The Validity and Reliability of Perception of the Traffic Safety Survey Questionnaire for Active School Travel: A Pilot Study

Abstract

Background: Although there is a considerable body of research evidence on active school travel (AST), the number of children walking to school has decreased over the years in Australia, as well as around the world. Different factors influence AST; the most cited in Melbourne is traffic safety perception. Traffic safety perception is influenced by built environment elements, and improving the built environment can enhance parental perception of traffic safety. Studies have shown that lateral separation from traffic and the provision of a buffer improve the perception of traffic safety, and this has to be explored for children walking to school based on the existing ground situation on a typical street near a school in Melbourne. Based on this background, a pilot study was carried out before the main study to test the reliability and validity of the survey questionnaire. Methods: The survey instrument was developed based on perceptions and/or AST studies, and included safety and probability aspects of the perception construct to elicit responses on perception. The perception of traffic safety was to be rated based on the streetscape videos embedded in the survey. The reliability was tested using Cronbach’s alpha and validity was explored through exploratory factor analysis. The study also checked the feasibility of the recruitment method and whether there would be an observable outcome from the study. The participants were recruited online through community Facebook groups. Results: The survey instrument had excellent reliability (α = 0.945) and was valid. The recruitment method through a Facebook community group was apt for recruiting participants. The preliminary analysis of the pilot data revealed a difference in perception ratings based on the streetscape element intervention. Conclusions: The survey instrument can be used for similar AST research, as it proved to be reliable and valid.

1. Background

Active school travel (AST) involves commuting to school by active modes like walking, biking, etc. The number of children actively commuting to school has been declining worldwide. Research from Australia shows that the proportion of children commuting independently decreased sharply from 61% in 1991 to about 32% in 2012. In Melbourne, only 35% of children walked to school, while 60% were driven, and a very small percentage used bikes or public transport. The World Health Organization recommends at least 60 min of moderate to vigorous physical activity daily. Children with a sedentary lifestyle tend to experience poor health outcomes, such as reduced sleep, poor cardiometabolic health, decreased fitness, and behavioural issues.

Many factors influence AST, and traffic safety perception is among the main factors affecting the number of children actively commuting to school. Although there is substantial research evidence on AST, there is a lack of evidence that offers feasible and implementable results to improve the perception of traffic safety [1]. Research has shown that parental perception of traffic safety is associated with built environment elements such as intersection density [2], presence or absence of footpaths [3,4], road crossings [5], road type [6], traffic volume or density [7,8], and speed [6,9]. Many elements of the built environment cannot be easily modified, as doing so would require large-scale interventions that are not feasible.

On the other hand, a study on pedestrians in the United States has shown that lateral separation from traffic and the provision of buffers improve the perception of safety in pedestrians [10]. Street buffers, such as street trees, have been explored in many studies and have been found to influence the speed of cars by forming a visual boundary for drivers, as well as calming them psychologically and subsequently reducing traffic speed [11]. Barriers such as pedestrian fencing have been used near schools with a large volume of traffic and where there is dangerous traffic to stop pedestrians from stepping onto the road [12]. These studies have used mostly survey questionnaires to elicit responses on perception [13,14,15,16,17]. Others have used photos/3D visualization [6,18,19,20,21,22], field surveys with paper [10] or a field survey using a mobile GIS application [23]. Different scenarios of children playing on the footpath and crossing the road were presented in pictures [20]. In another study, references were made to built environment features, such as traffic, speed, or footpaths, to find the perception of safety [24].

Improving perception through small-scale intervention is yet to be explored for AST. Based on this background, a pilot study was conducted. The pilot study is at a smaller scale to determine whether the recruitment approach, methods, and survey instrument are feasible for the main study [25,26,27]. This study does not find answers to the research questions, but tests the methods and procedures before the main/large-scale study [27]. The study’s research question is to determine whether the perception of traffic safety varies with different streetscape elements, and to find why parents perceive the streetscape as safe or unsafe from traffic. In this pilot study, a survey questionnaire was developed and tested for reliability and validity. In addition, a preliminary analysis is carried out using basic statistical tools to determine if there would be answers to the proposed research question. The pilot study was carried out using a prototype street near a school as a base for interventions.

Socio-Ecological Model

Many health-related models exist for PA (physical activity) [28]. Of these models, the SEM (socio-ecological model) is used for research related to active travel (walking/cycling). The model indicates that multiple levels of factors influence human behaviour [29]. These factors are at individual, social (perception), environmental (natural/built environment), and policy levels. The factors at these different levels interact to influence PA behaviour changes [30]. For example, commuting distance (built environment factor) is influenced by school choice, urban sprawl policies, and the neighbourhood [29].

Among the multiple levels of factors, the built environment has come under observation recently, as there are reduced levels of PA [31]. The built environment is the man-made surroundings consisting of amenities like schools, houses, parks, etc., connected by a transport network [32]. The SEM model lacks built environment specifics that influence physical activity behaviour [33], and this could be the reason for the inconsistent approach in defining/measuring the built environment for physical activity research [31]. In addition, there is also a lack of specific behaviour-related interaction variables [34] or information on the degree to which the built environment influences the type of PA [33]. A study [35] has summarized factors of multiple levels of SEM based on the evidence/review papers. Parental perception of traffic safety is a social-level factor and is the result of interaction with the built environment (environment level). This is evident as perception was found to be associated with built environment determinants.

2. Methods

A concurrent mixed-method approach was adopted, wherein quantitative and qualitative data were collected simultaneously. The data were collected at the same time through an online survey. The reliability of the survey was tested using Cronbach’s alpha. The validity was checked using exploratory factor analysis. A preliminary analysis of the data was also conducted to check if there would be an observable outcome for the main study. The quantitative data analysis was carried out by finding the mean and standard deviation. The qualitative data were analyzed using thematic analysis. This analysis involved identifying, analyzing, and reporting patterns (themes). The questions of the qualitative section were analyzed together because the responses indicated that these questions were similar. First, the individual responses were coded, and then these codes were grouped to generate the themes.

2.1. Participant Recruitment

Owing to the COVID-19 pandemic, collecting data was challenging as there was no direct access to prospective research participants. The initial proposal to collect data by contacting schools was dropped as the Department of Education and Training, Victoria (Australia), halted the ethics application process for schools and access to schools due to the increased workload on schools arising from COVID-19. Most of the earlier research on AST [36,37] was conducted by contacting schools to recruit research participants. Given the situation, the data collection was carried out online on Facebook (2021, Meta, Menlo Park, CA, USA) after the ethics approval.

Based on the number of councils within metropolitan Melbourne (that is, 31 councils), a search was made for Facebook groups based on these council names. There were approximately 33 community groups for councils. The administrators of the 33 community groups were contacted, requesting data collection through their community members. Only 13 community group administrators responded and provided permission to post on their Facebook community group page. The project brief and the link to the survey questionnaire (see Figure 1) were shared on 13 Facebook community pages for 25 days in October 2021. The number of members on these pages ranged from 1800 to 21,000. The post had to be reposted from time to time to maintain visibility at the top of the page to recruit more research participants. Approximately 53 participants were recruited through online community Facebook groups in Melbourne, but only 31 provided valid responses that were used for the analysis. The participants were the parents/carers of children enrolled in public primary schools across Greater Melbourne.

2.2. Risk Perception Dimension

There is a lack of proper theory on how risk is processed in human brains [38]; therefore, studies have used various aspects to measure risk. Studies have used only one dimension, such as “how safe/would feel safe,” Refs. [10,23], “how dangerous,” Ref. [16], or probability of injury [15], to find the perception of traffic risk. These safety aspects have been defined as a general measure of risk [39], while in another study, it is defined as the affective aspect of the risk perception [14]. The literature on risk perception indicates that risk perception consists of affective (emotion/feelings experienced when exposed to the risk) and cognitive aspects (probability assessment). Rundmo and Iversen [14] have used safety and probability aspects to measure risk perception. The structure of this pilot study’s survey questionnaire is adapted from Rundmo and Iversen [14], which is about risk perception and driving behaviour before and after the traffic safety campaign. The research evidence has used safety, worry, and concern to measure the affective component and probability assessment for the cognitive aspect of the risk. These studies were used as a basis for developing a survey questionnaire. For the affective component, only the safety aspect has been considered for this study.

2.3. Survey Questionnaire

The survey questionnaire started with a project background, followed by an informed consent statement. The participants agreed to participate in the pilot study through a consent form. The main questionnaire was divided into three parts: (1) the demographics of the respondents (parents/carers), (2) the demographics of the child, and (3) perception-related questions (quantitative and qualitative). Some of the sociodemographic questions were adapted from [21,40,41]. The third section of the survey consisted of perception-related questions alongside streetscape intervention videos embedded within the survey. The questions were intended to measure the safety and probability aspects of the perception construct based on the streetscape intervention videos. The safety aspect of the perception-related question was “How safe do you think it is for your child to walk on this street?” to be rated on a 1–10 visual analogue scale from 1 (not safe at all) to 10 (very safe). It was followed by a second question on the probability aspect, which was “How likely is it for a child to be injured in a traffic accident on this street?” measured on the same scale from 1 (very likely) and 10 (very unlikely). The videos were 20 s long and were looped to be presented to the participants. The videos are point-of-view videos seen from the pedestrian’s perspective on the footpath.

The different streetscape intervention videos included the provision of pedestrian fencing, low- or high-canopy street trees, low or high speed, widened footpaths, and multiple-intervention streetscapes. The car motion was created via frame animations in Adobe Photoshop (Version 22.5.1, Adobe Inc., San Jose, CA, USA), and the traffic sound was incorporated from the Zapsplat www.zapsplat.com (accessed on 20 August 2021) free sound website. The car speed in these streetscape videos was high (to depict the existing speed limit of 40 km per hour near schools in Melbourne), except for multiple streetscape and low-speed interventions. Owing to technical limitations, the exact speed could not be set.

Table A1. Snapshot of streetscape intervention videos embedded in the survey questionnaire.

Sl. No.	Video Snapshot of the Streetscape Intervention	Interventions
1		High Speed: A prototype street near a school was photographed from a pedestrian’s viewpoint. The street is a two-way lane with an approximately 1.2 m wide footpath and a nature strip of the same width. Most of the streets in Australian residential settings are two-way streets with parking areas, footpaths, and nature strips [62].
2		Low-Canopy Trees: The low-canopy trees are incorporated into the nature strip.
3		Low Speed: The traffic speed is reduced in this case.
4		Widened Footpath: Based on the ground situation near schools in Melbourne, there is a need to provide a wide footpath to cater to the high volume of traffic during pick-up and drop-off hours at schools for comfortable and conflict-free walking. The footpath is widened to accommodate the high volume of pedestrians near schools. Only a 0.5 m buffer of the nature strip was maintained as a buffer from the traffic.
5		High-Canopy Trees: High-canopy trees were incorporated into the nature strip.
6		Pedestrian Fencing: Pedestrian fencing was provided.
7		Multiple-Intervention Streetscape: This streetscape represents one of the possible options for streetscape design near a school. In this, the traffic lane width is reduced, owing to the provision of a bike lane. The provision of narrower traffic lanes with other streetscape elements closer to the roadway reduces the traffic speed [63]. The footpath was widened, and street trees were provided, in addition to an increase in the distance from the vehicular traffic, owing to the provision of the bike lane.

in Appendix A shows snapshots of the videos embedded within the survey questionnaire that were presented randomly using a Qualtrics randomiser (2021, Silver Lake, Menlo Park, CA, USA).

The qualitative perception-related questions were designed to find the reasons why the streetscape elements were perceived as answered in the earlier section of the perception-related questions. The section starts with the question “Please provide reason(s) for the streetscape(s) that you have rated ‘very safe’ in the earlier part of this Survey. Which streetscape elements made it feel safe?” It was followed by the questions “Do you think that different streetscape elements make the street safe from traffic? Please explain why you think this?” These qualitative questions were to understand the reasons for perceiving as indicated in the earlier section of the perception survey.

3. Results

3.1. Participants

As per [42], the recommended sample for the pilot study is about 10% of the main study sample size, while [43] has recommended a size of 10–30. Specifically, Ref. [44] recommends 25–40 for the survey instrument preparation. The sample size for this pilot study is adequate. Among 31 respondents, the highest number of respondents were mothers, followed by fathers (refer to

Table 1. Participant (relationship with the child) frequency and percentage.

Relationship to Child	Number of Participants	Percentage
Mother	25	80.64
Father	5	16.13
Grandmother	1	3.23
Total	31	100

). These participants were mostly from the age range of 35–44 years, followed by the age range of 45–54 years, and the rest were from other age ranges (

Table 2. Age range of the participants.

Age Range of Participants	Number of Participants	Percentage
25 to 34 years	2	6.50
35 to 44 years	21	67.70
45 to 54 years	7	22.60
55 to 64 years	1	3.20
Total	31	100

). The number of participants was highest from Australia as the country of birth, and the rest (n = 9) were from Belgium, Bhutan, China, England, India, Malaysia, New Zealand, Ukraine, or the USA (

Table 3. Country of birth of the participants.

Participant’s Country of Origin	Number of Participants	Percentage (%)
Australia	22	70.96
England	1	3.23
India	1	3.23
China	1	3.23
New Zealand	1	3.23
Other countries	5	16.12
Total	31	100

). The representation of the sample from these countries conforms to the data of metropolitan Melbourne [45], with the highest representation from Australia as country of birth, followed by other countries, especially China, India, England, and New Zealand.

3.2. Reliability of the Survey Questionnaire

The survey instrument is reliable when it consistently measures what it intends to measure. Reliability is measured using a correlation coefficient, and the coefficient is the correlation between the variables measuring the same thing [46]. The reliability/internal consistency of the survey instrument was tested using Cronbach’s alpha in IBM SPSS Statistics 28.0 (IBM Corporation, Armonk, NY, USA). The items are split into two in every possible way, the correlation coefficient is calculated for these splits, and the average of these values is Cronbach’s alpha [47]. The reliability was checked for a total of 14 items/questions, with each item on either safety or probability for seven streetscape intervention videos. The overall reliability was 0.945 (Cronbach’s alpha), which is excellent reliability (see

Table 4. Internal consistency of the items.

Sl. No.	Perception of Safety from Traffic (Based on Streetscape Elements)	Number of Cases	Cronbach’s Alpha	Cronbach’s Alpha (for All the Items)
1.	Low-Canopy Trees	29	0.864	0.945
2.	Pedestrian Fencing	29	0.931
3.	Widened Footpath	27	0.821
4.	High-Canopy Trees	31	0.845
5.	Multiple-Intervention Streetscape	30	0.866
6.	High Speed	29	0.853
7.	Low Speed	30	0.846

). Moreover, the reliability for individual items was above good and excellent, as it was greater than acceptable Cronbach’s alpha (0.7).

3.3. Validity of the Survey Questionnaire

The different categories of validity are construct, content, and criterion-oriented validity [48]. The data from the survey question is interpreted for its validity. Validity is used to verify if the survey/tool accurately measures what it proposes to measure [49]. Perception as a construct is investigated for construct validity, as it is a measure of quality/attributes that cannot be operationally defined [48].

The exploratory factor analysis (EFA) is a statistical tool used to validate and develop psychological theories or survey tools [50]. In the EFA, the data are reduced as the variables load onto factors that represent the construct. It has been used in this study to understand how the different aspects/dimensions relate to the perception construct and to infer the construct validity. EFA can only be used if the majority of variables are correlated (>±0.30) [51]; therefore, the correlation was checked for the variables (see

Table 5. Correlation matrix (correlation <0.30 in colour) with legend of variables.

		1	2	3	4	5	6	7	8	9	10	11	12	13	14
Correlation	1	1.00
	2	0.69	1.00
	3	0.66	0.62	1.00
	4	0.65	0.87	0.67	1.00
	5	0.00	0.22	0.22	0.19	1.00
	6	0.06	0.41	0.11	0.31	0.78	1.00
	7	0.81	0.72	0.66	0.58	0.16	0.34	1.00
	8	0.66	0.82	0.63	0.77	0.15	0.44	0.83	1.00
	9	0.64	0.66	0.65	0.55	0.61	0.60	0.69	0.64	1.00
	10	0.46	0.64	0.47	0.68	0.61	0.59	0.45	0.64	0.76	1.00
	11	0.23	0.43	0.26	0.41	0.68	0.66	0.40	0.47	0.53	0.69	1.00
	12	0.28	0.55	0.32	0.47	0.73	0.80	0.51	0.53	0.63	0.71	0.87	1.00
	13	0.62	0.76	0.68	0.74	0.12	0.29	0.70	0.88	0.55	0.61	0.47	0.47	1.00
	14	0.70	0.71	0.77	0.65	0.38	0.28	0.74	0.71	0.75	0.66	0.54	0.55	0.76	1.00
a. Determinant = 2.79 × 10−9

Legend: 1 Widened_Footpath_Safe; 2 Widened_Footpath_Likely_Injury; 3 HighCanopy_Tree_Safe; 4 HighCanopy_Tree_Likely_Injury; 5 MultipleIntervention_Safe; 6 MultipleIntervention_Likely_Injury; 7 HighSpeed_Safe; 8 HighSpeed_Likely_Injury; 9 LowSpeed_Safe; 10 LowSpeed_Likely_Injury; 11 Fencing_Safe; 12 Fencing_Likely_ Injury; 13 LowCanopy_Tree_Likely_Injury; 14 LowCanopy_Tree_Safe.

). A larger number of correlated variables could indicate that these variables are measuring the same underlying dimension. On the other hand, the highly correlated variables are not recommended [52], nor are variables that are perfectly correlated, to avoid extreme multicollinearity [47]. The correlation matrix (refer to Table 5) shows that the majority of the variables are significantly correlated. However, there are a few variables with low correlation < +0.30, while there are no highly correlated variables (>±0.90). Therefore, there is mild multicollinearity, which is not a problem for conducting EFA. The variables in the correlation matrix, for example, Widened_Footpath_Safe represent the data for the perception question on safety aspects related to widened footpath videos. Similarly, Widened_Footpath_Likely_Injury is for probability aspects of the widened footpath streetscape intervention video.

A further test, Bartlett’s test of sphericity, was carried out to check whether EFA could be used. This test is considered an objective test, whereas the correlation matrix test is subjective. Bartlett’s test of sphericity tests whether a correlation matrix contains ones diagonally and zeros off-diagonally. The test is represented by a significant chi-square value [50]. The test is followed by a third test, the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy, for which the value is measured from 0.00 to 1.00. EFA is unsuitable if the KMO value is less than 0.50, and EFA is appropriate if the KMO value is greater than or equal to 0.70 [53]. The KMO value for the pilot study is 0.77, and the chi-square (approx. 364.41) is significant (see

Table 6. KMO measure of sampling adequacy and Bartlett’s test of sphericity results.

Kaiser–Meyer–Olkin Measure of Sampling Adequacy		0.77
Bartlett’s Test of Sphericity	Approx. Chi-Square	364.407
	df	91
	Sig.	<0.001

). These tests show that EFA is appropriate for the matrix, and EFA was conducted.

EFA was conducted by selecting Analyze, then Data Reduction in SPSS (28.0, IBM Corporation, Armonk, NY, USA), and using the principal component method with direct oblimin rotation. The variables included for EFA were the perception score ratings for seven streetscape interventions (14 variables), which were presented in a correlation matrix. Since the underlying construct dimensions are related, direct oblimin rotation was used. The two factors (1 and 2) that have eigenvalues greater than one account for 78% of the variance (see

Table 7. Variation is explained by different factors represented by eigenvalues and the percentage of variation.

Initial Eigenvalues				Extraction Sums of Squared Loadings			Rotation Sums of Squared Loadings a
Component	Total	% of Variance	Cumulative %	Total	% of Variance	Cumulative %	Total
1	8.368	59.774	59.774	8.368	59.774	59.774	7.619
2	2.508	17.916	77.690	2.508	17.916	77.690	5.352
3	0.775	5.536	83.226
4	0.589	4.206	87.433
5	0.502	3.588	91.020
6	0.377	2.694	93.715
7	0.300	2.144	95.858
8	0.180	1.284	97.143
9	0.107	0.764	97.906
10	0.097	0.693	98.599
11	0.091	0.650	99.249
12	0.059	0.421	99.670
13	0.028	0.199	99.868
14	0.018	0.132	100.000
Extraction Method: Principal Component Analysis.

^a When components are correlated, sums of squared loadings cannot be added to obtain a total variance.

) and were extracted to find the factor loadings.

The pattern matrix shows the loading of each variable on factors 1 and 2. The loading represents the relationship between the variables and factors. A higher loading factor indicates that the variable represents that factor. A simple pattern matrix, which is easy to understand, will have significant loading on one factor compared with the other. As observed in the pattern matrix table (refer to

Table 8. Pattern matrix showing the loading of variables on extracted factors 1 and 2.

	Component
	1	2
Widened_Footpath_Safe	0.942	−0.232
Widened_Footpath_Likely_Injury	0.845	0.108
HighCanopy_Tree_Safe	0.858	−0.098
HighCanopy_Tree_Likely_Injury	0.828	0.054
MultipleIntervention_Safe	−0.190	0.976
MultipleIntervention_Likely_Injury	−0.060	0.921
HighSpeed _Safe	0.864	0.010
HighSpeed_Likely_Injury	0.870	0.086
LowSpeed_Safe	0.570	0.463
LowSpeed_Likely_Injury	0.440	0.596
Fencing_Safe	0.130	0.810
Fencing_Likely_Injury	0.183	0.846
LowCanopy_Tree_Likely_Injury	0.871	0.017
LowCanopy_Tree_Safe	0.804	0.166
Extraction Method: Principal Component Analysis. Rotation Method: Oblimin with Kaiser Normalization a

^a Rotation converged in 5 iterations.

), all the variables load onto one or the other factor significantly, except for LowSpeed_Safety and LowSpeed_Likely_Injury, where there is a cross-loading. This implies that a variable represents these two factors equally, which is conceptually wrong. This error may be owing to a technical limitation of speed representation. As speed is an important factor that influences objective and safety perceptions, the low-speed variables were included in the main study, which included an improved video with an exact speed representation to avoid cross-loading across the factors. Overall, the analysis showed that the survey is valid.

3.4. Preliminary Data Analysis

Quantitative data analysis revealed that the multiple-intervention streetscape was rated highest, followed by pedestrian fencing, low speed, high-canopy trees, low-canopy trees, widened footpath, and finally high speed (refer to

Table 9. Streetscape intervention with perception of traffic safety mean scores.

Streetscape Intervention	Perception of Traffic Safety (Mean Score)
Multiple-Intervention	7.19
Pedestrian Fencing	6.67
Low Speed	5.73
High-Canopy Trees	5.65
Low-Canopy Trees	5.58
Widened Footpath	5.40
High Speed	5.28

) in terms of safety aspects. The higher the perceived safety, the lower the probability of injury in a traffic accident.

Among the 31 participants in the pilot study, 25 responded to the qualitative questions. The qualitative data offer overall insights into what is perceived as a traffic safety risk along a footpath. Parents worry that their children might play or walk into the road or stumble onto it. Some parents also fear the risk of a car mounting the curb, while other participants doubt whether an accident can happen on the footpath. Streetscape elements influence perceptions of traffic safety because they act as barriers, with trees or fences specifically mentioned by some participants. Others point out that streetscape elements can reduce the impact of a crash on a pedestrian if an accident occurs. The presence of streetscape elements increases the distance from traffic and lowers the risk of children stumbling onto the road or cars mounting the footpath. Additionally, streetscape elements create space for errors made by drivers or children. While most participants believe these elements provide a sense of safety from traffic, one participant disagreed, stating that the streetscape creates a false sense of safety, which may lead pedestrians to take more risks.

4. Discussion and Conclusions

The traffic safety perception questionnaire for children walking to school is reliable and valid. The questionnaire can be used for similar research. The perception is measured using safety aspects (also referred to as a general risk measure) and the probability of injury in the event of collisions. However, it has to be noted that exploratory factor analysis was conducted on a small sample of 30 and this may affect the reliability of the factor analysis. There are many rules for the required sample size. Kass and Tinsley [54] recommended having 5–10 participants per variable or, on the higher end, about 300 per variable as recommended by Tabachnick and Fidell [55]. Meanwhile, Guadagnoli and Velicer [56] argue that as long as there are four or more variables that have >0.6 loadings, the factor analysis is reliable, irrespective of the sample. Additionally, a larger number of factor loadings of values less than 0.6 requires a larger recommended sample size. Based on this background on sample size for factor analysis, the sample size for this pilot study is appropriate. Moreover, the Kaiser–Meyer–Olkin measure of sample adequacy indicated that factor analysis is feasible for this sample size.

The recruitment approach through Facebook online communities was successful for the pilot study. However, the feasibility of this recruitment method for the main study is doubtful due to the large sample required. The rate of recruitment for the pilot study was, on average, one response per day. In addition, there were other concerns about not gaining access to a large number of Facebook community groups for the main study. The pilot study did provide the experience of recruitment via social media. Furthermore, the survey instrument is valid and reliable for research on parental perceptions of children walking to school. The preliminary data analysis revealed that there was an observable outcome from the study.

To my knowledge, this study is the first to add the dimension of probability of an injury in case of a traffic collision to the perception construct for AST research. AST research on perception has only used safety aspects [16,22,23] to determine perceptions of traffic safety. Moreover, there appears to be no proper theory that provides a holistic picture of the risk perception process in the human brain [38]. This study uses safety and probability aspects to measure perceptions for children walking to school.

Risk perception studies have been carried out for drivers [57], the public [10,58], adolescents [14], and cyclists and drivers [59]. There are few studies on perceptions related to children walking to school [16,23] and this study adds to the research evidence.

A limitation of this study is the visual representation of the streetscape videos. A basic form of 3D modelling was used, which may have had an influence on the result. The use of virtual reality modelling would have provided a flexible view of the traffic environment and accurate results. Having said that, earlier studies [6,20] have used pictures to elicit responses on perception, and these studies have produced reliable results.

The data was collected on Facebook, and this may be a limitation, as the access to a certain group of participants may have been impacted. As per the data, Facebook users are mostly aged 18 to 29 years old [60], limiting access to other age groups. In addition, only 13 out of the 33 Facebook groups gave permission to post the survey on their website. This may have also affected the survey results’ validity and reliability.

This study provided insights to improve the survey questionnaire for the main study and highlighted the potential challenges of participant recruitment through Facebook community groups. Collection of data from Facebook community groups would be a daunting task, requiring daily posting and an enormous amount of time [61], especially for the main study with its large sample size. The streetscape intervention videos were proposed to be redeveloped for better representation and especially to represent the exact speed. The survey instrument was improved for clarity, especially the qualitative section, where two questions had similar responses.

Future research, when developing a survey questionnaire for AST, could include worry/concern (affective component) or severity of the consequences in case of collision (cognitive component) or both as aspects of risk perception, in addition to the aspects used in this pilot study.