Perceived risk of using online transportation during the COVID-19 pandemic: A MIMIC-model approach

This study aims to assess the construct validity of the perceived risk scale on online transportation users during the COVID-19 outbreak at once to test whether this scale can produce information that is not influenced by individual characteristics using Multiple Indicators and Multiple Cause (MIMIC) models. The 251 participants from different regions in Indonesia, such as East Java, Central Java, West Java, and outside Java participated in this study. The confirmatory factor analysis (CFA) approach was used as the main method in data analysis. The result showed that this six-aspect contextualized version of the scale was fit. The MIMIC model showed that this scale was not influenced by covariate variables, such as age, domicile, sex, marital status, length of time using online transportation, types of online transportation services, and occupation (except private employee). For further research, the exploratory study of risk perception and analysis of the roles of socioeconomic factors are suggested to do.


Introduction
The World Health Organization (2020) stated that Corona Virus Disease 2019 (COVID-19) has become a global pandemic and still continues to this day.In Indonesia, Satgas Penanganan COVID-19 (COVID-19 response acceleration task force) reported that the number of positive cases of COVID-19 in Indonesia has reached more than one million cases (Satuan Tugas Penanganan COVID-19, 2021).Data from Komite Penanganan COVID-19 dan Pemulihan Ekonomi Nasional (Committee for Handling  shows that on the 16 th of June 2021, there were 1,531,005 positive cases of COVID-19 in Indonesia.This number increased by 8,161 compared to the previous day.The number of active cases also reached 83,700.Based on the detail per province DKI Jakarta ranked the highest in the positive case with 440,071 cases, followed by West Java with 328,940 cases and Central Java with 215,684 cases (Komite Penanganan Corona Virus Disease 2019 (COVID-19) dan Pemulihan Ekonomi Nasional, 2021).The increase in COVID-19 cases encourages the Indonesian government to make national-scale regulations because it threatens the security of all citizens (Everts, 2020).
New Habit Adaptation (NHA) is one of the Indonesian Government's policies to be applied by all Indonesian citizens.NHA consists of several points, such as 1) always use a mask when going out of the house; 2) avoid touching the eyes, nose, and mouth; 3) take a distance of more than one meter from other people when outside the house; 4) wash your hands with soap frequently; 5) always follow the development of information from trusted sources (Komite Penanganan Corona Virus Disease 2019 (COVID-19) dan Pemulihan Ekonomi Nasional, 2020a).Like the previous outbreak, most countries have remained the society to apply NHA especially when we travel to densely populated areas, visit hospitals, and use public transportation (Wang, 2014;Zhang, et al., 2019).
Another national-scale policy to minimize COVID-19 spread in Indonesia is Pemberlakuan Pembatasan Sosial Berskala Besar or PSBB (largescale social restrictions) referring to the social distancing concept.Society applying social distancing or PSBB tends to avoid the public area with large crowds like shopping centers, markets, education institutions, and hospitals (Majid et al., 2020).As a response to PSBB enforcement, people who live in business, education, and other sectors are continuing their activities by adapting new normal habits.It is because they assume that restrictions and new normal adaptation are very effective to minimize COVID-19 transmission (Alauddin, et al., 2020).
Compliance with the implementation of the PSBB also depends on the risk perception of the COVID-19 outbreak.How risk perceptions are owned is the basis for community decision for choosing to do preventive behaviors or not and travel or stay at home during the pandemic (Hilyard, et al., 2010;Majid, et al., 2020;Rayani, et al., 2021;Wong & Sam, 2011).Those who perceive that they have a lower risk of being infected with COVID-19, tend to keep doing outside activities and violate quarantine regulations even though they know the consequences.This phenomenon also occurs in other pandemic situations like SARS and H1N1 (DiGiovanni et al., 2004;Hernández-Jover et al., 2012;Kamate et al., 2009).
In Indonesia, society seems not to realize that going outside during PSBB is harmful.A survey conducted by Komite Penanganan COVID-19 dan Pemulihan Ekonomi Nasional (2020b) proved that only 38.75% of participants rarely left the house during the PSBB, while the rest was still left the house.Those who left the house chose online transportation, as 20.60% of the participants used online motorcycle taxis (ojek) and another 20.03% chose online taxis.
According to the survey, the number of people who tend to go outside using online transportation was still high; online transportation providers need to have a standard risk perception instrument in the COVID-19 pandemic setting to produce representative data with the user perception.It is crucial since the behavior changing and perception toward online transportation in the pandemic era will affect users' behaviors in the post-pandemic.
All instruments from previous studies have demonstrated good psychometric properties in reliability and construct validity.However, there has not been a single risk perception instrument of the COVID-19 outbreak that has been used or created in the context of online transportation during the COVID-19 pandemic.So far, we can only find the Perceived Risk Scale developed by Kusumayani et al (2019) in the context of online transportation before the COVID-19 outbreak in Indonesia.Therefore, this study aims to assess the psychometric properties of the Perceived Risk Scale of Online Transportation Customers in the COVID-19 outbreak.
This study also used the Multiple Indicators Multiple Cause (MIMIC) model to test construct validity by involving several covariate variables.By using the MIMIC model, the researcher wants to obtain a scale that is invariant or able to produce information that is not disturbed by the characteristics of the participants.MIMIC model is needed because if the scale is prone to population heterogeneity (demographic background, social and economic status, and environment) as covariate variables, it will tend to produce biased information (Widhiarso, 2012).Information bias from the data generated by the risk perception instrument may occur because the public's response to risk itself is influenced by factors such as daily experience and educational background (Dony et al., 2017;You, 2011).Thus, online transportation providers can enrich their understanding of risk perceptions among their users which can be used as consideration in policy or programs during and after the pandemic.

Participants
Convenience sampling was used to select participants from the population.The participants are 251 participants who used online transportation during the pandemic, with 192 participants are female and 59 are males.The average age of participants was 28.1 (SD = 8.52).In terms of marital status, 100 participants were married and the rest was single.Participants' domicile was divided into four categories: East Java (n=142), Central Java (n=44), West Java (n = 35), and outside Java (n = 30).The majority of participants were taking or had finished the bachelor program (n = 126).Regarding the occupations, 80 participants were private employees, 29 were entrepreneurs, 26 were civil servants, 45 participants were unemployed or working in other occupations, then the rest (79 participants) were students.For additional information, this research enabled participants to give information about how long they had used online transportation (M = 2.9; SD = 1.18) and how many service types were used in the pandemic situation (M = 2.16; SD = 1.23).

Measures
The researchers used a modification of the Perceived Risk Scale of Online Transportation Customers in Indonesia (Kusumayani et al., 2019).It consists of seven components called financial risk (RN), social risk (RS), time risk, physical risk (RF), functional risk (RG), psychological risk (RP), and overall risk (RU).As reported by developers this 7-component scale had good internal consistency (α = 0.888).In this study, the researchers did not include time risk as mentioned in the original scale because it was intended to measure the delay of a product could satisfy customers or not.Therefore, the researcher considers the time risk component irrelevant to the context of using online transportation during the COVID-19 pandemic.
The Perceived Risk Scale of Online Transportation Customers modified consists of 18 items with three items in each component.Each item would be rated by the participant on 5 points on Likert type scale (1 = extremely unlikely; 5 = extremely likely).Several examples items that had been contextualized are "Saya khawatir akan tertular COVID-19 setelah menggunakan berbagai macam layanan angkutan online" (I am worried that I will catch COVID-19 after using various online transportation services); "Saya merasa tidak aman menggunakan layanan angkutan online di masa pandemi" (I don't feel safe using online transportation services during pandemic); "Saya khawatir layanan yang diberikan oleh pengendara angkutan online di masa pandemi tidak sebaik biasanya" (I am worried that the services provided by online transportation drivers during the pandemic are not as good as usual).

Procedures
First, the researchers did the modification of the Perceived Risk Scale of Online Transportation Customers by adjusting the scale context to COVID-19 pandemic situations and the services that be used (online transportation).Next, the researchers did data collecting using the online method.Data was obtained using an online questionnaire (Google Form).This method was chosen because it could reach a wide range of participants and is quite helpful for improving the research process (Creswell & Creswell, 2018).Moreover, the data collection process was performed during a pandemic.Even though this study was using the online method, researchers are able to obtain the data from participants (see Participants) who represent internet users in Indonesia.This is based on the results of a survey by Asosiasi Penyelenggara Jasa Internet Indonesia (2020) which found that the majority of internet users in Indonesia are female, occupied as students or private employees, and are domiciled on Java.
The participants were asked to fill in the demographic variables such as gender, marital status, domicile, education level, and occupation.Afterward, they were asked to fill out the modified risk perception scale.Then, the researchers did data cleansing and created the dummy variables of socio-demographics, such as domicile and occupation.Data analysis was done using several statistical applications, such as Jeffreys's Amazing Statistics Program (JASP), Analysis of Moment Structure (AMOS), and Jamovi.

Data analyses
Participant risk perception was analyzed using a Multiple Indicator Multiple Cause (MIMIC) method, the expansion of confirmatory factor analysis (CFA) involving covariates in the model.The MIMIC model tried to facilitate the heterogeneity of the population by involving a set of predictors or covariates in the model (Muthén, 1989).The main benefit of MIMIC was that this method can be applied to a smaller sample size because there was only one model in each analysis process (Widhiarso, 2012).In this study, CFA was applied using the item parceling technique in each component of risk perception.
The mentioned sociodemographic variables and information about the use of online transportation were used as a covariate in the MIMIC model (see Table 1).The researchers created some dummy sociodemographic variables and selected a referred group to be compared with other groups.The dummy sociodemographic variables consisted of sex (males as a referred group), marital status (single as a referred group), domicile (East Java as a referred group), level of education (graduate program as a referred group), and occupation (civil servant as a referred group).Meanwhile, age, duration of online service use, and type of services used were needed as continuous variables.
The chi-square value, Tucker-Lewis Index (TLI), Comparative Fit Index (CFI), and Root Mean Square Error of Approximation (RMSEA) were used as the model accuracy index.Each index had its critical value to visualize the model fit.The model had a chi-square value more than the table chi-square value (p>0.05); the CFI and TLI values more than 0,95, and the RMSEA below 0,06 could be identified as a fit model (Hu & Bentler, 1999;Kline, 2014).

Internal consistency reliability
Table 2 presents the internal consistency reliability of the Perceived Risk Scale of Online Transportation Customers.Reliability analysis was carried out by employing the single-trial administration approach using the Cronbach Alpha coefficient.From this analysis, it was observed that this scale had good internal consistency (α = 0.894).In addition, the correcteditem total correlations for the items of the scale were strong, ranging from 0,349 to 0,689.

Confirmatory Factor Analysis
After performing item score packaging to the same component, the researchers did a CFA analysis of six components of risk perception in online transportation users during the COVID-19 pandemic.Six components of risk perception exclude covariate variables and without correlation between residual indicators showed non-fit model accuracy index (p = 0.000, CFI = 0.833, TLI = 0.721, RMSEA = 0,235).Through suggestions from modification indices (M.I), the researchers found a relatively high score of model-constrained covariance between several components.Next, the researchers did additional CFA analysis using components residual correlation as suggested by M.I.This process resulted in an acceptable model fit index (p= 0.049, CFI= 0.992, TLI= 0.975, RMSEA= 0.070).The modified model of CFA was applied in the MIMIC model analysis.

MIMIC Model of Risk Perception
The result of model fit showed transformation when covariates were included in the analysis (see Table 3).The result of the MIMIC model illustrates that the addition of sociodemographic variables implied to model fix index p = 0.002, CFI = 0.978, TLI = 0.945, RMSEA = 0.046).Nevertheless, the fit index of the MIMIC model was still acceptable, and all factor loadings remained significant.
The covariate difference among significant sociodemographic groups in the factor latent score or risk perception was only found in the participant's occupation, specifically private employee (see Table 4 and Figure 1).Private employees had a latent score of 0.364 higher than civil servants.In addition to private employees, students with a latent score of 0.275 showed that marginally significant there is a slight difference in risk perception with civil servants.However, the risk perception measurement was invariant or not affected by population heterogeneity as a covariate variable i.e., age, domicile, gender, marital status, level of education, length of time using online transportation, types of online transportation services, and occupation (except private employee).

Discussion
This study showed the use of a perceived risk model with six of 18 items verified by CFA.After that, it was verified using CFA in an overall sample that showed an average index fit model; next, this result was used for MIMIC model analysis.The risk perception of online transportation customers measurement using the MIMIC model proved that it was invariant to covariate variables such as age, domicile, sex, marital status, length of time using online transportation, types of online transportation services, and occupation (except private employee).In consequence, Borsboom ( 2006) stated that the invariant scale was able to accommodate the population heterogeneity.Before measuring risk perception, it is important to know that there are two ways to understand risk perception as a feeling or experimental system and risk as an analysis or rational system (Epstein, 1994;Slovic et al., 2004).A sociodemographic covariate is a rational answer because it contains personal data that can be identified as personal identity.Whereas the answer related to the length of time using online transportation and types of online transportation services represent the result of participants' experiment during the use of online transportation services.
The findings showed that private employees had a higher level of perception of the risk of the COVID-19 outbreak when using online transportation services than civil servants.We argue that private employees are the most vulnerable to losing their jobs or becoming unemployed during this pandemic.Therefore, the private employees who are vulnerable to losing their jobs will consider the use of economic resources, not least when using online transportation services.Previous research has also shown that unemployed individuals have a higher perception of risk as a cause of fear than working individuals (Green et al., 2021;Mahmood et al., 2020).This may be due to the uncertainty of the COVID-19 crisis which has caused economic instability, so they become increasingly anxious (Wang et al., 2020).
Moreover, students who use online transportation during pandemics have a marginally significantly different risk perception than civil servants.This finding indicates that students tend to not have appropriate risk perception, thus they tend to underestimate to reduce their fear and anxiety, which disturbs the preventive action (Mant et al., 2021).How the perception of the risk of COVID-19 in students is based on the behavior of seeking health information related to COVID-19 (Rayani et al., 2021) Thus, this study had proven that the Perceived Risk Scale of Online Transportation Customers during the COVID-19 pandemic when simultaneously used on a population consisting of private employees and students would tend to produce less consistent measurements.This needs to be taken into account because one of the most important measurement properties is the stability of the measurement results when applied to different individual characteristics (Widhiarso, 2012).In line with Chauvin (2018) that not considering the background related to how participants understand the risk before taking measurements can lead to inaccurate conclusions.
Generally, in the COVID-19 pandemic context, the health system and technology are related to society's risk perception.COVID-19 pandemic causes changes in several aspects, such as social patterns, demographic, and environmental.These changes cause the risk to the operation of the health system and can elicit the new measurement, ways, strategies, and risk management (Alauddin et al., 2020;de Amorim & de Andrade Guerra, 2020).Likewise, the machine learningbased technology used by online transportation providers can help them to modify their applications.Modifications that can be applied such as automatic tracking to notify if someone is infected by the virus, an early warning system to detect the mask use, and density of public transportation routes or safe lanes from the virus spreading areas (Moss & Metcalf, 2020).
As a response to the high public desire to travel during the COVID-19 pandemic, the online transportation provider makes some policies to prevent COVID-19 spread while using online transportation.For example, Gojek applies J3K policy or jaga kesehatan (take care of health), jaga kebersihan (keep clean), and jaga keamanan (keep safe) (Gojek, 2020).Grab as another online transportation provider also makes some policies to ensure the safety and comfort of its customer like restricting physical contact, providing medical tools (e.g.mask and hand sanitizer), and doing routine body temperature checks for Grab's partner (Grab, 2020).These regulations in this pandemic era can be seen as an adjustment between providers and customers to prevent the transmission of COVID-19 (Mondada et al., 2020).In addition, online transportation providers can use their internet-based applications to spread a balanced combination of efficacy messages and threat appeals to increase the self-protection responses which intend reduced customer risk perceptions.Birhanu et. al. (2021) have proven that internet-based media or applications are the main sources for people to get this information.

Conclusions
According to the results, it can be proven that the risk perception scale of online transportation users which has been contextualized with the COVID-19 pandemic situation has good internal consistency and the construct of this scale is fit.This scale can be classified as an invariant scalein other words, this scale can produce consistent measurement results because it can accommodate population heterogeneity.However, after finding out the difference in risk perception in online transportation users between occupations, the suggestion for further research is to continue the study of the roles of socioeconomic covariates toward risk perception.Researchers

Figure 1 .
Figure 1.Multiple Indicator Multiple Cause (MIMIC) model showing the impact of covariate variable on the risk perception as latent variable

Table 3 .
Factor loading and Model Fit Indices of Initial CFA Model, Modified CFA Model, and MIMIC Model