Perceptions of Autonomous Vehicles: A Case Study of Jordan

Highlights

• The ordinal logit model was deployed to determine the factors attributed to individual acceptance of AVs, such as the cost, security, privacy, along with the environmental impact, among others.
• The results indicated that the cost of AVs greatly influences purchasing decisions, though if the cost is affordable, respondents were more interested in using AVs.

This study helps to enhance the current understanding by highlighting road user perceptions, with practical implications for practitioners.

Abstract

Technologies for automated driving have advanced rapidly in recent years. Autonomous Vehicles (AVs) are one example of these recent technologies that deploy elements such as sensors or processing units to assist the driver. The effective integration of these vehicles into public roads depends on the drivers’ acceptance and how they adjust to this new generation of vehicles. This study investigated the acceptance and willingness of Jordanians to purchase AVs in Jordan. The ordinal logit model was deployed to determine the factors attributed to individual acceptance of AVs, such as the cost, security, privacy, along with the environmental impact, among others. The findings of a national survey conducted on 582 Jordanians to assess their perception about AVs revealed that Jordanians were generally interested in using AVs. However, their decisions about purchasing AVs are influenced by several factors. The results indicated that the cost of AVs greatly influences purchasing decisions, though if the cost is affordable, respondents were more interested in using AVs. The findings also revealed that there is a substantial relationship between the level of security and the likelihood of buying a self-driving car, as respondents are concerned about the level of security and privacy. Furthermore, the results revealed that environmentally friendly AVs are more likely to be owned compared to conventional vehicles. This study helps to enhance the current understanding by highlighting road user perceptions, with practical implications for practitioners.

1. Introduction

Technologies for automated driving have evolved rapidly in recent years [1,2]. Autonomous Vehicles (AVs) or self-driving cars are one example of these recent technologies that deploy elements such as sensors or processing units to assist the driver in; for example, steering, braking, lane changing, or parking [3].

The technology of automating driving has several levels, as described by the International Society of Automation Engineers [4]. These levels vary from no driving automation, where drivers complete all tasks (i.e., level zero), to full driving automation, where vehicles could take complete control under all conditions (i.e., level five). The robotisation of driving tasks has the potential to dramatically reduce driver errors resulting from fatigue, distraction, or other driver errors and improve traffic safety [5,6]. This is consistent with the findings of [7] who investigated the impact of connected and autonomous vehicles on traffic safety by simulating connected and conventional vehicles under various penetration rates. They concluded that the condition of traffic safety is greatly improved with the increase in the penetration rates of connected vehicles compared to conventional vehicles. Human errors are a major contributing factor in crashes that account for most of road crashes in Jordan according to the Jordan Traffic [8]. Eliminating these errors will significantly improve road safety.

While AVs could improve traffic safety by more than 90% by eliminating driver errors [9], other factors might increase the rate of collision (e.g., system malfunction) as the AVs’ safety is still in the early stages. One can refer to [2] for more discussion about the safety of automated driving.

Several studies have discussed the merits of AVs for various road users (i.e., drivers, pedestrians, or cyclists). For instance, in [10] the authors discussed the benefits of autonomous driving, including increased road safety and reduced traffic congestion. Other studies, see for example [11,12], investigated the impact of advanced driver-assistance system on pedestrians and cyclists safety. These studies highlighted how aiding driver systems support drivers as they overtake cyclists or pedestrians to avoid or mitigate crashes. In [13], the authors discussed the advancements in connected and autonomous vehicles (e.g., using 5G networks), highlighting the need for direct communication between vehicles to ensure safe and comfortable travel experiences. Refs. [14,15,16,17,18], among others, discussed how AVs are environmentally-friendly and can reduce fuel consumption, and mitigate air pollution.

In [19] the authors discussed how AVs are comfortable and improve the travelling experience. The authors of [9,20] investigated the influence AVs exert on people; for instance, safety concerns, as driverless systems malfunctions could compromise the safety of passengers, as well as the potential risks of privacy invasion during the process of information exchange. However, if properly regulated, AVs will likely improve the transport system by eliminating vehicle crashes.

The question then arises: are drivers willing to use AVs? Earlier studies discussed the willingness to use AVs in several countries: for example, the USA [21], China [22,23], Turkey [24], Japan, UK, and Germany [3], Poland [25], Belgium [26], Austria [27] and KSA [28], among others, as summarised in Section 2. The results of these studies revealed that attitudes, in general, are in favour of automated driving systems. The acceptance of AVs is still unknown in Jordan. The deployment of AVs will depend on road users’ acceptance. To this end, this study examines road users’ perceptions, interest, and concerns about using AVs in their daily commuting. This study has adopted the ordinal logit model (i.e., logistic regression) to investigate the acceptance of AVs, but will differ from the others, as it examines the factors in two groups: demographical factors (e.g., age, income, or gender) and AV characteristic factors (e.g., safety, security or cost). To this end, this paper investigates and discusses the factors that impact the acceptance of using AVs in Jordan.

The following sections provide a literature review, outline the questionnaire survey scenarios, detail the development of the proposed model, and discuss the main findings.

2. Literature Review

Effective implementation of AVs technology will require public acceptance, with respect to purchase and use [29]. In [30], the technology acceptability was defined as:

“the evaluation of that technology before having any interaction with it” (p. 253).

The literature on acceptability of AVs embraces a broad range of studies that implemented in several countries via surveys on automated driving systems. For instance, [30] investigated the acceptance of AVs in France. Their findings indicated that about 52% of the sample expressed interest in using AVs. However, gender differences were observed, with men being more inclined to purchase or use AVs compared to women. A later study [20] investigated road user acceptance and concerns to purchase an AV across 109 different countries. Their findings indicated that men are more willing to pay more for AVs than women. Their findings are consistent with another three studies [1,22,23] conducted in China to provide insights into the public’s willingness to adopt self-driving vehicles. These studies concluded that young participants with high levels of education and high income would pay more for AVs than other groups. In contrast, the study by [31] found that Indian women were willing to ride driverless vehicles. This is in line with [28] as they found women favoured AVs more than men in KSA. One can refer to [32] for a systematic review about AVs.

Based on the reviewed literature, the widespread acceptance of AVs is influenced by several factors. These could be grouped into:

• Demographics: age, gender, or income;
• AVs related characteristics: privacy and security, environmental impact, or cost.

This paper investigated the acceptance of AVs based on privacy, security, environmental impact, or cost. It is worth noting that this study focuses on level five automation systems (i.e., complete control under all conditions). This assumption was made for several reasons:

• Level five AVs have not been yet introduced in Jordan, unlike other automation levels. Therefore, investigating their potential impact on society can provide valuable insights into the acceptance and adoption of autonomous vehicle technology in the future.
• While level five AVs are not currently available in Jordan, this level of automation may become more prevalent in the future. Thus, there is a need to understand how potential road users perceive and accept this technology.
• Focusing on level five AVs allows the research objectives to be more narrowly defined, with a specific emphasis on the potential implications of fully autonomous driving.

Several statistical models were implemented to understand the willingness to own or use AVs and its determinants, such as partial least squares, a Tobit model, and ordinary least squares analysis [1], technology acceptance model [23], stated choice survey [28], the multinomial logit [24,33], Mann–Whitney U test [27], the ordered logit model [3], the binary logit model [34], and the Analysis of Variance (ANOVA) model [22]. However, the choice of statistical analysis may have its own assumptions and the outcomes may be affected by these assumptions. To this end, the ordered logit model was implemented in this study, as this model is used when the goal is to model the probability of falling into each category of the outcome variable, rather than falling into a specific category. This gives a better understanding of the correlation between each category in the predictor variable with the outcome variable. In this study, we hypothesis the willingness of buying a self-driving car as the dependent variable, with privacy and security, environmental impact, or cost as independent variables.

Overall, much work has been conducted to investigate the factors influencing the acceptability of AVs; however, there is still scope for more improvement. This study aims at investigating the acceptance of AVs in Jordan, along with understanding the factors with influence on the acceptance of AVs using the ordinal logistic model. Setting reasonable expectations regarding the acceptance of AVs before users interact with them may be critical before deploying AVs in Jordan. The study’s findings will help policymakers and planners to estimate short- and long-term adoption of AVs, and develop policies to achieve optimal adoption rates. This is in line with the 2030 vision of a new smart city in Jordan [35].

3. Ordinal Logistic Regression Model

The ordinal logistic regression model is used to model the correlation between an ordered dependent variable (i.e., such as the decision to purchase a self-driving car) and independent variables (i.e., privacy, security, environmental impact, or cost). This model was selected among others as the values of the outcome variable have an order; for example, the five-point Likert scale [36,37,38]. It is used when the goal is to model the likelihood of falling into each category of the outcome variable, rather than a specific category [38]. Therefore, the event being modelled not only has an outcome in a particular category, but also preserves information about response categories that are ordered [39].

A general description of the ordinal logistic regression model (OLM), based on [36,37,38,40], is given below. The OLM is similar to the equation for simple logistic regression, but includes the cumulative link function [40]. The general equation of the ordinal logistic regression is:

$$P(Z=k)=F(\alpha k+{\lambda }_1{X}_1+{\lambda }_2{X}_2+\dots +{\lambda }_p{X}_p)$$

(1)

where $P(Z=k)$ is the probability of the observation Z falling into k category of the dependent variable; F is the cumulative link function, which is typically the logistic Cumulative Distribution Function (CDF) or the probit CDF; $\alpha k$ is the intercept for category k; ${\lambda }_1,{\lambda }_2,\dots ,{\lambda }_p$ are the coefficients for the independent variables $X_1,X_2,\dots ,X_p$. However, the specific form of the equation will depend on the cumulative link function used. For example, if the logistic CDF is used, the equation will be as follows:

$$P(Z=k)=\frac{exp(\alpha k+{\lambda }_1{X}_1+{\lambda }_2{X}_2+\dots +{\lambda }_p{X}_p)}{(1+exp(\alpha k+{\lambda }_1{X}_1+{\lambda }_2{X}_2+\dots +{\lambda }_p{X}_p))}$$

(2)

and if the probit CDF is used, the equation will be:

$$P(Z=k)=\mathsf{\Phi }(\alpha k+{\lambda }_1{X}_1+{\lambda }_2{X}_2+\dots +{\lambda }_p{X}_p)$$

(3)

where $\mathsf{\Phi }$ is the standard normal CDF.

For this reason, several models were proposed, such as the Partial Proportional Odds Model (PPOM), the Proportional Odds Model (POM), the Cumulative Logit Model (CLM), and the Cumulative Probit Model (CPM) [1,41,42,43]. Each of these models is a variation of ordinal logistic regression and differs in the assumptions made about the correlation between the predictor variables and the ordinal outcome variable. Some models, e.g., the POM, make the assumption that the effect of the independent variables on the outcome variable is the same for all categories of the outcome variable, while others, e.g., the PPOM, disregard this assumption and allow for different effects of the predictor variables at different categories (i.e., levels) of the outcome variable. The constrained cumulative logit model is the most widely used model, also is known as POM [39]. The POM could be implemented to evaluate the correlation between the dependent and independent variables [38,39] with no need to fit separate models for each dependent variable category. The POM can be formulated as below:

$$ln\left(\frac{p(Z=k\mid X)}{p(Z=k-1\mid X)}\right)=X^{\prime }{\lambda }_k$$

(4)

where $p(Z=k\mid X)$ is the probability that the response variable takes on the level (of k levels), given the values of the predictor variables X; X is the matrix of predictor variables; Z is the ordinal categorical outcome variable; and ${\lambda }_k$ is the vector of coefficients for the kth level of the outcome variable. The ordinal logit regression model based on the POM is adopted in this study to evaluate the relationship between independents and dependent variables.

4. Method and Implementation

To investigate the factors that affect the acceptance of self-driving cars in Jordan, a survey was designed and analysed using the ordinal logit model as discussed below.

4.1. Design Overview

A questionnaire- based survey approach was adopted to obtain data from a diverse range of respondents in Jordan. The questionnaire survey was based on design principles described by [30,44]. The first section of the questionnaire compromises of demographical questions, which are mainly utilised to investigate age, gender, education level, employment, income, mobility disability, location and other demographic information of the participants. The second section includes factors that measure the acceptance of using AVs, such as privacy and security, environmental impact, the cost of AVs, and the public’s decisions about purchasing an AV.

The five-point Likert scale is implemented by previous research studies [45,46], and it was used in this study to measure each item in the second part of the questionnaire. In this part of the questionnaire, questions were designed based on previous studies, along with modifications of experts’ comments to suit the research context. In addition, since AVs have not appeared in Jordan yet, to avoid difficulties in understanding the the study, the first page of the questionnaire includes a clear description of the context to guide the respondents for a better understanding of the research context.

The statistical assessment of the model was conducted using IBM SPSS Statistics 22 Software (SPSS) package.

4.2. Population and Sample Profile

According to the Jordanian Department of Statistics (DoS), the country had a population of about 11,057,000 in 2021 [8]. The targeted sample for this study is about 6,159,480, which covers all Jordanians over the age group of 18 years (i.e., those who can legally drive a car). According to [47], the sample size can be calculated by using Equation (5) as given by:

$$\mathrm{Sample}\mathrm{size}=\left(\frac{z^2\ast std(1-std)}{e^2}\right)/\left(1+\frac{z^2\ast std(1-std)}{e^2\ast n}\right)$$

(5)

where n is the population size (i.e., 6,159,480), e is the margin of error (i.e., 0.05), std is the standard deviation (i.e., 0.5), z is the z-score (i.e., 1.96). Based on Equation (5), the acceptable sample size must be above 384.13. A total sample size of 582 was collected to ensure that it would be representative of the population of Jordan.

4.3. Data Collection Instrument

An online survey was used to collect the data from participants. This research made use of social media since it can offer access to a broad population. The participants were contacted to participate from June to August 2022. The survey, which was distributed through social media and websites that cater for many segments of society and different ages, invited the public to participate in an online survey about their attitudes toward AVs. The time required to complete the questionnaire was estimated to be less than 10 min. A total of 631 of the forms were returned, and 49 were discarded due to the participants responses with patterns. Final data collected came from 582 forms. This gives a response rate of about 92%.

4.4. Descriptive Statistics

The questionnaire was filled out by 582 respondents, all over the age of 18 and residing in Jordan.

Table 1. Respondents’ demographic information (N = 582).

Variable		Frequency	Percentage
Driving License	Yes	324	55.7
	No	258	44.3
Car/Household	0	73	12.5
	1	269	46.2
	2	165	28.4
	≤3	75	12.9
Age (years)	18–28	443	76.1
	29–39	65	11.2
	40–50	63	10.8
	50+	11	1.9
Gender	Male	187	32.1
	Female	395	67.9
Education	Undergraduate	58	10
	Bachelor/diploma	498	85.6
	Postgraduate	26	4.5
Employment	University student	312	53.6
	Employee	169	29
	Unemployed	78	13.4
	Other	23	4
Family income (JOD) 1	<1043	305	52.4
	1043–1460	67	11.5
	1460–1670	18	3.1
	1670–1880	10	1.7
	1880–2090	17	2.9
	>2090	20	3.4
	Prefer not to answer	145	24.9
Mobility disability	Yes	60	10.3
	No	522	89.7
Governorate	North (Irbid, Ajloun, & Jerash)	41	7
	Mafraq	7	1.2
	Salt	17	2.9
	Zarqa	86	14.8
	Amman	414	71.1
	Madaba	8	1.4
	South (Ma’an, Tafilah, Karak, & Aqaba)	9	1.5

¹ JOD 1 = USD 0.7.

summarises the social demographic characteristics of the study sample. Among the 582 participants, about 32% are male, and 68% are female. About 76% of the respondents are 18–28 years old, 11% are 29–39, 11% are 40–50, and 2% are older than 50. Approximately 56% of the respondents have a driver’s license. Almost 13% of the participants’ families do not have a car, 46% have one car, 28% have two cars, and 13% have more than two cars. Monthly income of most of the respondents is less than JOD 1043 (about USD 1470) (53%), followed by JOD 1043 to JOD 1460 (12%), JOD 1670 to JOD 1880 (3%), JOD 1880 to JOD 2090 (3%), more than JOD 2090 (4%), while 25% of the participants preferred not to answer. About 86% of the participants hold a bachelor’s degree. About 71% of the respondents are based in the capital city of Jordan (Amman).

5. Results and Discussion

The below sections summarise the results, along with a discussion of the major findings.

5.1. Ordinal Logit Model and Hypothesis Development

The below assumptions must be checked before implementing the ordinal logit model:

• The dependent variable and at least one of the independent variables are measured on an ordinal scale;
• There must be no multicollinearity among the independent variables as the presence of multicollinearity results in difficulty to determine the relationship between the dependent and independent variables;
• The dependent variable categories are affected equally by each independent variable.

Therefore, the following tests were carried out:

• The Variance Inflation Factor (VIF) was used to check multicollinearity.
• The goodness of fit was implemented to compare the observed data to the expected data generated by the model.
• The proportional odds assumption was applied to model ordinal categorical dependent variables.
• The parallel lines test was utilized to check whether the odds of belonging to one category compared to another category are the same for all levels of the independent variables.

5.2. Parallel Lines Test

An ordinal regression model’s validity relies on the proportionate odds assumption, as stated in the method section. In the event of a failure to satisfy this assumption, one cannot deploy an ordered logit model. Therefore, a test of parallel lines is implemented to assess the proportional odds assumption which states that the odds of a unit change in the independent variable having an effect on the outcome variable are constant across all levels of the dependent variable. This test involves plotting the predicted probabilities of each level of the dependent variable against the independent variable and examining the parallelism of the lines. If the lines are parallel, it indicates that the proportional odds assumption is met, and ordinal logistic regression can be used. If the lines are not parallel, it suggests that the proportional odds assumption is invalid, and an alternative statistical model may need to be used.

As can be seen in

Table 2. Parallel lines test.

Model	−2 Log-Likelihood	Chi-Square	$\mathrm{df}$	p
Null Hypothesis 1	995.568
General	937.191	58.376	48	0.145 1

¹ The null hypothesis states that the location parameters (slope coefficients) are the same across response categories.

, the Chi-square was 58.376 and the p-value was 0.145, indicating that there is no significant variation in the slopes of the regression lines among dependent variable categories. This shows that the correlation between the predictor and outcome variables is constant across categories. The results of the test of parallel lines validates the OLM. As a result, the assumption of proportional odds is confirmed.

It should be noted that a significant result from the parallel lines test does not guarantee that the ordinal logistic regression model is appropriate for this data. It is one step in the process of verifying the model’s assumptions and ensuring that the model is an appropriate fit for the data. Therefore, the residual approach is used to evaluate the goodness of fit of the assumptions model. The residual plot, as depicted in Figure 1, is almost horizontal between −2 and 2. This indicates that the model fits the data well and the residuals are randomly distributed. A horizontal residual plot also indicates that no pattern exists in the residuals, which implies that the model represents the underlying correlation between the predictors and the outcome variable.

5.3. Multicollinearity Check

The absence of high multicollinearity is another assumption for the ordinal regression analysis validity. To check for multicollinearity in an ordinal logistic regression model, various statistics can be used, including tolerance and the VIF. Tolerance measures the proportion of the variance in a predictor that is not explained by other predictors, with a lower tolerance indicating higher multicollinearity. On the other hand, the VIF measures the inflation of the standard error of the regression coefficients due to multicollinearity, with a higher VIF indicating higher multicollinearity. Typically, a VIF greater than ten is considered a sign of severe multicollinearity, whereas a VIF between five and ten indicates moderate multicollinearity. Based on the values shown in

Table 3. Multicollinearity statistics.

Model	Unstandardised		Standardised		95%CI			Collinearity
	B	Std.Error		t	Sig.	LB	UB	Tolerance	VIF
Predictors 1	2.129	0.186		11.433	0	1.763	2.495
Cost 2	0.013	0.046	0.012	0.281	0.779	−0.078	0.103	0.97	1.03
Privacy 3	0.032	0.04	0.036	0.801	0.423	−0.047	0.111	0.873	1.145
Enviro. 4	−0.039	0.041	−0.042	−0.95	0.342	−0.119	0.042	0.888	1.126
Security 5	−0.007	0.048	−0.007	−0.156	0.876	−0.102	0.087	0.993	1.007

¹ Dependent variable: are you thinking of buying a self-driving car? Independent variable: Cost, Privacy, Environment, and Security; ² Cost: what do you think about the cost of this type of vehicle?; ³ Privacy: these vehicles rely on large-scale data collection (user travel behaviours, travel time, work and home location, mobile phone number, …) to ensure a high level of security and improve traffic flow. Is this data collection a concern?; ⁴ Environment impact: how do you classify these vehicles from an environmental point of view?; ⁵ Security: these vehicles contain a system that enables the vehicle to identify the owners of the vehicle with voice, fingerprints and network detection, how should the level of safety and protection of the vehicle from thefts and harm to passengers be evaluated?

, the tolerance ranges from 0.970 to 0.993 and VIF values range from 1.030 to 1.145 are below ten which implies that that the multicollinearity in the model is low to moderate [48].

5.4. Goodness-of-Fit Check

The goodness-of-fit of a model is carried out to compare the observed data to the expected data in the model using some fit statistics [49]. This demonstrates the disparity between the values expected in the estimated model and the values actually observed. For ordinal logistic regression models, common goodness-of-fit tests include the Pearson and the deviance tests. These tests compare the observed and predicted frequencies of the outcome variable to determine the overall fit of the model.

As presented in

Table 4. Goodness-of-fit tests.

Model	Chi-Square	df	Sig.
Pearson	816.538	776	0.152
Deviance	684.650	776	0.992

, the Pearson and deviance tests are employed. The results of the Pearson test revealed a Chi-square value of 816.538 with 776 degrees of freedom and a significance value of 0.152. The results of the deviance test revealed a Chi-square value of 684.650, with 776 degrees of freedom and a significance value of 0.992. These results indicate that the OLM provides an adequate fit to the data, although the Pearson test suggests some discrepancy between the observed and expected values.

To evaluate the goodness of fit of a statistical model, model fitting information is used. The goodness-of-fit of the OLM is evaluated applying two measures; namely: the Pearson Chi-square and deviance Chi-square.

Table 5. The fitting information.

Model 1	−2 Log-Likelihood	Chi-Square	df	p
Intercept Only	1040.564
Final	995.568	44.996	16	0.000

¹ Link function: Logit.

shows that the final model has a lower −2 Log-Likelihood value (995.568) compared to the intercept-only model (−2 Log-Likelihood = 1040.564), indicating an improvement in model fit. The final model also has a Chi-square statistic of 44.996 with 16 degrees of freedom and a significance level of 0.000, indicating that the model provides a good fit to the data.

5.5. Ordinal Logistic Model (OLM) Testing

The OLM is suitable for hypothesis testing based on the measurements of assumption testing. In this model, thinking of buying a self-driving car is the dependent variable with five ordered categories. The independent variables (i.e., predictors) are individuals’ perceptions on the relationship between AVs and privacy, security, environmental impact, or cost. The findings of the ordinal regression analysis are summarised in

Table 6. Parameter estimates of ordinal logistic regression.

	Category	Estimate	Std.Error	Wald	df	p	Odds Ratio	95% C.I.
								LB	UB
Threshold	[BuyAV=0]	−1.879	0.758	6.151	1	0.013		−3.363	−0.394
	[BuyAV=1]	−0.187	0.747	0.062	1	0.803		−1.651	1.277
	[BuyAV=2]	1.637	0.75	4.762	1	0.029		0.167	3.108
	[BuyAV=3]	2.89	0.757	14.579	1	0		1.407	4.374
Location	[Privacy=0]	1.552	0.464	11.171	1	0.001	4.721	0.642	2.462
	[Privacy=1]	−0.437	0.322	1.836	1	0.175	0.646	−1.068	0.195
	[Privacy=2]	−0.439	0.274	2.564	1	0.109	0.644	−0.977	0.098
	[Privacy=3]	0.074	0.275	0.072	1	1.077	0.788	−0.465	0.613
	[Privacy=4]	0	.	.	0	.	.	.
	[Security=0]	0.319	0.368	0.751	1	0.086	1.376	−0.402	1.039
	[Security=1]	−0.084	0.272	0.094	1	0.059	0.920	−0.617	0.45
	[Security=2]	−0.274	0.214	1.641	1	0	0.759	−0.692	0.145
	[Security=3]	−0.082	0.216	0.145	1	0.004	0.921	−0.505	0.341
	[Security=4]	0	.	.	0	.	.	.
	[Environ.=0]	0.517	0.386	1.795	1	0.18	1.676	−0.24	1.274
	[Environ.=1]	0.059	0.278	0.044	1	0.833	1.061	−0.486	0.604
	[Environ.=2]	0.259	0.203	1.634	1	0.201	1.296	−0.138	0.656
	[Environ.=3]	0.327	0.223	2.14	1	0.143	1.386	−0.111	0.765
	[Environ.=4]	0a	.	.	0	.	.	.
	[Cost= 0]	0.816	0.707	1.33	1	0.249	2.261	−0.571	2.202
	[Cost=1]	1.329	0.717	3.432	1	0.034	3.778	−0.077	2.735
	[Cost=2]	0.854	0.73	1.369	1	0.242	2.350	−0.576	2.284
	[Cost=3]	0.931	0.79	1.388	1	0.039	2.539	−0.618	2.481
	[Cost=4]	0	.	.	0	.	.	.

.

Threshold values refer to the values that determine the cut-off points between categories in the dependent variable [50]. As presented in Table 6, five categories are ordered as follows: 0—very uninterested, 1—uninterested, 2—neutral, 3—interested, and 4—very interested. The thresholds determine the points at which a unit change in the independent variable is enough to change the predicted category of the dependent variable.

The predictor variable privacy measures the level of concern regarding large-scale data collection: 0 = not at all concerned; 1 = slightly concerned; 2 = somewhat concerned; 3 = moderately concerned; 4 = extremely concerned. The results showed the relationship between the two variables, the estimated effect of privacy on buying a self-driving car is statistically significant. The coefficients and their standard errors, as well as the Wald statistic, degrees of freedom, p-values, and 95% confidence intervals are presented in Table 6. When the privacy equals ’not at all concerned’, the estimated effect on buying a self-driving car is 1.552, with a standard error of 0.464 and a p-value of 0.001, indicating that the relationship is statistically significant. The 95% confidence interval ranges from 0.642 to 2.462, meaning there is 95% certainty that the true effect of privacy on buying a self-driving car lies between 0.642 and 2.462 for the privacy category ’not at all concerned’ equal to 0.

The independent variable level of security measures the security of AVs in a five-point scale ranges from high insecure to high secure. The results revealed that there is a significant relationship between the level of security and the possibility that someone may consider purchasing an AV. The estimate for buying an AV being very desirable (3) is 2.890, and the p-value of this estimate is less than 0.05 (0.000), indicating that this relationship is statistically significant. The estimate for buying an AV being very undesirable (0) is −1.879, with a p-value of 0.013, indicating that this relationship is statistically significant. The 95% confidence interval for this estimate ranges from −3.363 to −0.394, which further supports the correlation between the level of security and the likelihood of someone thinking of buying a self-driving car.

The independent variable environmental impact is a categorical variable with five possible outcomes; namely: strongly not friendly = 0; not friendly = 1; neutral = 2; friendly = 3; strongly friendly = 4. The results revealed that the categories of the “environmental impact” variable have varying effects on purchasing an AV. For instance, a change from ‘strongly not friendly’, to ‘not friendly’, increases the log odds of the dependent variable by almost 0.059, which is not statistically significant (i.e., p = 0.833). A shift from ‘neutral’ to ‘friendly’, increases the log odds of the dependent variable to 0.327 with a value of p = 0.143, which is statistically not significant. This indicates that self-driving cars that are classified as environmentally friendly are more likely to be bought.

Finally, the outcome variable cost measures the willingness to purchase an AV based on their cost. The cost categorical variable ranges from ‘very high cost’ to ‘very low cost’. The results indicated that the cost of AVs has a significant impact on the possibility of purchasing them. For instance, when the cost of the AV decreases, the likelihood of buying it increases. The p-value for the cost category ‘very high cost’ is 0.039, which is significant. The p-value for the cost category ‘high cost’ is 0.034, which is also significant. While the p-value for the cost category ‘average’ is 0.242, which is not significant. This also the case for the cost category ‘low cost’, with a p value is 0.239.

To assess the impact of the predictors (privacy, level of security, environmental impact, and cost) on the dependent variable (willingness to buy an AV), odds ratios were computed for each category change as presented in Table 6.

Regarding the predictor privacy, the change in the odds of being in a higher category of ’buy AV’ associated with a one-unit increase in Privacy, while holding other variables constant; for example, Privacy = 0 vs. Privacy = 4. An odds ratio of 4.721 means that compared to individuals who are most privacy-conscious, those with Privacy = 0 (the least privacy-conscious) had approximately 4.721 times higher odds of being in a higher category of ’buy AV’ (e.g., moving from 0 to 1, or from 1 to 2, etc.). It should be noted that the ’buy AV’ category where [buy AV = 4] is considered the reference category, and the corresponding odds ratio is not calculated since it would be compared to itself.

The odds ratio of 1.376 for the predictor level of security implies that, holding other predictors constant, individuals who rated security as the least important factor (Security = 0) had about 1.376 times higher odds of being in the ’buy AV’ category 0 compared to those who rated security as the most important factor (Security = 4). This suggests that the level of importance placed on security may affect the willingness of individuals to buy autonomous vehicles. This means those who value security highly may be more cautious and hesitant about adopting AV technology until they are convinced that it is safe and secure.

Regarding the predictor environmental impact, individuals who rated environmental impact as the least important factor (Environ = 0) had approximately 1.678 times higher odds of being in the ’buy AV’ category 0 compared to those who rated environmental impact as the most important factor (Environ = 4). This suggests that the level of importance placed on environmental impact may influence the willingness of individuals to buy an AV. Those who prioritise environmental considerations may be more likely to adopt an AV as it is perceived as being more environmentally friendly than conventional vehicles.

Regarding the predictor cost, individuals with Cost = 0 have approximately 2.26 times higher odds of being in the ’buy AV’ category 0 compared to individuals with Cost = 4. This implies that the cost may affect the willingness of individuals to buy an AV. Those who prioritise cost considerations may be more willing to adopt an AV if it is perceived as being more cost-effective compared to conventional vehicles.

Overall, the study analysed the impact of four predictors (privacy, level of security, environmental impact, and cost) on the willingness of individuals to buy autonomous vehicles (AVs). The odds ratios were calculated for each predictor, indicating the change in the odds of being in a higher category of ’buy AV’ associated with a one-unit increase in the predictor, while holding other variables constant. The results showed that the level of importance placed on each predictor had a significant effect on the willingness of individuals to buy AVs. Those who ranked privacy, environmental impact, and cost as low had higher odds of being in the ’buy AV’ category 0 (not interested at all), while those who highly ranked security had lower odds of being in the ’buy AV’ category 0 (not interested at all). These findings suggest that understanding individuals’ priorities and concerns related to AV adoption is critical to designing effective policies and strategies to promote the widespread adoption of AV technology.

6. Conclusions

This paper has presented results for the acceptance of AVs in Jordan. We hypothesise the willing of buying a self-driving car as the dependent variable with privacy and security, environmental impact, along with the cost as independent variables. The results indicated that there is significant correlation between the level of security and the likelihood of buying a self-driving car. The results also indicated that the self-driving cars that are classified as environmentally friendly are more likely to be bought, compared to those classified as ‘neutral’. Finally, the cost of the self-driving car has a significant effect on the likelihood of buying the car. As the cost of the car decreases, the likelihood of buying it increases.

The limitations of this study could be summarised as follows: the sample could be biased toward women. For example, about 68% of the participants in the study were women compared to 32% men. According to the Jordanian Department of Statistics [8], the population consists of 48% women compared to 52% men. Therefore, the gender impact on purchasing AVs was not carried out.