This paper aims to contribute to the literature on the labor market impacts of Uber’s entry in Brazilian capital cities. The analysis first characterizes patterns of employment flows, earnings dynamics and shifts in the composition of self-employed drivers using descriptive statistics and transition analysis. Subsequently, it leverages the staggered rollout of Uber and 99 across cities using a difference-in-differences framework to estimate the causal effects of platform entry on labor market outcomes in Brazil.
Using microdata for 2012–2021 from Brazil’s Continuous National Household Sample Survey, a nationally representative, quarterly rotating labor-force survey, we combine (1) descriptive statistics, (2) year-to-year transition matrices and (3) a staggered difference-in-differences design that exploits the city-level timing of app entry to isolate causal effects on unemployment, overall employment, working hours, earnings and hourly earnings.
Platform entry triggered rapid growth of self-employed drivers, disproportionately Black and secondary-educated. Most entrants were already employed and remained driving, while a modest share of unemployed workers transitioned into the occupation, suggesting a buffer role. The difference-in-differences estimates reveal no statistically significant effect of Uber and 99’s market entry on city-level unemployment rates. Among self-employed drivers, employment counts increase in later periods, yet the initial earnings uptick proves transitory: average labor income and working hours decline over time, resulting in statistically unchanged hourly earnings.
City-level samples limit statistical power, and the one-year panel may miss longer-term mobility. Future research could aim at investigating the impact on the formal labor market, trying to grasp how much the flexibility benefit of this type of occupation crowded out employment in a formal labor market context.
This study provides one of the first causal assessments of ride-sharing platforms’ labor-market impacts in Brazil and supplements it with detailed descriptive and transition analyses drawn from nationally representative survey microdata. It sheds light on how digital platforms reshape employment dynamics in economies with persistent informality.
1. Introduction
With the advent of transportation apps, new technologies have given rise to new forms of work. These apps have low entry costs and offer flexibility regarding when, where and how much to work, characteristics often associated with higher employment levels. However, these platforms have also been criticized for increasing precarity among drivers, who are considered independent workers rather than employees. In Brazil, the end of the last decade was marked by economic recession characterized by rising informal employment. Between 2016 and 2019, formal jobs decreased by 0.4%, while informal employment rose by 12% (Barbosa, 2019). This growth coincided with the emergence of platforms such as UberX and 99POP, which began operations in three major Brazilian cities in late 2015 before expanding nationally in 2016.
This study contributes to the literature by analyzing the entry of Uber and 99 into Brazilian capital cities through a three-step analysis using microdata from Brazil’s Continuous National Household Sample Survey (Pesquisa Nacional por Amostra de Domicílios Contínua, PNAD-C), a nationally representative labor-force survey conducted quarterly by the Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatística, IBGE). First, descriptive statistics evaluate sociodemographic changes among self-employed drivers before and after app entry. Second, a panel analysis examines transitions among different labor statuses before and after the apps’ entry, net of unobservable individual characteristics. Third, we propose to identify the effect of the introduction of ride-sharing apps on the labor market in general, on self-employment and on self-employed drivers. Taking advantage of their staggered entry into Brazilian cities, we apply a staggered difference-in-differences econometric model, using these different entries as potential treatments.
The literature on gig platforms in high-income contexts suggests drivers value flexibility and earning wages comparable to taxi drivers (Hall & Krueger, 2018). Critics argue these analyses overlook vehicle costs, potentially overstating real earnings (Berg & Johnston, 2019). Recent studies leverage the phased expansion of Uber to isolate causal effects on labor market outcomes. Chang (2017) shows reduced incomes among taxi drivers after Uber entry, while Li, Hong, and Zhang (2021) and Omberg (2024) find increased labor participation and lower unemployment rates in the US. These findings imply that, in the United States, app-based gig work can serve as a buffer for individuals between jobs, mitigating frictional unemployment rather than simply displacing existing work.
In middle-income countries, the labor market context can be remarkably different due to higher informality rates, weaker labor protections and sharper income inequalities. Segmentation and search models suggest that, in downturns, the informal sector absorbs workers who lack access to formal jobs, while low fixed-entry costs select those with liquidity but limited outside options; simple entry models further predict that any early earnings premia dissipate as competition intensifies.
Ridesharing rapidly gained traction in Brazil around 2015, precisely when unemployment rose and formal employment shrank. This context helps frame the surge of self-employed drivers; however, the overall consequences in such a setting remain underexplored. Empirical studies in Brazil indicate Uber entry negatively impacted taxi drivers’ earnings or showed mixed effects (Esteves, 2015; Oliveira & Machado, 2021). Carvalho and Nogueira (2023) highlight gig workers’ lower earnings and social protections, while qualitative research (Magaldi, Azaïs, Razafindrakoto, & Roubaud, 2024) notes drivers value flexible schedules despite these drawbacks, corroborating the patterns found by Hall and Krueger (2018). Nazareno (2023), employing a difference-in-differences framework, confirms increased driver entry but reduced earnings and hours worked.
The present article advances the literature by adopting a broader perspective on the labor market impacts of ride-hailing platforms. Rather than focusing narrowly on a single occupational group, we examine labor market transitions over time, incorporating both Uber and 99 – two major platforms that entered the Brazilian market around the same time. Using a rotating panel from PNAD-C, we trace individual transitions into and out of ride-hailing over time, enabling a dynamic view of labor reallocation. Our analysis encompasses multiple employment statuses and demographic subgroups, including variations in education, race and prior employment categories, to assess whether ride-hailing disproportionately attracts – or displaces – specific segments of the workforce. Methodologically, we move beyond a basic difference-in-differences approach by incorporating transition analyses and enriched event-study techniques, allowing us to capture temporal heterogeneity and dynamic effects. This extended framework offers a more comprehensive understanding of how ride-hailing apps reshape labor patterns in an emerging economy context, shedding light on both the short-term appeal of flexible work and the longer-term risks of declining earnings and weakened social protections for drivers.
Results show an increased representation of Black and higher-educated self-employed drivers, alongside declining hourly earnings. Transition analysis reveals rising movements from unemployment to driver occupations and increasing persistence in this type of work over time. The difference-in-differences event study indicates delayed employment increases following platform entry, initially accompanied by income growth that subsequently declines. Additionally, working hours decreased after the app introduction, with early adopters experiencing positive outcomes while late adopters faced negative effects. Finally, the entry of ride-sharing apps in Brazilian capital cities did not reduce unemployment rates. Importantly, diagnostic checks fail to support the parallel-trends assumption for some outcomes, which limits the strength of causal interpretation.
This article is organized as follows: first, we describe in detail the institutional background and entry of Uber and 99 in Brazilian cities and provide initial statistics of the phenomenon. Next, we present the data used and the empirical strategy chosen. Finally, we present the results, followed by a brief conclusion.
2. Institutional background and Brazilian context
Launched in 2010, Uber pioneered the e-hailing concept – requesting taxis via the internet rather than phone calls or street-hailing. By connecting passengers and drivers within one digital ecosystem, the platform reduces transaction costs by efficiently matching users based on proximity and ride distance, optimizing pricing and estimated arrival times. Additionally, Uber’s integrated navigation lowers entry barriers, enabling drivers unfamiliar with city routes to easily transport passengers.
Compared to traditional taxis, becoming an Uber driver in Brazil involves fewer requirements. Taxi drivers face strict regulations such as mandatory training and costly permits, restricting flexibility in market entry and exit. Uber drivers, conversely, primarily need vehicles that meet basic age and condition standards, significantly easing their market participation.
Uber began Brazilian operations in 2014, coinciding with the onset of a significant economic and political crisis. Available in the Supplementary materials section, economic growth rates were negative from mid-2014 through 2016, accompanied by rising unemployment and informality from 2015 onward. Even as the economy recovered in 2017, informality continued its upward trajectory, temporarily disrupted during the pandemic but resuming growth thereafter.
The growth trend of the informality rate was driven primarily by an increase in self-employment (Pero, Machado, & Fontes, 2022). As illustrated in Figure 1, the evolution of the number of self-employed workers by occupation highlights the drivers in red, revealing significant growth after 2015, which marks the entry milestone for UberX in many cities (99’s entry occurred later, in 2016). This trend may partly result from the entry of Uber and 99. However, it is difficult to make a definitive assertion, as the country was in the midst of an economic recession that began in 2014. Therefore, this change could also be attributed to the high and growing unemployment rate during this period.
Evolution of the number of self-employed by occupation. Source: Authors’ own elaboration with PNAD-C/IBGE data
Evolution of the number of self-employed by occupation. Source: Authors’ own elaboration with PNAD-C/IBGE data
Regarding its entry into Brazilian cities, Uber’s geographical expansion in Brazil went through several stages. Uber entered five major capitals in 2014–2015, expanded to coastal and nearby cities in 2016, and covered all 27 capitals – including smaller northern ones – by 2018. 99POP (the ride-hailing service created by local firm “99”) debuted in São Paulo (late 2016) and Rio (early 2017); most other capitals followed in the second half of 2017 (see Supplementary materials).
When analyzing the transportation sector data through the PNAD-C, we observe a significant increase in the employed population in this sector after 2015. There’s some stability and even a possible decrease in the employed population in the transportation sector from 2012 to 2015. After mid-2015, the employed population increased significantly, rising from 1.6 m in mid-2015 to 2.2 m by the end of 2019 (see Supplementary materials). Breaking down this growth into the occupations within the transportation sector, it can be noted that the growth is due to one specific occupation: drivers of cars, taxis and vans (see Supplementary materials). Although this occupation is broad and not limited to app drivers, it is notable that the growth began in 2016, the year when Uber’s and 99’s presences intensified in Brazilian cities.
Moreover, this evolution is particularly evident in the changes in occupational status. The self-employed category diverges from its previous pattern, starting to increase significantly from the years of greater presence of the apps in the cities. Delving into the average income, there is a sharp drop in earnings among the self-employed (charts available in the Supplementary materials section). While income levels for formal and informal workers remained relatively stable throughout most of the period, the self-employed experienced a steady decline, particularly after 2015. Notably, before 2015, self-employed workers in the transport sector had higher average earnings than formal workers. However, this pattern reversed in the subsequent years, with formal employment consistently yielding higher earnings than self-employment from 2016 onward.
3. Data and empirical strategy
A custom-built database was created, containing UberX entry dates across Brazil’s 27 capital cities, ranging from May 2015 to April 2018. Information on 99POP was also collected for 23 capitals through manual searches in newspaper articles.
The main data source for the analysis is the Continuous National Household Sample Survey (PNAD-C), conducted quarterly by IBGE. The PNAD-C gathers socioeconomic information – such as demographics, employment type, income and hours worked – from around 550,000 individuals in each quarter. However, the survey only indicates whether individuals reside in a capital city, not their specific municipality, so only capital city respondents were retained.
Each quarter, about 211,000 households are surveyed using a rotating 1–2(5) scheme: a household is interviewed in one month, excluded for two, then re-interviewed, repeating this five-cycle process before permanent exclusion. This structure creates an 80% sample overlap between consecutive quarters, enabling the construction of short-term panels to analyze labor market transitions across up to five quarters.
Two databases were used in the analysis. The first contains all surveyed individuals in a repeated cross-section format, where each row represents one person in a given quarter. The dataset includes individuals from 27 capital cities, with considerable variation in sample sizes across cities. Since the PNAD-C began in 2012 and Uber entered its first city in mid-2015 (using UberX as reference), the sample spans from Q1 2012 to Q4 2022.
Given PNAD-C’s rotating panel design, a second dataset was constructed as an incomplete panel, allowing for the analysis of labor transitions, as explained below. Individuals were tracked using identifiers based on the methodology by Ribas and Soares (2008). To reduce bias related to sampling weights, we followed Monteiro (2019), applying a two-step correction. First, weights were adjusted for attrition relative to the original sample. Second, sub-sample weights were corrected so that population estimates by sex and age group matched those of the full PNAD-C sample within each of the 77 estimation strata. The resulting calculation is as follows:
Where:
= new weight in T1, calculated for each individual j in the sample;
= weight for each individual j in the Continuous PNAD sample (V1028), in T1;
= total number of people interviewed in the geographic area represented by estimation stratum d, in T1;
= total number of people in the subsample, after losses, in the geographic area represented by estimation stratum d, in T1;
= population estimate produced by IBGE for estimation stratum d, sex s and age group i, on the reference date.
= population estimate produced with the subsample for estimation stratum d, sex s and age group i, on the reference date.
Furthermore, since the occupation of app driver does not have an occupational code, the following classifications were adopted to consider an individual as such: (1) Economic activity: 49,030 (Road passenger transport); Occupation: 8,322 (Private driver) and Occupational position: Self-employed.
The self-employed occupational category was considered to ensure that private drivers not affiliated with platforms are properly excluded. Although this category includes more than just app drivers, encompassing private drivers and self-employed taxi drivers, this longitudinal analysis allows for comparisons of this group before and after the entry of apps. While private drivers do not overlap with app drivers, taxi drivers may pose an issue. Since app drivers provide a service very similar to that of taxis, their entry also affects taxi drivers’ income. Therefore, any measures generated for this occupational set should be understood as net of the dynamics between taxi drivers and app drivers. Thus, from this point forward, we will refer to the individuals belonging to this occupational set as self-employed drivers.
The empirical strategy follows a three-step analysis to evaluate the effects of ride-sharing apps’ entry in Brazilian capitals. First, a descriptive statistical analysis by sociodemographic groups is conducted one year before and one year after the apps’ entry in the capital cities.
Second, we conduct a transition analysis to investigate how self-employed drivers were positioned in the labor market one year prior. This involves examining individuals identified as self-employed drivers in interview 5, focusing on their sociodemographic aspects during their first interview.
Third, we estimate the causal effect of Uber and 99’s entry with a staggered difference-in-differences (DD) model using the method proposed by Callaway and Sant’Anna (2021) for estimating group-time average treatment effect (ATT(g,t)).
Regarding the transition analysis, the panel component of the PNAD-C makes it possible to identify job transitions from different labor force statuses to self-employed driver before and after the entry of ride-sharing apps and vice versa (from self-employed driver to different labor force statuses). We select each self-employed driver who appears as such in their final interview and check their labor force status in the same quarter of the previous year (or check their labor force status one year later if they appear as self-employed drivers in their 1st interview). We then calculated for each quarter the proportion of self-employed drivers according to their labor force status in the previous period.
We consider four labor force statuses (Bouvier et al., 2022): “occupied,” “unemployed,” inactive “discouraged” (potential labor force; person is available to work but not seeking a job) and other “inactive” (e.g. students, retirees and out of the labor force). Within the “occupied” category, we further assess occupational positions: formal (employees with a formal labor contract, military personnel and civil servants); informal (employees and domestic workers without a formal labor contract, and unpaid family workers); self-employed and employer. Finally, we select only transitions from the job statuses of occupied workers to self-employed drivers to calculate changes in income. We investigate whether the transition from formal employee to self-employed led to increased mean earnings, if differences exist between job statuses, and how these income variations change over time.
Lastly, to estimate the causal effect of apps’ entry into a city, one can consider the presence of the apps as a treatment within a framework of the difference-in-differences (DiD) estimation model. As the treatment effect may have a dynamic character over time, using two-way fixed effects regression (Angrist & Pischke, 2008) would lead to a negatively biased estimator (Callaway & Sant’Anna, 2021). Therefore, the traditional estimation method is not sufficient to capture the proposed causal effect in this study. Thus, we investigate the literature of recent advances in DiD to accommodate the peculiarities of the research object; due to its flexibility and ease of implementation, the approach proposed by Callaway and Sant’Anna (2021) is chosen, with the following equation to be estimated:
Where represents the labor market variables to be analyzed: unemployment rate, number of people occupied in the city, labor income usually received per month, weekly worked hours and income per hour; and represents, respectively, the group and period dummy; represents the group-time ATT, since represents the group-time interaction dummy that allows the calculation of the ; finally, represents a vector of covariates.
The selected covariates are intended to account for potential inter-city differences that may affect the validity of the parallel trends assumption. These factors are likely to influence both the timing of ride-sharing app entry and labor market outcomes for self-employed drivers. Specifically, the covariates include the city’s location within a metropolitan area, the quality of public transportation and the overall level of economic prosperity.
A city’s location within a metropolitan region is an important factor, as municipalities situated in larger metropolitan areas were among the first to be served by ride-sharing platforms and may exhibit distinct transportation dynamics, potentially influencing drivers’ hourly earnings. The quality of public transportation services is another critical variable; the robustness and efficiency of these systems can shape both the necessity for and the attractiveness of ride-sharing services, thereby affecting overall demand. In this analysis, public transport quality is proxied by the year in which each municipality formalized its urban mobility plan. Lastly, a city’s level of economic prosperity, measured by its GDP, can influence both the demand for ride-sharing services and the economic viability of such platforms. Given that these factors may simultaneously affect both treatment assignment and outcome variables, their inclusion as covariates in the model is needed to reduce potential confounding bias.
4. Results and discussion
In this section, we present the results of the three-step analysis to evaluate the effects of ride-sharing apps’ entry on the labor market in Brazilian capitals.
4.1 Descriptive statistics of before and after ride-sharing apps
Table 1 shows the differences in the profile of employed workers one year before and one year after Uber and 99 entries, comparing the labor market in Brazilian capitals with the selective group of workers in the transportation sector, drivers and self-employed drivers. First, the table shows a gender difference, with women more prevalent in the overall labor market than in transport, and 93% of drivers are men, a share unchanged since ride-hailing apps arrived.
Demographic characteristics by entry timing of apps
| Characteristic | Capitals | Transport sector | Driver | SE driver | |||||
|---|---|---|---|---|---|---|---|---|---|
| 12m before UberX | 12m after UberX | 12m before UberX | 12m after UberX | 12m before UberX | 12m after UberX | 12m before UberX | 12m after UberX | ||
| Gender | Male | 47% | 48% | 86% | 90%* | 93% | 93% | 92% | 93% |
| Female | 53% | 52% | 14% | 10%* | 7% | 7% | 8% | 7% | |
| Color | White | 47% | 45%* | 41% | 38% | 49% | 43%* | 53% | 44%** |
| Black and Indigenous | 53% | 55%* | 59% | 62% | 51% | 57%* | 47% | 56%** | |
| Education | Incomplete middle education | 32% | 30% | 20% | 18% | 21% | 16% | 21% | 15%* |
| Complete middle education | 15% | 14% | 20% | 18% | 20% | 14%*** | 20% | 14%** | |
| Complete secondary education | 31% | 31%* | 54% | 58% | 54% | 60%*** | 52% | 60%** | |
| Complete higher education | 22% | 24%*** | 6% | 7% | 6% | 10%*** | 6% | 11%* | |
| Ocupational types | Employee | 69% | 66%*** | 54% | 50% | 17% | 16%** | – | – |
| Employer | 4% | 5% | 3% | 1%*** | 5% | 0%** | – | – | |
| Self-employed | 20% | 23%*** | 43% | 49%** | 78% | 83%*** | 100% | 100% | |
| Others | 6% | 6% | 0% | 0% | 0% | 0% | 0% | 0% | |
| SE driver | SE Driver 1 Job | – | – | – | – | – | – | 98% | 97% |
| SE Driver 2 Job | – | – | – | – | – | – | 2% | 3% | |
| Age | Average age | 34 | 35 | 41 | 42 | 45 | 43* | 46 | 43*** |
| Income | Average income (R$) | 3,827 | 3,796 | 2,971 | 2,683** | 3,583 | 3,034 | 3,610 | 3,117*** |
| Average income per hour (R$/h) | 16 | 18 | 10 | 11 | 12 | 12** | 12 | 13* | |
| Social security contribution | Yes | 73% | 72% | 67% | 66% | 53% | 52%* | 48% | 48%* |
| No | 27% | 28% | 33% | 34% | 47% | 48%* | 52% | 52%* | |
Note(s): *p < 0.1, **p < 0.05 and ***p < 0.01 represent statistical significance levels based on a Student’s t-test for means
Source(s): Authors’ own elaboration with data from PNAD-C/IBGE
Second, Black workers are already overrepresented in capitals, transport and driving; before the apps, however, self-employed drivers were mostly white. After entry, Black participation rose across all groups and, within a year, surpassed whites among self-employed drivers.
Analyzing the educational profile, the proportion of workers with secondary education is higher in the transport sector and among drivers than in the labor market. On the other hand, the proportion of workers with higher education is lower in the transport sector. However, there has been a significant increase in the proportion of workers with completed secondary and higher education among self-employed drivers, from 58% before the entry to 71% after the entry.
The average age of employed people in Brazilian capitals was 34 years old before apps’ entry and rose to 35 after apps’ entry. The average age in the transportation sector is higher and has risen over the period from 41 to 42. In the case of drivers, the average age is even higher but decreases after apps’ entry. For self-employed drivers, it went from 46 before apps’ entry to 43 after apps’ entry.
The occupational types in the capitals have a higher proportion of self-employed in the transport sector and among drivers than in the capitals’ labor market. In contrast, all other occupational types have a lower proportion. This higher share of the self-employed is even greater among drivers and increases in all the groups analyzed.
In terms of labor income, both median income and hourly income are higher in the capital and have remained almost constant over the period. The income of transport workers and drivers fell significantly after apps’ entry. However, when evaluating hourly income, there was a slight increase.
Finally, the percentage of workers who contribute to social security in the capitals is much higher than in the transport sector and among drivers. There has been a slight decrease over time. It is worth noting that most self-employed drivers do not contribute to social security (52%).
This analysis has shown that there have been significant changes in the profile of self-employed drivers in terms of color (an increase in the proportion of Blacks), education level (an increase in the proportion of workers with secondary and higher education) and mean labor income, which has decreased significantly after apps’ entry.
Regarding the statistical significance of these changes, we conducted an analysis using a t-test for the difference in means. The test provided statistical support for some of the previously described trends. Although the difference in gender does not present significance, the deepening of racial inequality is statistically significant among both the general driver sample and self-employed drivers. Regarding educational changes, all prove to be significant, suggesting an increase in the number of individuals with completed secondary or higher education, with a more significant increase in those with secondary education. Given that this occupation requires both a prior qualification (driver’s license) and the possession or rental of a car, both of which positively correlate with income and, consequently, education, the larger share of those employed with completed secondary education makes sense.
In terms of occupational positions, the increase in self-employed workers, although part of a broader labor market trend, appears to be related, in the transport sector, to the emergence of this new form of work. This new form of work also seems to have changed the age composition of drivers, as there was a statistically significant reduction in the average age. Both income and social security contributions show changes; although social security contributions have only decreased slightly, income shows more substantial changes.
4.2 Changing dynamics to self-employed drivers
To provide a more detailed overview of the increase in drivers' self-employment, we conduct a dynamic analysis of transitions from different labor force statuses (occupied, unemployed, discouraged and inactive) to self-employed driver over time. The analysis of individual transitions paints a more accurate picture of the dynamics at work, highlighting phenomena that are inaccessible to traditional analysis. We will look at the original status of individuals before they became self-employed drivers as well as the changes in their income following this transition. The key question to address is which type of transition has changed the most since the entry of ride-sharing apps. Has the proportion of transitions from unemployment to self-employed drivers increased after the entry of ride-sharing apps? Or has the increase been more significant from inactive status?
Assessing the annualized proportion of individuals who became self-employed drivers in the following year, categorized by labor force status, many drivers were already employed in some occupations before transitioning to self-employed drivers. However, an increase in transitions from unemployed status over the years is identifiable, indicating the attractiveness of this form of work for individuals (these evolutions can be verified in the Supplementary materials section).
When examining the labor force status of the self-employed drivers in the previous period, many of them have already the same occupation, and their proportion has increased during the pandemic, rising from 48% in 2012 to around 60% in 2019 and 2020. But in parallel, many individuals from other occupations have also migrated to become self-employed drivers. The growth of the unemployed portion after the entry milestone is evident, along with a slight decrease in the inactive (Others) category. The new occupation became attractive, possibly due to its flexibility, making it easier for the unemployed to find employment despite the unusual working hours and lack of formality.
For the occupied, we can restrict the analysis to examine their occupational positions more closely. The self-employed category still has a high proportion, suggesting that informality may facilitate this transition. The informally employed category also shows slight growth, while those employed with a formal contract remain stable initially but decrease in 2019 and 2020 (see Supplementary materials section).
Regarding hourly income, the income per hour differential shows a stability with a slight increase over the quarters before the entry milestone and, after the apps’ entry, proceeds to fall, reaching the negative value of around −2.5 at the end of the series, indicating that, of those occupied in the previous year, all those who transitioned to the self-employed driver occupation experiencing a decline in income. Although quarters preceding the apps’ entry have a slight positive value, the income differential stays in negative values all the way further (see Supplementary materials). Expanding the analysis to different occupational positions, Figure 2 below disaggregates the income differential for formal and self-employed occupations.
Income differential by formal and self-employed occupations. Source: Authors’ own elaboration with data from PNAD-C/IBGE
Income differential by formal and self-employed occupations. Source: Authors’ own elaboration with data from PNAD-C/IBGE
Since the sample size is significantly reduced, the analysis is less robust, as indicated by the erratic movements of both segments. The other occupational positions were omitted because their values were so volatile that they compromised the analysis. The figure shows that the decline was mainly due to individuals who were previously in formal occupations (blue line). While the self-employed (yellow line) did not show variations over the years, those in formal employment experienced a slight increase in income differential immediately after the apps’ entry, possibly capturing a period when the self-employed driver occupation was so competitive that transitioning to the self-employed driver occupation led to income gains. However, after 2017, there is a progressive decline in the income differential.
Annualizing the values allows for the construction of income measures representing the average hourly income at the beginning and end of these transitions, making them less volatile (see Supplementary materials). Although all transitions lead to becoming a self-employed driver, the hourly income obtained as a driver varies depending on the initial occupational position, particularly when comparing those from formal occupations to those from informal occupations or self-employment. Formerly, formal workers had a higher income level as self-employed drivers compared to others. This difference may stem from the ride categories. Formally employed individuals tend to have higher income levels than those in other occupations, allowing them to own better cars, which are more likely to qualify for more exclusive ride categories (such as the comfort and Black categories in the case of Uber). Through these rides, especially during the initial years of app entry when the popular category was less widespread, this group could achieve a higher hourly income.
Unlike formal and informal workers, the self-employed consistently experienced a negative income differential in all the years following the app’s entry. While formal and informal workers mostly had negative differentials, they experienced positive average income transitions in 2016 and 2017.
Interestingly, the individuals who remain self-employed drivers the following year show positive increases in hourly income until the pandemic years, when a decrease occurs. However, there is a small drop in hourly income in the first year post-entry. This small drop may be attributed to the influx of new self-employed drivers into the occupation, suggesting that those already in the occupation previously had higher hourly income.
4.3 Impact evaluation of ride-sharing apps’ entry
In this section, we present the results of the DiD model regarding five variables: unemployment rate, number of occupied, labor income, working hours and income per hour. One important note regarding the unemployment rate is that it is a city-level aggregate rather than an individual-level measure; therefore, once we examine this metric, our “observation unit” effectively becomes the city itself instead of individual survey respondents. In other words, while the other four variables are estimated at the person level, the city-level unemployment rate necessitates a separate approach and will thus be shown and discussed independently. The results for the other four variables will be presented for all workers, for the self-employed and for self-employed drivers in Brazilian capital cities.
Estimating the results for the unemployment rate, we observe a complete lack of statistical significance (see Supplementary materials), indicating that, unlike Omberg (2024), there is no evidence that the introduction of ride-sharing apps has impacted this variable. Because of PNAD-C’s sample limitations, this lack of significance may stem from the small sample size – only 27 cities – whereas Omberg (2024) analyzed 389 metropolitan areas.
Aggregating the ATT dynamically we can measure it by each passing quarter prior and after the intervention. There is a complete absence of effect of apps’ entry on the unemployment rate. Although there is a decline followed by growth after the intervention, the broad confidence interval indicates that any observed changes in each quarter lack statistical significance (see Supplementary materials).
The lack of effect, aside from potentially arising from the limited sample size in our analysis, also highlights the possibility that this new form of work may impact high-informality-rate countries differently than those with lower informality rates. In highly informal settings, individuals may shift from other informal types of occupation (e.g. self-employed in sales or other service) into ride-sharing, producing mainly internal reallocation and only marginal reductions in overall unemployment. Essentially, app-based driving replaces one precarious role with another – characterized by greater flexibility in working hours and somewhat more predictable earnings, rather than generating a net job increase.
The Table 2 below provides a summary of the average treatment on treated effect (ATT) for the other set of variables.
Overall summary of ATT’s (“simple” aggregation)
| Labor market | Self-employed | SE – private driver | |
|---|---|---|---|
| Number of occupied | |||
| ATT | 9,764,048 | 13,099 | 2,475 |
| Std. Error | 303,476 | 34,259 | 3,289 |
| 95% Conf. Int | [(−585,038) – (604,566)] | [(−54,047) – (80,245)] | [(−3,971) – (8,921)]* |
| Income | |||
| ATT | −59.37 | 20.76 | −308.15 |
| Std. Error | 52.43 | 119.11 | 31.84 |
| 95% Conf. Int | [(−162.14) – (43.40)] | [(−212.70) – (254.22)] | [(−370.57) – (−245.73)]* |
| Worked hours | |||
| ATT | −57.23 | −10.23 | −1.54 |
| Std, Error | 48.67 | 72.13 | 0.22 |
| 95% Conf, Int | [(−154.7) – (36)] | [(−151.6) – (131.1)] | [(−1.97) – (−1.10)]* |
| Log hourly income | |||
| ATT | 0.0252 | 0.0495 | 0.0172 |
| Std, Error | 0.0104 | 0.0216 | 0.0522 |
| 95% Conf, Int | [(0.0048) – (0.0456)]* | [(0.0072) – (0.0919)]* | [(−0.0851) – (0.1195)] |
Note(s): The simple aggregation represents a weighted average of all group-time average treatment effects with weights proportional to group size
∗p < 0.1; ∗∗p < 0.05 and ∗∗∗p < 0.01. Signif. codes: “*” confidence band does not cover 0
Control Group: Not Yet Treated. Anticipation Periods: 0. Estimation Method: Doubly Robust
Source(s): Authors’ own elaboration
The difference-in-difference regression analysis was conducted on three samples (all workers, the self-employed and self-employed drivers), but the most compelling insights arise from the self-employed drivers’ sample. Analyzing the labor market as a whole is challenging due to the wide range of jobs and employment types, making less accurate before-and-after comparisons. Similarly, analyzing the entire self-employed segment is complex, as it encompasses a variety of occupations with different dynamics, potentially obscuring the specific effects of ridesharing apps.
The self-employed drivers sample provides the most interesting perspective, as the comparison is more direct and more explicit: it contrasts the scenario of private drivers before ridesharing apps (when the occupation consisted solely of traditional private drivers) with the scenario after their entry (when the occupation includes both traditional and self-employed drivers). This targeted comparison allows us to isolate the impact of ridesharing apps on employment and income. Figure 3 shows the ATT for the number of self-employed drivers relative to platform entry. Before entry, effects stay near zero and are not significant, so there is no sign of anticipation. After entry, the estimates rise, becoming positive and statistically significant in later quarters. This points to a delayed but substantial expansion of self-employed drivers attributable to ride-sharing platforms.
Average effect of apps’ entry on the number of private driver occupied (self-employed driver). Source: Authors’ own elaboration with data from PNAD-C/IBGE
Average effect of apps’ entry on the number of private driver occupied (self-employed driver). Source: Authors’ own elaboration with data from PNAD-C/IBGE
ATT estimates show a brief, statistically significant income uptick for early adopters, followed by a persistent and significant decline for self-employed drivers after platform entry. Pre-treatment trends are not flat, which suggests possible violations of the parallel-trends assumption and incomplete covariate balance, so any causal interpretation should be cautious (Figure 4).
Average effect of ride-sharing apps’ entry on income (self-employed drivers). Source: Authors’ own elaboration with data from PNAD-C/IBGE
Average effect of ride-sharing apps’ entry on income (self-employed drivers). Source: Authors’ own elaboration with data from PNAD-C/IBGE
Regarding the ATT worked hours estimates, the violation of the parallel trends assumption is a serious concern. However, after the entry, predominantly negative effects are observed, potentially indicating that the apps’ entry led to a reduction in working hours. This result may be due to some large variance between the groups (see Supplementary materials).
Disaggregating by city and grouping according to the quarter in which Uber entered each market, we reveal underlying dynamics in working hours that reflect changes in the opportunity cost faced by drivers over time (see Supplementary materials). Initially, the supply of drivers may have been limited, allowing those already on the platform to earn substantial compensation. This likely incentivized many to adopt the app as their primary source of income. However, as the driver supply increased, the earnings from the app may have diminished, leading to a subsequent reduction in average working hours.
Finally, we detail the results for the income per hour (available in the Supplementary materials). While the parallel trends assumption is not a threat to the analysis, the hourly income does not show any sign of variation, indicating that it is not possible to affirm that the introduction of the ridesharing apps significantly impacted the hourly wages of self-employed drivers.
In conclusion, the estimations cannot be considered definitive evidence of causal impacts. However, the results provide interesting insights. Regarding the number of jobs, there was an increase, with no variation in the effect before the apps’ entry. Concomitantly, the average income progressively declines over time after the entry of Uber and 99.
Additionally, the analysis shows a reduction in working hours after the apps’ entry, although the coefficients prior to the treatment are volatile, potentially due to large variance between groups. The effect on working hours varied by group, with those adopting ride-sharing occupations earlier experiencing a positive effect, while those adopting it later experienced a negative effect. Hourly income shows no significant impact.
This income dynamic may be related to the drivers’ opportunity cost. Initially, the supply of drivers was low, allowing drivers on the platform to earn substantial compensation, making the app their primary income source. However, as the supply of drivers grew, the compensation obtained from the app decreased, which may have discouraged longer working hours and led to a decrease in total income. This suggests that hourly income remained unchanged because drivers adjusted by working fewer hours.
5. Conclusion
This article examines the impact of the entry of ride-sharing applications on labor market outcomes. Employing a three-step empirical strategy – comprising descriptive statistics, transition analysis and causal impact evaluation – we seek to assess both the labor market dynamics associated with the expansion of these platforms and their direct causal effects. Our descriptive analysis indicates notable shifts in the composition of self-employed drivers, including an increased proportion of non-white workers and those with secondary education as well as a decline in average hourly earnings and social protection.
The transitions analysis yields several important insights. Most drivers were already employed before transitioning to app driving. Moreover, most of these drivers remain in their occupation over time, challenging the common perception of ride-hailing as a predominantly temporary or transitory form of employment. However, it could be that the one-year panel duration is not sufficient to capture longer-term occupational mobility. Simultaneously, we observe a slight increase in the number of unemployed individuals transitioning into self-employed drivers after the apps’ entry, indicating, to some extent, their role as a buffer for jobs and earnings.
In terms of income, hourly earnings exhibited slight increases prior to the entry of ride-sharing apps into the market. However, earnings declined following their entry, suggesting a general decrease in income for individuals transitioning to self-employed drivers. Formal employees transitioning to drivers initially experienced income gains, but these gains declined after 2017. In contrast, self-employed workers consistently faced negative income differentials. The analysis of annualized income transitions shows that formal employees who became self-employed drivers had higher income levels than informal or self-employed workers. This may be attributed to their greater capacity to acquire higher-quality vehicles eligible for premium ride categories, leading to higher earnings as self-employed drivers.
Our difference-in-differences estimates provide suggestive, not definitive, evidence because the parallel-trends assumption does not hold for every outcome. Contrary to Omberg (2024), Uber’s arrival in Brazilian capitals appears not to have lowered unemployment; in a highly informal economy the platform likely reallocates workers rather than creates jobs. Employment therefore rose, yet income climbed only briefly before declining, and average working hours fell. Coefficients before treatment were volatile, probably because of large variance between treated and control groups. The effect on hours was heterogeneous: early adopters worked more, whereas late adopters reduced their schedules. Early adopters benefited when driver supply was scarce, but as supply expanded, compensation dropped, prompting drivers to cut hours in order to preserve roughly stable hourly earnings.
Overall, the study offers a first look at the new occupation’s dynamics. While the apps create opportunities and short-run earnings gains for people in lower-quality jobs or unemployed, maintaining a large workforce in this role may be problematic. In the long run, drivers could struggle to return to formal employment, producing an undesirable equilibrium. A longer observation window is therefore essential. Future work might explore how platform heterogeneity influences earnings trajectories.
From a public-policy standpoint, an urgent priority is to push for greater platform transparency. Requiring ride-hailing companies to provide anonymized trip-level data and algorithm-pricing documentation to regulators and researchers will enable the monitoring of pay dispersion, discrimination and market power. Platform better transparency would be key for the drivers themselves, enabling them to negotiate fairer labor conditions and to reduce information asymmetries. In this respect, policies promoting drivers’ organization (through trade unions, associations or other structures) would be a good way to give them a voice. Additionally, the creation of a portable social-protection floor so that contributions to pensions and sickness and accident coverage travel with drivers across platforms and back into formal employment would offer gig workers a basic safety net, reduce long-run vulnerability and facilitate smoother transitions to formal jobs.
Finally, these findings underscore the regulatory challenge posed by platform-based work: how to preserve the freedom to choose when, where and how much to work – a key element of what makes ride-hailing attractive to workers – while addressing the precariousness that emerges over time, manifested in declining earnings, lack of social protection and limited longer-term mobility. Designing policies that balance these tensions is crucial not only to improve job quality in the gig economy but also to ensure that new forms of work do not deepen labor market segmentation in contexts of high informality.
The supplementary material for this article can be found online.
