The effect of information on changing opinions toward autonomous vehicle adoption: An exploratory analysis

Abstract

There is extensive theoretical literature that looks at factors that make people more or less likely to change their opinions as additional information is gathered. People whose opinions are less likely to change in response to information may have strong anchoring effects (commitments to initial opinions) or may support their initial opinion by selectively processing information to confirm their initial opinion (confirmation bias). Selectively processing information can also result in opinion polarization where opinions become more extreme as additional information is provided. While theoretical literature has been relatively abundant on this topic, there has been limited empirical evidence with transportation-related opinions as to how anchoring effects and confirmation bias may affect changing opinions and possible opinion polarization. The intent of the current paper is to provide some initial evidence of changing opinions and possible polarization as it relates to the potential adoption of autonomous vehicles, which will likely be a key element in future sustainable transportation strategies. Specifically, the paper studies how people’s initial autonomous-vehicle adoption likelihoods change after being asked a common set of questions that leads them through an assessment of factors involved in adoption. A series of discrete outcome models were estimated to determine the factors that influence the likelihood of people changing their initial opinions. Although the empirical models identified many variables associated with opinion change, it is argued that traditional transportation surveys may not be gathering the type of data needed to truly understand how people’s transportation-related decisions evolve in response to new information.

Keywords:

• anchoring effect
• attitudes
• autonomous vehicle adoption
• confirmation bias
• opinion formation

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 As will be shown, the experimental design that is used in this paper cannot directly account for authoritative-source effects. However, it is important to keep in mind that one of the determinants of a strong anchoring effect is an authoritative-source effect, and that people with certain measurable characteristics may be more or less susceptible to such an effect.

2 An alternative to this experimental approach would be to gather longitudinal data and track changing opinions. However, in addition to the high cost associated with the acquisition of such longitudinal data, there would be significant challenges in survey design to capture the effects of people’s information gathering from media, social networks, and other sources. While the experimental approach adopted herein allows for a much tighter control on information availability, exploring this opinion-change issue with a detailed longitudinal survey is a promising direction for future research.

3 Consideration must also be given to the primacy/recency effect with regard to the questions in the appendix. That is, the fact that questions presented at the beginning (primacy) and the end (recency) are likely to be given more weight than questions in the middle. A potentially fruitful direction for future research would be to randomize these questions, or present respondents with a finite group of alternate orderings of these questions, to study the possible extent of the primacy/recency effect in this context.

4 It should be pointed out here that the approach of asking questions as a means of having people think about elements of autonomous vehicle adoption is fundamentally different than providing them with specific information, or have them gather information themselves through various media and social networks. Thus, some caution should be exercised in extending the findings to these other forms of information gathering.

5 Please note that the experimental approach used is potentially susceptible to hypothetical bias. That is, in unfamiliar contexts (such as autonomous vehicle adoption) individuals may not fully understand or perceive how the hypothetical decision they are making will differ from an actual decision (Rakotonarivo, Schaafsma, & Hockley, 2016). This point should be kept in mind in assessing the forthcoming empirical findings of this paper.

6 In addition to the normal distribution, models were estimated with a number of other distributions, but no other distribution produced estimation results that were significantly better than the normal distribution.

7 There is also the possibility that people may not remember their initial opinion selection and just select a new opinion by chance even though their core opinion has not changed. However, this possibility is believed to be highly unlikely since the survey’s focus was on autonomous vehicle adoption which implies this question would have been given careful thought before and after the informational questions.

8 Respondents without an initial opinion (those who are initially uncertain) are not considered because the study focuses on anchoring effects and polarization. Those without an initial opinion will not have an anchoring effect, will not engage in confirmatory hypothesis testing, and thus will not technically polarize.

9 Although the response data are still technically ordered (extremely unlikely, unlikely, etc.), conditioning on the initial adoption likelihood reduces the ranges of responses considerably. Given this, and the additional flexibility inherent in traditional non-ordered response models such as the multinomial logit, an unordered response modeling approach is chosen. Please see Mannering and Bhat (2014) for an extensive discussion of this point.

10 Because the focus of the paper was on anchoring effects and polarization, recall that the statistical analysis did not address the changing opinions of the 248 people who initially indicated that they were uncertain. After going through the informational questions, 55% of these respondents remained uncertain, 26% became likely, 17% became unlikely, 2% became extremely likely, and 0% became extremely unlikely. These rather substantial shifts suggest that additional information definitely affects the likelihood of remaining uncertain. A study focusing on the effects that information has on uncertainty in the context of autonomous vehicle adoption would be a fruitful area for future research.

11 The possibility of heterogeneity in means and variances was also considered (Behnood & Mannering, 2017a, 2017b; Seraneeprakarn et al., 2017). However, likelihood ratio tests showed that these formulations did not significantly improve the model estimation results.

12 Latent-class logit models were also estimated but these did not result in statistically different classes. This adds additional support indicating that unobserved heterogeneity was not playing a significant role in the model estimations (Mannering et al., 2016).

Additional information

Funding

The authors gratefully acknowledge support provided by the Center for Teaching Old Models New Tricks (TOMNET), a University Transportation Center sponsored by the US Department of Transportation through Grant No. 69A3551747116.