired behaviors while preserving their freedom of choice (Thaler & Sunstein, 2009). However, persuasion emphasizes interaction and communication between a persuader and the persuadee (i.e., a person receiving the persuasive technique).
Persuasive technology is defined as a “computing system, device, or application intentionally designed to change a person’s attitudes or behavior in a predetermined way” (Fogg, 1999, p. 27). Persuasion techniques have been categorized in various ways. For instance, the persuasive systems design model includes four categories: (a) primary task support, (b) dialogue, (c) system credibility, or (d) social support. Primary task support strategies assist a user in carrying out a primary task. For example, reduction techniques reduce the effort needed to reach a target behavior. Dialogue strategies involve human-system communication. For example, reward techniques communicate incentive structures to the user for performing a target behavior. System credibility refers to strategies that prompt persuasion by increasing a system’s credibility. Examples include authority, which refers to leveraging authoritative figures to add credibility to a target behavior. Finally, social support strategies use social influence principles to prompt a target behavior. This paper explores social support strategies to promote a target behavior. Particularly, we focus on the two types of social norms: descriptive norms and injunctive norms.
Social Norms
In the focus theory of normative conduct, Cialdini et al. (1991) distinguished descriptive norms and injunctive norms based on their relevance to fundamental goals that people pursue. Descriptive norms refer to behaviors that are perceived to be prevalent or typical. These norms appeal to people’s desire to behave efficiently or accurately. By motivating people to follow what most people typically do, descriptive norms function as heuristic rules (Cialdini & Goldstein, 2004). In contrast, injunctive norms refer to behaviors perceived to be approved by others. Injunctive norms appeal to people’s desire to build and maintain social relationships. By motivating people to follow what most people approve of doing or expect them to do, injunctive norms provide people with information about how to behave appropriately in a certain social context (Cialdini & Goldstein, 2004). There has been substantial evidence demonstrating the effects of these two types of social norms on changing people’s behavior in various contexts, such as littering (Cialdini et al., 1990), energy conservation (Goldstein et al., 2008), and mobile phone use while driving (Nicolls et al., 2022).
However, the potential persuasive power of descriptive and injunctive norms in the context of driving automation has been underexplored. Previous research showed that the relative influences of descriptive norms and injunctive norms may depend on which norm is salient in each context (Kallgren et al., 2000), and the persuasive power can be the strongest when those two norms are consistent with each other (Cialdini, 2003). In the current research, we examined the effects of making descriptive or injunctive norms salient on the use of driving automation, as little work has investigated the effects of these norms in automated driving contexts. Thus, in the directive messages presented to participants, we described using driving automation as either what most people choose to do (descriptive norms) or what most people think people should choose to do (injunctive norms). We targeted the use of the automated driving mode because of its benefits, as suggested in the existing literature on driving automation, such as reducing traffic congestion (e.g., Chen et al., 2012; Fagnant & Kockelman, 2015; Zheng et al., 2011).
When framing directive messages to promote the use of driving automation, it is essential to consider additional factors unique to this context. While there are benefits that justify the motivation for promoting the use of driving automation, it is unlikely that such benefits are shared among members of the current society. This distinguishes promoting the use of driving automation from other choices like avoiding littering or not texting while driving. Therefore, promoting the use of automation may require additional explanation to encourage its adoption.
The orientation of the message is another factor to consider when framing the directive messages. Prior research suggests that a message can be more persuasive when framed to be self-oriented, highlighting its relevance to the recipient rather than other-oriented, highlighting its relevance to others (Burnkrant & Unnava, 1995; Scannell & Gifford, 2013). To test this, we varied the message orientation to explore whether emphasizing relevance to the self could increase the persuasive impact of the directive messages.
Although driving is strongly associated with safety, not all everyday driving situations involve serious safety concerns like car crashes. While highlighting a message’s relevance to safety may increase its persuasive power, it is important to compare it with messages that are not associated with safety-critical outcomes. To do this, we created and compared two sets of messages portraying driving situations: one set involving car crashes and another set involving traffic flow.
Research Questions
This study addresses four research questions with a survey study of social norm messages. We used four measures— preference, effectiveness, trustworthiness, and feeling of control—but research questions refer to one measure for simplicity. The independent variables that we manipulated in the survey are boldfaced:
RQ1. Which
RQ2. Does the
RQ3. When the explanation is present, which
RQ4. Do people’s preferences for the messages vary between
To answer the questions, first, we examined whether directive messages framed as either descriptive or injunctive norms could persuade people to use the automated driving mode (RQ 1: effect of norm type). Second, we explored whether presenting explanations for why using the automated driving mode is beneficial would enhance the persuasiveness of the messages (RQ 2: effect of explanation). Third, we investigated whether describing the impact of the participants’ decisions on themselves (local impact) or other drivers (global impact) would influence the persuasiveness of the messages (RQ 3: effect of spatial scale). Lastly, we tested whether safety-related driving situations affected the persuasiveness of the messages (RQ 4: effect of outcome criticality).
Method
Participants
Twenty-four people participated in this study (Male = 9, female = 15, mean age = 32.7, SD of age = 2.9). Participants were recruited from the U.S. and compensated for their time ($10). Eligibility criteria for participants included: (a) between the ages of 25–55, (b) holding a valid U.S. driver’s license for at least two years, and (c) driving a minimum of 2,000 miles per year. The study lasted approximately 30 minutes for each participant.
Study Design
We used a within-subject design with a 2x2x2x2 factorial structure: norm type (descriptive vs. injunctive), explanation (absent vs. present), spatial scale (local vs. global), and outcome criticality (safety-critical vs. non-safety-critical). See Figure 1 for message samples and how each factor was manipulated.

Illustration of the survey design with example messages.
Survey Implementation and Procedure
The survey was created using the Qualtrics platform. The survey included a consent form, demographic questions, driving history, use of driving automation, propensity to trust automation (Merritt, 2011), the main survey (26 items per set), and follow-up questions. To assess participant engagement, two attention check items were added, and only the surveys that passed both items were used in the analysis.
The survey presented a scenario of drivers using partially automated driving who intend to disengage from driving automation, and the vehicle displays a message to encourage the driver to continue using driving automation. Participants were asked to compare a pair of messages and choose the message that (a) they preferred, (b) was effective in changing their minds about overriding vehicle control, (c) was more trustworthy, and (d) gave them the feeling that they had more control over the vehicle (see Figure 2). The follow-up questions explored participants’ selections by requesting them to rank the importance of each reason for choosing a message. Reasons for choosing a message for each measure were extracted from previous literature: preference (Kaley, 2022; Munro, 2020; Nielsen, 1994, 2001), effectiveness (Meske & Potthof, 2017; Torning & Oinas-Kukkonen, 2009), trustworthiness (Bews & Martins, 2002; Roy & Shekhar, 2010), and feeling of control (Thompson et al., 1998).

Survey item example. Participants were instructed to compare a pair of messages and select one for each measure.
Analytic Method
We used the choice-based conjoint (CBC) analysis to assess how the independent variables (IV) affect participants’ choice of one of the two messages. CBC is typically used in marketing research to assess how various attributes of products (e.g., price, design, etc.) affect consumers’ choices to purchase one product from several alternatives (Desarbo et al., 1995). For the survey analysis, we fit a multinomial logistic regression model using R (R Core Team, 2022).
Equation (1) shows the assumption of the multinomial logistic model from Kuhfeld (2010). Specifically, it assumes that the probability that a participant will choose one message (c!) between the two messages (m) in a survey item (C) is:
x! is a vector of dummy-coded independent variables. x! is a linear part-worth utility function determining the utility for the message c! (U(c!) ). From (1), we estimated each message’s choice probability ((c!|C)). We also used the partworth utility function (x!β) to calculate part-worth utilities for the IVs. In turn, the part-worth utilities provided a measure of the IVs’ impact on participants’ choices between the two messages in each survey item.
Results
The analysis was performed separately for the non-safetycritical set and the safety-critical set for simplicity, and the results were compared between the two sets.
Non-Safety-Critical Outcome
For messages related to non-safety-critical outcomes, participants selected descriptive norms over injunctive norms for all four measures, as shown in Figure 3 (left). The effect of explanations and spatial scale was only significant for trustworthiness, indicating that participants deemed messages with an explanation and messages with an explanation about the estimated time of arrival (ETA) to be more trustworthy.

Beta weights and 95th percentile confidence intervals for factors in the non-safety-critical outcome set (left) and in the safety-critical outcome set (right).
Safety-Critical Outcome
Regarding messages related to safety-critical outcomes, norm type, explanation, and spatial scales were statistically significant for all measures, showing a consistent trend as depicted in Figure 3 (right). In general, messages using descriptive norms, providing an explanation, and using an explanation related to the crash probability of the participant’s vehicle (versus other drivers’ crash probability) had higher probabilities of being selected. The coefficients’ directions were consistent with the non-safety-critical messages, suggesting a general trend regardless of the outcome criticality. However, the impact of message characteristics was more pronounced when the outcome was safety-relevant.
Rank-Ordered Subdimensions of the Measures
Participants ranked the subdimensions (reasons) of each measure at the end of the survey. We analyzed the data to investigate the participants’ decision-making process. Only the top three reasons are listed in Table 1 for comparison, as the number of reasons varied across measures (6-11 reasons).
Top three reasons of each measure (the mean rank).
Participants preferred messages that were concise, useful, and simple. Message directness, format, content, and targeted behavioral change were associated with effectiveness. Trustworthiness was associated with the reasons such as competency, history of interactions, and openness. Regarding feeling of control, the most frequently selected reason was the responsibility to make the final decision, followed by the knowledge of driving behaviors and outcomes, and obsession with vehicle control.
Discussion and Conclusion
The current study explored how social norms might persuade drivers to continue using fully automated driving in situations where they intended to disengage the automation. To achieve this end, social norm messages were designed and tested. To understand how social norm messages influence the use of driving automation, four message characteristics (IVs) were manipulated: norm type, presence of explanation, spatial scale, and outcome criticality.
We found that messages framed as a descriptive norm were preferred and rated as more effective and trustworthy in both non-safety-critical and safety-critical situations (RQ1). The effect of explaining the benefits of using driving automation was significant for safety-critical outcomes (RQ2). Framing the explanation as relevant to the self was preferred (for both safety-critical and non-safety-critical situations), effective (for safety-critical situations), trustworthy (for safety-critical situations), and gave participants a greater sense of control (for safety-critical situations) (RQ3). Safety-critical situations made the effects of providing an explanation and framing the explanation relevant to the self significant for preference, effectiveness, and feeling of control, compared to non-safetycritical situations (RQ4). Follow-up questions showed that while participants preferred concise and simple messages, in safety-critical situations, they wanted explanations that increased message length. This presents a design challenge that might lead designers to consider alternative modalities beyond visual cues, such as auditory feedback, for both concise and informative messaging.
A previous meta-analysis of choice architecture interventions categorized choice architecture interventions into decision information, decision structure, and decision assistance (Mertens et al., 2022). They found decision structure interventions to be more effective in general, but decision information interventions (including social norms) should be used when persuasion depends on personal values and goals. This study found that social norms can enhance message effectiveness.
Future studies should validate the effect of social norms on driver behavior in a driving simulator. Additionally, combining social norms with other persuasion or nudging techniques (e.g., increase procedural friction to override) should be investigated to promote the appropriate use of driving automation. To enhance the effectiveness of persuasive messages, future research should consider personalizing messages based on user characteristics such as attitudes toward automation.
This study highlights the importance of message characteristics in persuasion interventions aimed at promoting the use of driving automation. The results suggest that framing messages as descriptive norms and providing explanations that are relevant to the driver might encourage drivers to continue using automation, especially in safety-critical situations.
