Social Impacts of Ridesharing Platforms: Evidence from Platform-Based Drivers in a Southeast Asian Context

In recent years, the sharing economy has emerged as a disruptive force that reshapes how individuals interact, access services, and engage in economic activities. Among its most transformative applications is urban transport, where ridesharing platforms have redefined the traditional taxi industry and introduced new dynamics to labour markets and social relationships. These platforms enable flexible work arrangements for drivers, foster digital customer interactions, and promote value systems rooted in peer-to-peer labour exchange. However, their rapid expansion has also raised questions about labour rights, regulation, and long-term social impacts.

While existing studies have extensively examined the economic and employment effects of ridesharing—such as increased labour force participation, improved accessibility, and price competition—most of this literature is concentrated in developed contexts and focuses on supply-demand or consumer behaviour dimensions [1]. Much less attention has been paid to the social consequences of ridesharing platforms, particularly from the perspective of driver-partners. These include changes in personal well-being, social connectivity, and perceptions of work autonomy, which remain underexplored, especially in developing countries.

This study addresses this research gap by focusing on Vietnam, where Grab has become the potential ridesharing provider in urban areas. The country presents a relevant context for investigating how digital platforms alter labour conditions and social life, given its rapid urbanisation and evolving digital infrastructure. This paper explores how ridesharing platforms contribute to new labour market characteristics and emerging digital values, and how these dimensions, in turn, influence drivers’ social perspectives.

The novelty of this research lies in its use of primary data from over 2,000 Grab drivers across Vietnam and the application of Covariance-Based Structural Equation Modelling (CB-SEM) to examine latent relationships among labour structure, values, and perceived social outcomes. By integrating perspectives on digital labour and value systems, the study provides a multidimensional understanding of how platform work is reshaping urban social life.

Accordingly, this research aims to examine how the ridesharing economy affects drivers’ social perspectives by analysing their perceived autonomy, engagement, and satisfaction. Specifically, the study seeks to:

(1) Identify key indicators that shape new values and labour structures in the platform economy;

(2) Explore the interrelationship between these two constructs;

(3) Assess the extent to which they influence drivers’ social well-being and integration.

An approach combining literature review, quantitative data collection and statistical tests was used in this study. The first step presented a review of the current literature to single out measures that can enhance social perspectives regarding wages, job fulfillment and social inclusion of partners in the ridesharing economy. This section describes the method, statistical tests and structure of this study. The whole process is illustrated in Figure 1.

1.png

Figure 1. Theoretical framework for analyzing

3.1 Quantitative sampling and data collection

Quantitative data were collected through a nationwide survey conducted in 2020 across the South, North, and Central regions of Vietnam, where the Grab company provides pilot ridesharing to customers. The sampling process followed a combination of purposive and convenience techniques. Driver-partners were selected based on their current activity on the Grab platform and were required to have at least six months of driving experience to ensure their familiarity with the ridesharing work environment.

The questionnaires were pre-tested with 50 driver-partners before official distribution to ensure clarity, consistency, and ease of understanding. After final revision, two formats of the questionnaire—both printed and online—were distributed. Printed copies were delivered directly at Grab offices and driver hubs, while online forms were shared through social media groups and driver communities. This mixed approach facilitated broader access and increased participation.

Approximately 2,300 questionnaires were distributed, and 2,213 valid sets of responses were received, resulting in a response rate of over 96 percent. Among the respondents, 2,091 (94.49%) were male, 120 (5.52%) were female, and 2 (0.09%) identified as other. In terms of age, the largest group was between 30 and 34 years old (23.14%), followed by those aged 35 to 39 (19.79%) and 25 to 29 (17.98%). Regarding experience on the Grab platform, 31.19 percent of respondents had worked between one and two years, 22.94 percent had two to three years, and 18.18 percent had more than three years.

In addition to questions for demographic and descriptive purposes, a 5-point Likert scale was used for key questions to measure responses, where 5 was the highest score and 1 the lowest. The Likert scale is suitable for comparing among groups and is appropriate for use in structural equation models.

Although some participants did not answer every question, missing values were minimal and handled using listwise deletion in the modelling process. These omissions did not significantly affect the analysis results. Table 1 presents key demographic characteristics of the respondents.

Table 1. Demographic characteristics of the respondents

Notes: Some respondents work both as motorbike and car partners for Grab company

3.2 Data cleaning and reliability analysis

Before modeling multi-equation structures, it is necessary to clean the data and delete outliers so that results from the statistical models are reliable sources of evidence for proposing suggestions. The cleaning process involved data inputting staff checking the answers as several respondents did not select the choices using the 1 to 5 scale properly. In some cases, if there were no responses from interviewees, inputted values were treated as missing. Firstly, exploratory factor analysis and principal component analysis were used to ensure the deletion of any items for which values did not reach the accepted level. Fortunately, the values of all exogenous and endogenous variables were higher than 0.4, so they were retained for modeling purposes, see Table A1 in the Appendix for more details.

To ensure the items in the models were suitable and reliable, Cronbach’s alpha was calculated. Cronbach’s alpha presents the reliability or internal consistency of each essential item and the whole set of latent variables. The accepted value of Cronbach’s alpha is from 0 to 1 and excellent values are at around 0.9, so the calculated values for the items in this study were within the permitted limits [13]. There were no items deleted as a result of this reliability test for three sets of latent variables. Cronbach’s alpha values for each item are shown in Table 2.

Table 2. Non-parametric analysis of observed indicators

Notes: The accepted value of Cronbach's alpha is 0.7. *** and ** response to significant levels of 1% and 5% in consequence

3.3 Kruskal-Wallis analysis

The Kruskal-Wallis test, similar to an analysis of variance test, was adopted to check whether the medians in different groups of driver partners were statistically significant at p< 5%. The results were used to determine whether partners in distinct groups responded differently to economic, social or behavioural changes after joining the ridesharing economy. As presented in Table 2, the Kruskal-Wallis chi-squared values show the significance of medians among items such as income, happiness level, connections and increased interaction with relatives, family members and friends or customers. Only three questions from the latent constructs of the new labour market and new values of the respondents’ medians were statistically significant at 1% and 5% (c171, c179 and c185). This means that most of the items within the two latent exogenous variables can be assessed in one measurement. However, further analysis on social views could be carried out for specific age groups partners with regard to income, happiness level or exchanging more with others.

3.4 Modeling structural equations

Structural Equation Modelling (SEM) is a multivariate technique suitable for testing relationships among latent constructs and observed variables. This study employed Covariance-Based SEM (CB-SEM) rather than Partial Least Squares SEM (PLS-SEM), based on both theoretical and statistical considerations. CB-SEM is more appropriate when the research objective is to confirm theory-driven hypotheses and test overall model fit using global goodness-of-fit indices [14, 15]. Given that our model is grounded in a well-defined conceptual framework—linking new labour dynamics and value systems to social perspectives—CB-SEM enables a rigorous assessment of model structure, error terms, and factor loadings. Moreover, the large sample size (n > 1300 after missing data exclusion) meets the assumption requirements for maximum likelihood estimation in CB-SEM, further justifying its use.

SEM is used with confirmatory factor analysis (CFA) to measure contributions and influences among indicators and constructs. Specifically, CFA was conducted to validate the measurement structure of the latent constructs within new values, the new labour market and social aspects relating to the self-estimation of partners. With CFA results, if any items have more influence than others, suggested implications would be proposed for further implications.

3.5 Model fitness

Unlike the technique of PLS-SEM, in which R squared is appropriate to manifest the goodness of a model, the results in CB-SEM are checked with several criteria, including Chi-Square, Root Mean Square Error of Approximation (RMSEA) and Comparative Fit Indices (CFI). Table 3 provides more information about these criteria calculated from the initial model and the final model. In this table, other indices such as AIC, BIC, Standardized Root Mean Square Residual (SRMR) or CD are also presented to measure the goodness of fit after running CFA and SEM.

Table 3. Model fit indices

Notes: *** denotes statistical significance probability with p < 0.001

Different indices are applied in checking the goodness of fit. The two most popular indices, which are commonly used to determine model suitability are RMSEA, which is preferred at intervals of 0.01 to 0.08 (0.08 to 0.1 represents a fairly good fit) and CFI or TLI, which should be close to 1, as shown by Chen et al. [16].

3.6 Hypothetical model

CFA was used to understand the two main latent variables and their impact on the social perspectives as an endogenous latent variable. Each of the two latent explanatory variables included ten items and the endogenous latent construct had four indicators. The directional coefficients between the two exogenous variables and the endogenous construct provide a foundation for interpreting their relationships. To signal an addition, Figure 2 illustrates the initial model of interactions among these items and constructs in which the first order includes three reflective measurements and the second order consists of specific items within the variables of new values, new labour market and social perspectives.

To improve model fit, two error covariances were added based on Modification Indices (MI) suggested by Stata. The first was between c172 and c173, both measuring perceived efficiency in utilizing idle private assets, which are conceptually and semantically related. The second was between c179 and c1710, which both reflect drivers’ social engagement and network formation through the platform. These covariances were theoretically justified and led to improved model fit indices in the final model.

2.png

Figure 2. Initial model of social perspectives and the ridesharing economy

3.7 Confirmatory factor analysis

CFA was used to measure the latent variables composing the indicators as scheduled. The total sample size was more than 2200, but the statistical program excluded missing cases. Thus, the observations in the models contained 1384. Maximum Likelihood technique and standardized results were reported in the final model, as shown in Table 4.

Table 4. Results of initial model hypothesis

Notes: P – Significant probability; *** denotes statistical significance probability with p < 0.001

Improvement to the goodness of fit of the initial model was achieved through modifications made with suggestions from the statistical program by adding covariances between the error terms of the indicators. When adding these co-directions, this study carefully considered the theoretical links between items within new values, new labour market and social perspectives. Figure 3 illustrates the final results for indicators, latent variables and impacts, and Table 5 provides the results in detail, including estimates, standard errors, z values and significant signals.

3.png

Figure 3. Final model of social perspectives and the ridesharing economy

Table 5. Results of final model hypothesis

Notes: P – Significant probability; *** denotes statistical significance probability with p<0.001

5.1 Summary findings and contributions

This study examined the social impacts of ridesharing platforms on urban transport labour by focusing on two key constructs: the emergence of new labour market characteristics and the formation of new value systems. Using CB-SEM on survey data from over 2,200 Grab driver-partners in Vietnam, the findings revealed that both constructs have a significant and positive effect on drivers' social perspectives, including their perceived well-being, connection with others, and sense of professional identity.

The results suggest that digital labour platforms not only reshape employment structures by enabling autonomy and flexibility but also facilitate new forms of value co-creation and community engagement. These insights contribute to the growing body of literature on platform economies, particularly by highlighting social dimensions that extend beyond economic outcomes. The study also adds empirical evidence from a developing country context, where limited research has focused on the lived experiences of ridesharing workers.

From a practical standpoint, the findings have implications for policymakers and platform managers. Policy frameworks should recognize platform workers as a distinct labour group with evolving needs, including access to training, digital tools, and forms of social protection. Platforms may consider supporting driver communities and improving engagement practices to sustain satisfaction and long-term retention. Furthermore, insights from this study can inform the development of labour codes and smart mobility strategies that align with inclusive and sustainable urban growth.

5.2 Limitations and future research directions

This study has several limitations. First, the research was conducted in the specific context of Vietnam, where labour laws, cultural norms, and platform development differ from other regions. Second, the analysis was based on cross-sectional data, which may not capture evolving perceptions over time. Third, the study focused solely on Grab drivers, excluding other platform types or service models. Future research should explore longitudinal perspectives, comparative studies across countries, or qualitative approaches to better understand driver narratives and hidden forms of control or stress.

Despite these limitations, the study provides valuable insights into how platform work reshapes social experiences in urban transport, offering a foundation for further academic inquiry and informed policy design.