The Perils of Laundering Control through Customers: A Study of Control and Resistance in the Ride-hail Industry

Abstract

Customer abuse of frontline service workers is widespread. Yet despite growing recognition of this problem, we know very little about the role that service companies play in potentially enabling customers’ abusive behaviors. This phenomenon deserves attention because one of the recent trends in service management is giving customers a direct role in managing and evaluating workers’ performance. In this article, the author explores how granting customers direct access to organizational power over workers, what the author develops conceptually as “laundering control through customers,” explains how customer abuse emerges. Drawing on a sample of 486 Uber and Lyft drivers, the author examines how the companies’ use of the “five-star” evaluation system enables customers to engage in a range of different forms of abuse and how workers resist this configuration of control. This study contributes to the customer triangle literature by bringing in evidence from the gig economy and uncovers new implications for the “dark side” of customer service work.

Keywords

conflict management dispute resolution employment relations digital technology labor and employment policies

Customer abuse of frontline service workers is increasingly recognized as a problem in the academic literature and the media. Research has found workers suffer a number of negative consequences from this abuse, including lower performance, emotional exhaustion, depression, and greater work–family conflict (Groth, Wu, Nguyen, and Johnson 2019). Curiously, however, this research has stopped short of examining how service organizations may enable the abuse of frontline service workers by elevating customers to a position of power over these workers. In this study, I contribute to the literature by exploring how empowering customers to evaluate and monitor frontline service workers contributes to the abuse of these workers.

I refer to this process in which customers are used as an additional layer of managerial control as “laundering control” through customers. This type of control moves beyond saying “the customer is always right” to granting customers direct access to organizational power, such as directing, monitoring, and evaluating workers. In this article, I examine two research questions about this method of control. First, how does laundering control influence the types of abuse that workers face? Second, how do workers respond to this system of control? Platform companies, such as Uber and Lyft, provide a useful empirical site to examine this style of control because these companies rely almost exclusively on customer evaluations for the purposes of performance management and monitoring. Although almost all previous research on customer abuse has focused on single-company examples, a one-company case is insufficient to fully understand the impact of customer abuse in the platform setting because workers can easily move between companies, decreasing any single company’s control over these workers and increasing their options for resistance.

Using qualitative and quantitative data from Uber and Lyft drivers who are located across the United States, I explore how laundering control sets the foundation for customer abuse to occur and the novel methods drivers use to punish these organizations for orchestrating a system that enables their abuse. In doing so, this article reveals that customer abuse is not merely about proximate causes—the irritated or unsatisfied customer—but that organizations enable these behaviors by granting customers power over workers. This article concludes with policy recommendations for how platform companies can be integrated into existing labor and employment laws.

Conceptual Framework

Customer Abuse and the Dark Side of Service Work

It is common to hear service organizations say “the customer is always right” or the “customer is king.” These phrases capture service companies’ goal of differentiating themselves by competing on the basis of service quality to enhance profitability and shareholder value (Anderson and Fornell 2000). Frontline workers play an important role in customer retention because these workers act as boundary-spanners who represent the organization and provide services to the customers (Leidner 1993; Batt 1999). To curate and control the customer–worker interaction, service companies prescribe “service rules,” or scripts of mandatory behavior, that require frontline service workers to maintain their roles (usually friendly and deferential) regardless of customers’ behavior (Solomon, Surprenant, Czepiel, and Gutman 1985; Leidner 1993; Wang, Liao, Zhan, and Shi 2011).

Service companies engage in performance management and monitoring to ensure workers adhere to these service rules. Traditional performance monitoring relies on managers to directly observe the customer–worker interaction (Holman, Chissick, and Totterdell 2002; Lee, Batt, and Moynihan 2019). While many companies use this method, it is only episodic and can suffer from line-of-sight limitations. More recently, service companies have attempted to extend the organizational line of sight by delegating some performance monitoring authority to customers in the form of customer feedback surveys (Groth 2005). This method of monitoring creates the threat that a customer may report deviations from the prescribed service rules to management. Alternatively, in electronically mediated service arrangements, such as call centers, service companies can use electronic monitoring systems that track both the technical elements of service work (i.e., call length, information provided) and the social elements of the exchange (i.e., workers’ friendliness during a call) (Holman et al. 2002; Carroll 2008; Anteby and Chan 2018; Lee et al. 2019). In contrast to direct observational monitoring, electronic systems provide continuous monitoring capabilities and can store call logs for future managerial review (Rosenthal 2004; Aleks, Maffie, and Saksida 2020).

Embedded within these forms of monitoring is the concept that customers should be placed in a position of power to penalize workers for poor (or undesirable) performance (Hochschild 1983; Allan and Gilbert 2002; Groth et al. 2019). Yet customers are not always benevolent with this power. In fact, frontline service workers experience rampant customer abuse (Harris and Daunt 2011), with some call center employees experiencing verbal abuse more than 10 times a day (Grandey, Dickter, and Sin 2004). Existing scholarship has documented a range of types of customer abuse, including demeaning, disrespectful, uncivil, and even aggressive behaviors (Caruana, Ramaseshan, and Ewing 2001; Harris and Daunt 2003). Although most of these types of abuse are in response to customers’ demands for individual treatment or service, customers can also engage in abuse out of thoughtlessness or nefarious intent (Lovelock 1994; Harris and Daunt 2004).

With growing recognition of this problem, researchers have started exploring how customer abuse affects workers (Koopmann, Wang, Liu, and Song 2015), including how customer abuse can be emotionally or personally taxing (Brotheridge and Grandey 2002; Brotheridge and Lee 2002; Dormann and Zapf 2004; Wang et al. 2011). In this research, customers who engage in abusive behaviors require more attention and energy, resulting in employees feeling depleted and less able to perform their work (Rupp and Spencer 2006; Spencer and Rupp 2009). Using this theory, scholars have linked customer abuse to lowered performance, emotional burnout, and incivility toward customers (Dormann and Zapf 2004; Sliter, Jex, Wolford, and McInnerney 2010; Van Jaarsveld, Walker, and Skarlicki 2010; Shao and Skarlicki 2014; Sliter and Jones 2015). Furthermore, some scholars have extended this literature to examine how emotional exhaustion affects events outside of work, such as work–family conflict (Greenbaum et al. 2014; Chi, Yang, and Lin 2016).

Alternatively, some scholars take a justice-based approach to explain how and when frontline service workers engage in acts of retaliation or sabotage against customers (Skarlicki and Folger 1997; Colquitt et al. 2001; Groth et al. 2019). In these works, customer abuse amounts to a violation of interactional or informational justice expectations, resulting in workers engaging in retaliation to get back at customers (Van Maanen 1999; Rupp and Spencer 2006; Spencer and Rupp 2009; Van Jaarsveld et al. 2010). Both of these approaches, however, focus on how employees experience and react to acts of abuse and only implicitly acknowledge the role of the employer in creating an asymmetric power relationship between customers and workers.

Power and Customer Abuse on Digital Platforms

In stark contrast to saying “the customer is king,” it is common to hear platform companies, such as Uber and Lyft, say that workers “are their own boss” (Rosenblat 2018). Platform companies use this phrase to distinguish their services from ones that require workers to follow a set of service rules when interacting with customers. The absence of service rules is strategically advantageous for these companies because it (to date) helps them to avoid providing workers with a number of statutory labor and employment rights, such as unemployment compensation, overtime, or the right to engage in protected collective activity (Prassl 2018; DeVault, Figueroa, Kotler, and Maffie 2019). Similar to more traditional service companies, however, platform companies are dependent on developing a large customer base in order to create a defensible market position and eventually reach profitability (Rysman 2009). These two factors place platform companies at an impasse: These companies are reliant on workers to provide high-quality service experiences to customers, but in order to classify workers as “independent contractors,” they also must refrain from directing workers how to do so.

Ride-hail companies resolve this dilemma by delegating performance management and monitoring authority to customers (Schor 2014; Rosenblat and Stark 2016; Ticona, Mateescu, and Rosenblat 2018). The ubiquitous method for doing so is a five-star rating scale, for which customers review workers’ performance after each transaction. Three facets of the five-star system develop a structural power asymmetry between drivers and customers. First, drivers must maintain a very high average performance rating, roughly 4.7 out of 5 stars, de facto mandating that drivers receive perfect performance reviews from customers (Lee, Kusbit, Metsky, and Dabbish 2015; Rosenblat and Stark 2016; DeVault et al. 2019). Second, negative customer reviews, such as a one-star rating (−3.7 below the ratings cutoff), disproportionally affect drivers’ average performance ratings compared to a perfect five-star rating (only +0.3 above the performance cutoff). Finally, these companies do not specify their minimum performance threshold, leaving drivers uncertain how close they are to termination (Rosenblat and Stark 2016). Combined, these three factors effectively require drivers to follow the directive that “the customer is always right” without ever being told to do so (Rosenblat 2018).

The five-star system represents an evolution in the way that service companies exercise control over workers by laundering control through customers. Laundering control embeds customers as an additional layer of managerial control by empowering customers to direct, monitor, and/or evaluate workers. In contrast to instructing workers to act as if “the customer is king,” laundering control imbues customers with organizational power to directly manage workers. This method of control presents a conceptual departure from traditional modes of managing frontline workers because it grants customers the power to both set their own service rules and evaluate workers’ adherence to those rules. Such a system allows platform companies to achieve their desired goal of ensuring that workers are attentive to customers while simultaneously allowing them to deny their role in managing the service exchange.

Laundering control presents a powerful control system because it allows companies to create directives through the customer. In ride-hail, for almost every driver “freedom,” companies emphasize a countervailing customer value or expectation (Rosenblat and Stark 2016; Rosenblat 2018). For example, companies say that drivers can select which rides to work, but Uber conditions driver bonuses on a high acceptance rate because customers want “reliable service” (previously, Uber would discipline or deactivate drivers who had low acceptance rates) (Uber 2019). Furthermore, companies tell drivers that top-rated drivers offer mints and water to customers, customers prefer professionally dressed drivers, and “going above and beyond” (such as opening a passenger’s door) is more likely to result in a five-star rating (Rosenblat and Stark 2016; Rosenblat, Levy, Barocas, and Hwang 2017). While this information could be seen as benevolent if offered in isolation, pairing these messages with a system that disproportionally punishes anything short of perfect service allows platform companies to launder these directives through the customer. These two systems work in tandem; the five-star system creates very little room for error, allowing companies to create implicit directives by suggestion.

Similar to more traditional methods of managing frontline service workers, laundering control is predicated on granting customers power over workers. Yet it is unknown how customers behave when they are given direct access to organizational power. As the research on frontline service workers turns to the “dark side” of service interactions (Skarlicki, Van Jaarsveld, and Walker 2008: 1335), this study examines two questions about laundering control. First, how do customers abuse their laundered control? Second, how do workers resist this configuration of power?

Study 1: Conflict and Power in the Ride-hail Industry

Very little is known about the process or content of the driver−passenger interaction in the ride-hail industry. In this study, I interviewed drivers to explore the relationship between laundering control and worker abuse. Based on these data, I developed patterns to describe how worker abuse manifests itself in ride-hail and how drivers fight back against this method of control.

Qualitative Research Strategy and Methods

Between fall 2016 and spring 2017, I engaged in field work for which I attended driver meetings in a large metropolitan area and conducted semi-structured interviews with 55 drivers located across the United States. Prior to entering the field, I developed my initial set of interview questions by listening to 30 driver interviews from online resources. I was able to obtain transcripts of 12 of these interviews and thematically code them for the elements of drivers’ experiences, including drivers’ views of Uber compared to Lyft, emotional labor, relationship to ride-hail companies, relationships with other drivers, and the five-star rating system. Examples of these questions are included in Appendix A.

Participant Recruitment

I recruited participants for this study in three ways. First, from a small initial list of drivers I used snowball sampling to obtain the names of additional drivers, and from this method was able to interview 21 ride-hail drivers. These drivers were located across the United States and possessed a range of experiences—some having worked fewer than 100 rides to others with more than 10,000 rides. Second, I interviewed 26 drivers outside of Uber’s offices in New York City. Finally, I recruited 8 participants from closed driver Facebook groups. Drivers recruited via Facebook operated in smaller towns and cities. In total, I interviewed 55 ride-hail drivers about their experiences working for Uber and Lyft. All interviews occurred in English, and only one participant was excluded due to their method of transportation (a bike).

Data Analysis

Data analysis proceeded in multiple steps. First, I iteratively coded the data after each interview (Strauss and Corbin 2008) for events that drivers identified as passenger conflicts. Next, I grouped these events into five categories: unruly passenger behaviors, filing a cleanup fee, squeezing too many people into a vehicle beyond the legal limit, incorrect compensation, and if drivers filed a formal complaint with their platform over passenger behavior (which was a common reaction). During this step, I eliminated outlier events that only one or two drivers identified as a conflict, such as health and wellness concerns from driving. After developing the core conflict categories, I returned to the data to examine their surrounding context. During this stage, one of the most intriguing elements of the study emerged: Drivers blamed ride-hail companies for enabling customers’ abusive behaviors because of the power imbalance created by the five-star system. Drivers who had previous transportation experience provided the clearest evidence here, with one former taxi driver stating that Uber and Lyft’s deference to customers over drivers led him to begin carrying a gun with him during rides.

Upon surfacing the central conflict categories, I delved into the platform governance literature to explore how platform designers structure relationships between users. This research yielded several insights regarding how digital intermediaries shape the types of interactions that occur within a platform (Tiwana, Konsynski, and Bush 2010; Winter, Berente, Howison, and Butler 2014; Song, Xue, Rai, and Zhang 2018). This literature proved useful because it frames platform design as an act of power: Intermediaries structure relationships in a fashion that either marginalizes or empowers people within their spaces (Deng, Joshi, and Galliers 2016). This body of work provided an important addition to the study because it brought to light how platform designers can strategically use power imbalances to further their own goals.

Study 1. Results and Discussion

Customer abuse under a laundered control system can be understood as a three-part process: scope, blame, and retaliation. In terms of scope, I found that the five-star system creates an environment in which driver abuse is an ever-present part of the industry and encompasses both traditional abuse (e.g., verbal) and a new category of abuse in which workers are placed in legal jeopardy. Examples of this abuse include passengers “squeezing” too many people into a vehicle (to avoid paying for a larger, more expensive vehicle) or sneaking an open container of alcohol into a vehicle:

I don’t want to say no [to passengers trying to pack too many people into a car] because I know that that’s going to lead to less than five stars. [Interview 015, M]

For drivers who work at the most lucrative times, nights and weekends, this form of abuse is a frequent occurrence:

It’s [squeezing] going to happen whenever I’m working a bar crowd. If you’re going to work the night, it’s going to happen, once or twice on average. It’s going to happen. Most of it is that the XL’s, I take Uber a lot—for example, I’m going to take Uber tonight because we have five people—it’s a little pricy. It’s almost cheaper to call two Ubers than to call an XL, which makes no sense in my mind. [Interview 017, M]

Don’t do Friday nights. What are you going to do, have a logical discussion with them [drunk passengers]? If someone thinks it’s ok to drink beer in a car, how do you convince them it’s not? They will just give you a bad rating. What do you say to someone who wants to get hammered in your car? They know it’s illegal. You know? [Interview 014, M]

Next, recognizing that the five-star system forced them into a submissive position vis-à-vis the customer, drivers blamed ride-hail companies as contributing transgressors in passengers’ abusive behaviors:

My biggest issue, and you have probably seen this if you are on some of the boards, is that it’s [the rating system] created—Uber in the middle—has created a passenger vs. driver situation. . . . They will rate you poorly and email Uber even if you did a fine job to try to get a free ride because Uber acquiesces a lot and just gives them a free ride if they claim they’ve had this bad experience. . . . I feel like it’s created a situation where passengers and drivers are at each other when the real problem is Uber. [Interview 026, F]

When I was driving a taxi, it was more fun because it wasn’t my car . . . with a cab, if someone is too drunk you can yell at them, you can curse at them, there were no repercussions. With Uber you have to pretty much brown-nose to everybody and kiss everyone’s ass because you are concerned about the rating system. [Interview 021, M]

Uber’s biggest mistake is that they don’t balance customer and drivers, they always take the customer’s side. We are also paying them for the rides. It needs to be balanced. [Interview 041, M]

Finally, without recourse against passengers, drivers retaliated against ride-hail companies by moving passengers onto services that provided them with greater protections. In this stage, design differences across Uber and Lyft became important: at the time, Lyft empowered workers to block future pairings with abusive passengers while granting workers greater leniency in its five-star system. As a result, Lyft’s power imbalance between workers and passengers was lower than Uber’s, leading drivers to poach passengers for Lyft. Doing so allowed drivers to move passengers away from platforms with harsher performance management systems while not directly harming their ability to earn future income.

I prefer Lyft much more than Uber. . . . One, in general, Lyft tends to treat their drivers better. The feeling I have, and I’ll give you examples, but the general feeling I have is Uber is all about bait-and-switch from the driver’s perspective. Whereas Lyft, what they say is what you might expect. . . . I’ll encourage Uber passengers to download the [Lyft] app. I’ll convert Uber passengers. And I’ve been called by Uber. They called me out, “we hear you are passing out coupon codes, and we’d ask you not to do that.” I said, “Fair enough.” Hung up, stopped passing out coupon codes for a couple months, started back up again passing out Lyft coupon codes to Uber passengers. [Interview 025, M]

There’s a new company starting in New York called Juno. . . . [A]pparently they are addressing all of these negative points that drivers have about Uber and Lyft. . . . [I]f Juno comes down here, I’ll probably start handing out their cards to both Uber and Lyft passengers. [Interview 030, M]

The rating system for passengers is a joke. They [Uber] don’t care. They don’t give a shit. Passengers can have a terrible rating and I find that such a double standard. . . . [D]rivers rating passengers was just a way of making drivers feel better . . . it’s one of the reasons why I know drivers that prefer Lyft: If either the driver or the passenger rate each other at 3 or below, you will never be matched up with that driver or passenger again. And I love it. Every driver I know loves it. [Interview 016, M]

This three-part process identifies how power and control operate when customers are inserted as an additional layer of managerial control. Laundering control through customers creates an open-ended control system in which customers are empowered to establish and enforce their preferred set of service rules. This study illustrates that such a system results not only in documented forms of worker abuse, such as verbal abuse, but can also place workers in legal jeopardy. Realizing that ride-hail companies developed this power imbalance through the five-star system, workers blamed these companies as enabling customers’ abusive acts and retaliated against them by poaching customers and spending more time on services that provided them with superior working conditions. At the time of this study, this resulted in drivers recruiting passengers for Lyft because Lyft developed a slightly lower power asymmetry between drivers and passengers.

This three-part process can be rewritten as three hypotheses regarding the relationship between laundering control, customer abuse, and drivers’ retaliation against ride-hail companies:

H1: Laundering control results in workers blaming companies for customer abuse.

H2: Platform workers will move customers to service companies with less coercive forms of laundering control.

H3: Platform workers will reward companies with less coercive laundering control systems by spending additional time on their services.

Study 2: Testing the Relationship between Laundering Control and Worker Resistance

Study 2 empirically tests the hypotheses derived from Study 1 by gathering survey data on the frequency of driver abuse, drivers’ use of Uber and Lyft, and drivers’ passenger recruitment behaviors. Using these data, I test whether drivers blame platform companies for laundering control through passengers and whether drivers retaliate against companies that promulgate more coercive customer control systems.

Quantitative Research Strategy and Methods

Using the data from my qualitative study, I developed or repurposed existing scales to measure the incidents of driver abuse, drivers’ views on Uber and Lyft, and drivers’ time allocation across these platforms. As drivers’ ride-hail experiences will vary by geography, I sampled drivers in both rural and metropolitan areas. I administered the survey at two points in time to test if passenger abuse in one period of time is associated with drivers’ subsequent behaviors.

Choice of Companies

This study focuses on Uber and Lyft because my qualitative study provided the clearest evidence that workers vary their effort and recruitment behaviors across these two services. Focusing on these companies also presents two additional methodological benefits. First, they are the largest transportation network companies in terms of drivers and customers, which allows me to draw from the largest amount of comparative data across platforms. Second, these companies offer nearly identical rates in all of the markets where they compete head-to-head. A comparison of their rates in 50 large cities across the United States reveals only small differences in compensation per mile or per minute. This similarity in compensation means that models do not need to account for price per mile or per minute when measuring how workers allocate their effort across Uber and Lyft.

Sample Selection

I recruited 486 ride-hail drivers for study 2 in three ways. First, a worker organization from a Northeast metropolitan area notified their membership about the study, leading to 226 participants. Second, Harry Campbell, “The Rideshare Guy,” posted a call for participants on his website in summer 2017,¹ leading to 234 study participants. Finally, I engaged in targeted Facebook recruitment in an attempt to recruit drivers who had only recently started working ride-hail. Previous research shows that Facebook can be appropriate for statistical inference (Fenner et al. 2012; Ramo and Prochaska 2012). I recruited 26 participants from driver Facebook groups that require drivers to send moderators proof of driver status as a condition of membership. In total, 490 participants started the first questionnaire, with 486 providing useable data. A total of 359 drivers completed the follow-up questionnaire (73% retention rate).

Comparison of Sample Demographics to Prior Surveys

Table 1 displays the demographic characteristics of participants in this study. Two previous surveys help contextualize how participants in this study compare to the broader ride-hail industry. First, in 2014, Uber hired Benenson Strategy Group (BSG) to conduct demographic research on its drivers. Second, Harry Campbell conducted an online demographic survey from late 2016 to early 2017. BSG surveyed only Uber drivers whereas Campbell included drivers who worked for both Uber and Lyft. Consistent with these past surveys, roughly one-third of this study’s participants identified as full-time drivers and were not looking for other work. My survey includes a lower proportion of part-time drivers because it contained two additional driver status categories (transitional, other) that neither BSG nor Campbell included in their surveys. I included these categories because my qualitative study found that some drivers work ride-hail for reasons that are not captured by the full- versus part-time dichotomy.

Table 1.

Comparison of Author Sample to Previous Ride-hail Surveys (by Percent)

Variable name	2014 BSG survey (%)	Campbell, 2017 (%)	Author, 2018 (%)
Age 18–29	19.1	6.5	26.4
30–39	30.1	16	23.1
40–49	26.3	23.5	34.6
50–64	21.8	29.7	14.1
65+	2.7	24.3	1.6
Female	13.8	19	17.6
Less than HS	3	1.7	1.8
High school	9.2	8.5	9.8
Some college / associate’s	40	23.6	29
College degree	36.9	33.8	40.3
Postgraduate degree	10.8	16.4	18.9
White non-Hispanic	40.3	78.3	60.2
Black non-Hispanic	19.5	6.8	10.7
Asian non-Hispanic	16.5	4	12.5
Other non-Hispanic	5.9	3.5	5.3
Hispanic	17.7	7.1	10.4
Number of months driving (mean)	—	—	21.4
Vehicle	—	—
Small sedan	—	—	15.9
Medium sedan	—	—	35.6
Large sedan	—	—	11.2
Four-wheel drive SUV	—	—	22.7
Minivan			5.7
Other			9.7
Ride-hail insurance	—	—
Yes	—	—	38.4
No	—	—	59.6
Used to	—	—	1.8
Full time or Part time	—	—
Full time	38	—	38.9
Part time	61	—	30.3
Transitional work	—	—	17.4
Other	—	—	13.3
Transportation experience [Yes]	20	—	32.7
Number of drivers	601	1,150	486

BSG, Benenson Strategy Group.

Participants in this study have similar demographics to these past surveys, with the only areas of large deviations being the proportion of non-Hispanic white drivers and those over the age of 65. The BGS study was conducted in 2014; since then Uber and Lyft have started operating in nearly 100 new areas, resulting in tens of thousands of new drivers working for these services. Given the additional geographic areas in which these companies now operate, it is not clear how these areas have shaped the demographics of the industry. Yet the consistency across studies and the demographic makeup of this study suggest the models in this article are reasonably representative of the ride-hail industry.

Survey Design and Administration

I administered the survey from an online Qualtrics server. This method does not risk systematically excluding members of the population because ride-hail drivers are required to have a smartphone to work. Comparisons between telephone- and smartphone-administered surveys have found smartphone data collection provides virtually identical information to traditional phone-administered studies (Johansen and Wedderkopp 2010). Broadly, this data collection method has been used in a variety of research settings for decades (Shiffman, Stone, and Hufford 2008).

For this study, drivers were sent two surveys. In the first time period, June 2017, the survey collected data on drivers’ demographics, time on platform, passenger conflict events, and organizational relationship measures. I sent the second survey to drivers 30 days after they completed the first survey in order to avoid sample atrophy caused by drivers exiting the industry (Farrell, Greig, and Hamoudi 2018). Notably, the second round of data collection occurred during the first month of Uber’s 180 Days of Change program, a program that substantially redesigned the Uber service, but all data were collected before Uber began penalizing customers for receiving low ratings from drivers.

Scale Development and Variable Definitions

I developed original scales based on my qualitative research to measure constructs that are not found in the existing literature and adapted items from the literature to fit the specific conditions of the ride-hail industry. These constructs include a passenger–driver conflict scale, a driver−platform relationship measure, and drivers’ time allocation across Uber and Lyft.

Passenger Conflict Scale. I used the five conflict events identified in my qualitative study as scale items for the survey. Drivers responded with the frequency of these events, ranging from “Every time I drive” (6) to “Never” (1). I then analyzed these data using factor analysis. Research suggests that factor analysis should be conducted using both statistical and interpretative criteria (Guidroz et al. 2010). The five conflict items loaded on a single construct (DeVellis 2012), but conceptually, I removed one item, incorrect compensation, because it applied only to the relationship between drivers and platform companies. The remaining four items loaded above Tabachnick and Fiddell’s (2001) minimum item loading criteria (0.32). The final scale (see Appendix B) returned a Cronbach’s alpha of 0.75.

Organizational Loyalty Scales. I used Van Dyne, Graham, and Dinesch’s (1994) organizational loyalty scale to estimate the driver−platform relationship because the scale contains items that can reasonably fit the context of the ride-hail industry, such as if workers believe the company is a good place to work or represent the company favorably to others (the full list of items can be found in Appendix C). The resulting scales returned an acceptable Cronbach’s alpha for Uber (0.79) and Lyft (0.79). I differenced drivers’ Uber and Lyft organizational loyalty scores, calculated as Uber loyalty − Lyft loyalty, to create an additional variable that measures drivers’ relative loyalty across these organizations. The mean value (−0.62) suggests that drivers in this sample have a preference for Lyft over Uber.

Platform Recommendation. This item measures whether drivers would recommend working for a ride-hail platform to a friend. The scale ranged from 5 (“Highly Recommend”) to 1 (“Strongly Would Not Recommend”). Differencing drivers’ Uber and Lyft scores creates a comparison between these two companies. The variable ranged from −4 (would strongly recommend Lyft over Uber) to +4 (would strongly recommend Uber over Lyft). This item’s mean score, −0.64, suggests that the average driver would recommend a friend work for Lyft over Uber.

Time on Platform. This item measures the number of hours drivers worked on Uber and Lyft during the previous week. On average, drivers spent more time on Uber (time 1 = 25.84, time 2 = 23.28) than Lyft (time 1 = 18.04, time 2 = 15.44). In the second time period, 63 drivers worked zero hours, of which 49 indicated that they planned to continue working ride-hail in the future. Accordingly, I coded these 49 drivers as “0” hours worked in time period 2 and the 14 drivers who had left the industry as “NA.” Coding these 63 drivers as either “NA” or “0” does not meaningfully change the results of this study.

Fraction of Time on Uber versus Lyft. For the 317 drivers who provided working time for both Uber and Lyft, I divided their hours on Uber by their combined hours on Uber and Lyft. As the value approaches 1, it suggests that a driver is spending most of her or his time working on Uber relative to Lyft. In both time period 1 (mean = 0.57, SD = 0.21) and time period 2 (mean = 0.61, SD = 0.28), drivers, on average, spent more time on Uber than Lyft.

Recruitment Difference. On a five-point Likert scale from “Strongly Disagree” to “Strongly Agree,” drivers reported how likely they were to recruit passengers and drivers to Uber and Lyft. I differenced these two questions (Uber − Lyft recruitment) to measure how likely drivers were to recruit for Uber relative to Lyft. The mean reported in Table 2 suggests that, on average, drivers were more likely to recruit both drivers (mean = −0.70, SD = 1.46) and passengers (mean = −0.86, SD = 1.56) to Lyft over Uber.

Table 2.

Description of Continuous Variables

Variable name	Mean	SD	N
Uber loyalty scale (1−5)	2.56	0.81	430
Lyft loyalty scale (1−5)	3.14	0.75	367
Loyalty scale difference, Uber − Lyft	−0.62	0.98	306
Recommendation difference, Uber − Lyft	−0.64	1.40	410
Passenger conflict scale (1−6)	2.54	0.97	486
Weekly hours worked on Uber (Time 1)	25.84	20.56	436
Weekly hours worked on Uber (Time 2)	23.28	17.87	257
Weekly hours worked on Lyft (Time 1)	18.04	17.44	367
Weekly hours worked on Lyft (Time 2)	15.44	15.05	218
Fraction of hours worked on Uber, Time 1	0.57	0.21	317
Fraction of hours worked on Uber, Time 2	0.61	0.28	176
Driver recruitment, Uber − Lyft	−0.70	1.46	305
Passenger recruitment, Uber − Lyft	−0.86	1.56	304

Notes: The Uber and Lyft scales are 1 (lowest loyalty) to 5 (highest). The scale difference is the sum of these two scales and ranged from −4 (5 loyalty Lyft, 1 Uber) to +4 (5 loyalty Uber, 1 loyalty Lyft). The scale difference does not equal the difference between the individual scales because some drivers only worked for Uber or Lyft. SD, standard deviation.

Control Variables. This study’s demographic measurements mirrored Hall and Krueger (2017). Additional control variables include vehicle type, educational level, previous transportation industry experience, months driving ride-hail, and driver status (full-time, part-time, etc.). Full sample statistics can be found in Table 1.

Estimation Strategy

The structure of these data allows for several estimation methods. I used Tobit models to estimate the relationship between passenger conflict and organizational loyalty because the response variable is censored at 1. Testing this relationship using an ordinary least squares (OLS) model returns similar coefficients and strength of association between the variables of interest. I used OLS models to test the relationship between organizational loyalty and time on platform. As the dependent variable, time on platform, is a repeated measurement, I included random effects at the driver level (Hausman 1978). Random effects models also include a time dummy to absorb any exogenous variability associated with the measurement period. Finally, organizational loyalty and passenger recruitment rely on time period 1 data and therefore utilize OLS models. This relationship was also tested using Tobit models, but I did not find any differences in the models.

Quantitative Findings

Table 1 provides a proportional breakdown of the categorical variables while Table 2 provides the means and standard deviations for continuous variables. The models in Table 3 examine whether passenger conflict is associated with drivers’ loyalty toward their ride-hail companies. The independent variable of interest is drivers’ reported conflict frequency with customers. The dependent variable is their reported loyalty to Uber (model (1)) and Lyft (model (2)). These models suggest that each additional point on the passenger conflict frequency scale is associated with a 0.29 decrease in organizational loyalty to Uber (model (1)) or a 0.17 decrease in organizational loyalty to Lyft (model (2)). Both of these coefficients are statistically significant (p < 0.001).

Table 3.

Tobit Models Regressing Uber and Lyft Loyalty on Frequency of Passenger Conflict

Variable name	(1)Uber loyalty	(2)Lyft loyalty
Passenger conflict scale	−0.29***	−0.17***
	(0.04)	(0.04)
College [1 = Yes, 0 = No]	−0.16**	−0.10
	(0.08)	(0.07)
Female [1 = Yes, 0 = No]	0.01	0.38***
	(0.10)	(0.07)
Black non-Hispanic	−0.28**	0.21**
	(0.13)	(0.12)
Asian/API	−0.06	0.01
	(0.12)	(0.14)
Age	−0.12***	−0.03
	(0.04)	(0.03)
Transitional work	−0.19**	0.07
	(0.11)	(0.11)
Transportation experience [1 = Yes, 0 = No]	0.13	0.12
	(0.09)	(0.09)
Platform count	−0.06	−0.12***
	(0.04)	(0.04)
Number of observations	409	354
−Log-likelihood	−496.29	−365.82

Notes: The number of observations varies across regressions equations due to a different number of Lyft and Uber drivers in the sample. Some nonsignificant controls omitted for clarity. Standard errors listed below regression coefficients.

* = p < 0.10; **p < 0.05; ***p < 0.01.

The models in Table 4 examine whether drivers’ relationship with Uber or Lyft predicts their working time on these services. In these models, the dependent variable of interest is drivers’ reported time on Uber (models (1) and (2)) and Lyft (models (3) and (4)). Both Uber (model (1), p < 0.01) and Lyft (model (4), p < 0.015) loyalty scales are positively associated with workers’ total working time on the respective platform. A one unit increase in the Uber loyalty scale is associated with 3.96 more hours working on Uber and a one unit increase in the Lyft loyalty scale is associated with 3.17 more hours on Lyft. Models (2) and (3) in Table 4 report the relationship between the recommendation measure (Uber–Lyft) and drivers’ time on platform. These models suggest that a one unit increase in preference for Uber over Lyft is associated with working an additional 4.50 hours for Uber or 2.63 fewer hours for Lyft (p < 0.001). These results imply that drivers who would recommend Uber over Lyft are likely to spend more time working for Uber than those who recommend Lyft to Uber.

Table 4.

Linear OLS Models with Random Effects Regressing Time on Platform on Organizational Loyalty Measures

Variable name	(1)Uber time in hours	(2)Uber time in hours	(3)Lyft time in hours	(4)Lyft time in hours
Uber loyalty scale	3.96***	—	—	—
	(1.11)
Lyft loyalty scale	—	—	—	3.17**
				(1.24)
Recommendation difference	—	4.50***	−2.63***	−0.24
		(0.69)	(0.66)	(2.59)
Medium sedan	3.80	3.61	−0.23	−2.38
	(2.56)	(2.57)	(2.54)	(3.72)
Large sedan	4.42	3.90	−2.52	−2.00
	(3.31)	(3.33)	(3.53)	(2.95)
Four-wheel drive	4.72	5.32°	−1.40	−8.40**
	(2.87)	(2.91)	(2.88)	(4.05)
Minivan	−0.23	0.88	−10.18**	3.17**
	(4.80)	(4.58)	(4.06)	(1.24)
Black non-Hispanic	3.28	−0.71	−4.60	−3.59
	(3.09)	(3.23)	(3.19)	(3.07)
Asian/API	3.08	0	6.65*	6.70*
	(2.90)	(3.11)	(3.45)	(3.50)
Part time	−11.46***	−11.66***	−7.57***	−7.66***
	(2.23)	(2.28)	(2.27)	(2.35)
Transitional work	−4.05	−7.11	−0.99	−0.35
	(2.54)	(2.61)	(2.62)	(2.69)
Other work	−10.68***	−10.18***	−5.07*	−5.79*
	(2.88)	(3.02)	(2.94)	(3.05)
Transportation experience [Yes]	0.31	−0.79	−3.06	−3.13
	(2.01)	(2.08)	(2.04)	(2.08)
Number of observations	600	522	489	495
Number of groups	382	334	314	316
AIC	5,124.03	4,416.75	3,971.26	4,027.17

Notes: The number of groups varies across regressions equations due to participants driving for different platforms. Some nonsignificant controls omitted for clarity. “—” is a placeholder for variables not included in regressions. AIC, Akaike information criterion; OLS, ordinary least squares.

* = p < 0.10; **p < 0.05; ***p < 0.01.

The models in Table 5 examine whether drivers’ relationship with Uber and Lyft affects how they divide their time across these two platforms. In these models, the dependent variable is drivers’ fraction of time that they spend on Uber relative to their total time on Uber and Lyft. The independent variable is how drivers view Uber relative to Lyft, first using the difference in loyalty scales (Uber–Lyft loyalty, model (1)) and also the platform recommendation measures (Uber−Lyft recommendation, model (2)). The results displayed in Table 5 indicate that a driver’s preference for Uber over Lyft has a significant association with the fraction of time a driver works on these services (p < 0.001). The coefficient indicates that a one unit increase in loyalty toward Uber would result in a driver spending approximately 8% more time on Uber. A four-point change (approximately) in a driver’s loyalty would move the dependent variable by one standard deviation. Model (2) in Table 5 uses the recommendation measure as the independent variable of interest. This model returned a similarly strong association (p < 0.001) and coefficient magnitude (0.07). These results suggest that how drivers compare competing ride-hail services predicts how they allocate their labor across these companies.

Table 5.

Linear OLS Models with Random Effects Regressing Fraction of Time on Uber on Relative Preference for Uber over Lyft

Variable name	(1)Fraction of time on Uber	(2)Fraction of time on Uber
Loyalty scale difference	0.08***	—
	(0.01)
Recommendation difference	—	0.07***
		(0.01)
Medium sedan	0.09**	0.11***
	0.04	(0.04)
Large sedan	0.15***	0.19***
	0.05	(0.05)
Four-wheel drive	0.05	0.07*
	(0.04)	(0.04)
Minivan	0.11	0.14**
	(0.07)	0.07***
Hispanic	0.01	(0.01)
	(0.05)	(0.05)
Black non-Hispanic	0.15***	0.15***
	(0.05)	(0.05)
Asian/API	−0.13**	−0.07
	(0.05)	(0.05)
Part time	0.01	0.00
	(0.03)	(0.03)
Transitional work	0.01	−0.01
	(0.04)	(0.04)
Other work	−0.01	−0.00
	(0.05)	(0.04)
Number of observations	374	374
Number of groups	248	248
AIC	97.11	95.80

Notes: Some nonsignificant controls omitted for clarity. “—” acts as a placeholder for variables not included in regressions. AIC, Akaike information criterion; OLS, ordinary least squares.

* = p < 0.10; **p < 0.05; ***p < 0.01.

Finally, the models in Table 6 examine whether workers’ relationship with platform companies affects how they recruit drivers (models (1) and (3)) and passengers (models (2) and (4)). Again, relationship was measured using drivers’ Uber–Lyft loyalty and recommendation scores. For both passengers (p < 0.001) and drivers (p < 0.001), a driver’s organizational loyalty was strongly associated with how likely that driver would recruit passengers for Uber or Lyft. These models suggest that a 1.5 standard deviation shift in the independent variable (0.98) is associated with roughly a 1 standard deviation shift in independent variables (for drivers, 1.58; for passengers, 1.46). Additionally, these models reported an R² above 0.50, suggesting the relationship between drivers and platforms plays an important role in how drivers recruit passengers and drivers. The second independent variable, the platform recommendation item, also significantly predicted how drivers would recruit to Uber and Lyft (p < 0.001).

Table 6.

Linear OLS Models Regressing Likelihood of Driver and Passenger Recruitment on Platform Relationship

Variable name	(1)Driver recruitment difference	(2)Passenger recruitment difference	(3)Driver recruitment difference	(4)Passenger recruitment difference
Loyalty scale difference	1.18***	1.13***	—	—
	(0.07)	(0.06)	—	—
Recommendation difference	—	—	0.78***	0.74***
	—	—	(0.06)	(0.05)
2- or 4-year college degree [Yes]	0.11	−0.08	−0.08	−0.26*
	(0.13)	(0.12)	(0.14)	(0.13)
Female	0.09	0.23	0.02	0.16
	(0.17)	(0.16)	(0.19)	(0.18)
Hispanic	−0.19	0.28	−0.07	0.39*
	(0.21)	(0.20)	(0.23)	(0.22)
Black non-Hispanic	0.10	0.43**	0.25	0.61**
	(0.23)	(0.21)	(0.25)	(0.24)
Asian/API	0.13	0.13	0.04	0.01
	(0.22)	(0.21)	(0.25)	(0.24)
Minivan	0.62**	0.30	0.72**	0.41
	(0.31)	(0.29)	(0.35)	(0.33)
Platform count	0.16*	0.12	0.13	0.01
	(0.08)	(0.08)	(0.09)	(0.09)
Months driving ride-hail	0.00	0.00	0.01	0.01
	(0.01)	(0.00)	(0.01)	(0.01)
Number of observations	282	281	277	276
Adjusted R-squared	0.557	0.541	0.442	0.424

Notes: The number of observations varies across regressions equations due to some incomplete documents. Some nonsignificant controls omitted for clarity. Standard errors are listed below variable coefficients. “—” acts as a placeholder for variables not included in regressions. OLS, ordinary least squares.

* = p < 0.10; **p < 0.05; ***p < 0.01.

Limitations

There are several limitations to this study. Notably, self-reported data raise the question of a common method bias. To address this concern, dependent and independent variables were separated in the study documents, users were granted anonymity, and scales used distinct anchors (Podsakoff, MacKenzie, Lee, and Podsakoff 2003). Furthermore, questions that ask about discrete events, like those in the conflict scale, are less likely to be susceptible to a common method bias compared to psychometric questions (Spector 2006). Finally, the mixed methods nature of this study lends credibility to the association between passenger conflict and drivers’ views of ride-hail companies. That said, future research could improve on the passenger conflict construct in two ways. First, researchers could increase its precision by pairing passenger–driver data, coding dash-cam video, or asking drivers to record these events in real-time. Second, whereas this study uses an aggregate frequency measure, using a disaggregated measure would help clarify if drivers are responding to the totality of these events or if conflict operates on a threshold, wherein a requisite number of incidents are required to evoke a response.

Furthermore, the generalizability of these results could be improved by expanding on its geographic scope, languages of the surveys, and number of distribution sources. Within this study, however, I took several steps to expand its generalizability, such as recruiting from multiple sources and engaging in a mixed-methods approach. Although it would be very challenging to acquire such data, the findings of this study could be improved if replicated using administrative data from multiple platform companies.

General Findings and Discussion

These studies broaden our understanding about control and power in gig work and the organizational dynamics that enable customer abuse in frontline service work. Prevailing online platform research casts the five-star system as a novel and powerful method of regulating workers’ behaviors without extending labor rights to these workers (e.g., Rosenblat and Stark 2016). By examining workers’ behaviors across multiple platforms, however, this study finds that the five-star system is fragile because workers can co-opt companies’ point of control by moving customers across services, which allows workers to retaliate against companies without harming their own future income. This vulnerability had been obscured from extant studies of gig platforms because past research has relied on single platform research designs, ignoring how workers use competing services as an avenue for resistance. By contrast, this study finds that the presence of a competitor changes the power dynamic between workers and platforms because workers can migrate customers onto platforms that provide workers with superior working terms. This finding inverts the way that prior research on the gig economy conceptualizes the company−worker power dynamic because it means workers can capture companies’ point of control by moving customers across services.

More broadly, this study conceptualizes an emerging method of control—laundering control—in which companies embed customers as an additional layer of managerial oversight. This form of control manages a service exchange by controlling the distribution of power between customers and workers. Being able to determine this power imbalance means that platform companies are able to select who is subservient to whom while disavowing their role in tipping the scales in favor of one party. The end result is that companies attempt to recede to the background and allow the natural course of the power imbalance to play out: Workers must cater to customers’ every demand or risk losing their ability to work on a service. In developing this method of control, this article brings into sharp relief that the power used by more traditional service companies to instruct workers that “the customer is king” has merely been laundered through a new entity, customers. Instead of telling workers that the customer is king, laundering control makes the customer king by giving them direct access to organizational power, forcing workers to adhere to customers’ wishes in order to maintain their income. As court systems and legislators weigh how to incorporate platform work into existing legal infrastructure, this finding should make them cautious to accept the claim that platform workers are acting unencumbered by a digital intermediary. Instead, platform companies have the capacity to determine if workers are able to act free from coercion, yet many companies develop their services in a fashion that tips the scales in favor of customers because, similar to more traditional service companies, they are dependent on acquiring and retaining customers.

At the same time, with new configurations of power come new ways that customers can abuse them. As previous research has primarily examined frontline service work in industries that proscribe service rules, such as call centers or food service, scholars have developed a particular understanding of what constitutes customer abuse. By situating this study in a service environment built around laundered control, I find that customers can place workers in legal jeopardy by forcing them to choose between potentially losing their income or acquiescing to customers’ demands. This finding provides a wider context by which to understand growing reports of worker abuse in frontline service work. To date, management scholars, companies, and the media have framed customer abuse as a two-party phenomenon, limiting the discussion to customers and workers. By contrast, this study finds that service organizations place workers in a system where they are forced to endure customers’ abusive demands because of the asymmetric power relationship that service organizations create. Instead of seeing abuse as the consequence of proximate triggers, like an irate customer, this study widens our understanding of abuse to the system-level, centering on how organizations develop asymmetric power structures that force workers into a submissive position when interacting with customers.

Finally, this study illustrates some of the methods that workers use to fight back against this method of control. Absent recourse against customers, workers retaliated against these companies through their recruitment decisions and time allocation across companies. While existing service research focuses on how workers retaliate against their transgressors (i.e., customers), this research reveals that workers name service organizations as contributing transgressors for building an inherent power imbalance into the service exchange. For the wider study of frontline service work, this finding suggests that organizations bear the costs of whomever is granted use of their power and reframes how existing studies present the concept of worker retaliation. While more traditional studies cast worker retaliation as “getting back at” customers for their abusive actions, this study finds that workers are not only retaliating against abusive customers but also the systems that force them to endure these acts of abuse. In doing so, this study explains why service workers react to customer abuse by engaging in organizationally focused acts of retaliation, such as sabotage or neglect.

Conclusion

The “dark side” of service work does not disappear when it gains a user interface. In fact, quite the opposite occurs. This study shows how asymmetric power relationships are encoded into digital platforms through the customer evaluation system and how this system provides the framework from which customer abuse can emerge. The open-ended nature of this control system extends beyond harassment and verbal abuse into forcing workers to take on legal risk. Such behaviors are endemic to laundering control because this system of control provides customers with the capacity to both set the service rules—legal or otherwise—and evaluate workers’ adherence to those rules. By detailing how the management of frontline service workers has evolved from telling workers the customer is king to making the customer king, this article centers on how companies enable customer abuse by developing asymmetric power relationships between workers and customers. In doing so, this article clarifies that platform workers are far from “their own boss” and are instead subject to the same power dynamics found in more traditional frontline service work.

Footnotes

Appendices

Acknowledgements

The author gratefully acknowledges the feedback and insights provided by Alexander Colvin, David Lipsky, and Rachel Aleks. Additionally, I am indebted to J. Ryan Lamare and Ariel Avgar for their advice and suggestions on an earlier draft of this paper. Furthermore, this project would have not been possible without the generous support of Martin and Laurie Scheinman and the Scheinman Institute at Cornell University. Finally, I would like to thank Harry Campbell for his assistance with this project.

The data collection efforts for this article were supported by a research grant from Cornell University’s ILR School.

A preliminary version of this paper was presented at the 2018 LERA summer conference.

For information regarding the data and/or computer programs used for this study, please address correspondence to maffie@psu.edu. Detailed scale development information is available from the author.

ORCID iD

Michael David Maffie

1

Campbell is an influential member of the ride-hail industry. He has been quoted in major media outlets, such as the New York Times, CNBC, CNN, and Newsweek. In 2020, Campbell interviewed current Uber CEO, Dara Khosrowshahi, on his podcast. More information about Campbell can be found on his website: .

References

Aleks

Rachel

Maffie

Michael D.

Saksida

Tina

. 2020. The role of collective bargaining in a digitized workplace. In Pohler

Dionne

(Ed.), Reimagining the Governance of Work and Employment, pp. 85–111. Ithaca, NY: Cornell University Press.

Allan

Steven

Gilbert

Paul

. 2002. Anger and anger expression in relation to perceptions of social rank, entrapment and depressive symptoms. Personality and Individual Differences 32(3): 551–65.

Anderson

Eugene W.

Fornell

Claes

. 2000. Foundations of the American Customer Satisfaction Index. Total Quality Management 11(7): 869–82.

Anteby

Michel

Chan

Curtis K.

2018. A self-fulfilling cycle of coercive surveillance: Workers’ invisibility practices and managerial justification. Organization Science 29(2): 247–63.

Batt

Rosemary

. 1999. Work organization, technology, and performance in customer service and sales. Industrial and Labor Relations Review 52(4): 539–64.

Brotheridge

Céleste M.

Grandey

Alicia A.

2002. Emotional labor and burnout: Comparing two perspectives of “people work.” Journal of Vocational Behavior 60(1): 17–39.

Brotheridge

Céleste M.

Lee

Raymond T.

2002. Testing a conservation of resources model of the dynamics of emotional labor. Journal of Occupational Health Psychology 7(1): 57–67.

Carroll

Wendy R.

2008. The effects of electronic performance monitoring on performance outcomes: A review and meta-analysis symposium: The electronic workplace. Employee Rights and Employment Policy Journal 12(1): 29–48.

Caruana

Albert

Ramaseshan

Ewing

Michael

. 2001. Anomia and deviant behaviour in marketing: Some preliminary evidence. Journal of Managerial Psychology 16(5): 322–38.

10.

Chi

Nai-Wen

Yang

Jixia

Lin

Chia-Ying

. 2016. Service workers’ chain reactions to daily customer mistreatment: Behavioral linkages, mechanisms, and boundary conditions. Journal of Occupational Health Psychology 23(1): 58–70.

11.

Colquitt

Jason A.

Conlon

Donald E.

Wesson

Michael J.

Porter

Christopher O. L. H.

K. Yee

. 2001. Justice at the millennium: A meta-analytic review of 25 years of organizational justice research. Journal of Applied Psychology 86(3): 425–45.

12.

Deng

Xuefei

Joshi

K. D.

Galliers

Robert D.

2016. The duality of empowerment and marginalization in microtask crowdsourcing: Giving voice to the less powerful through value sensitive design. MIS Quarterly 40(2): 279–302.

13.

DeVault

Ileen

Figueroa

Maria

Kotler

Fred B.

Maffie

Michael D.

2019. On-Demand Platform Workers in New York State: The Challenges for Public Policy. Ithaca, NY: The Worker Institute at Cornell University.

14.

DeVellis

Robert

. 2012. Scale Development: Theory and Applications. Los Angeles: SAGE Publishing.

15.

Dormann

Christian

Zapf

Dieter

. 2004. Customer-related social stressors and burnout. Journal of Occupational Health Psychology 9(1): 61–82.

16.

Farrell

Diana

Greig

Fiona

Hamoudi.

Amar

2018. The Online Platform Economy in 2018: Drivers, Sellers, and Lessors. New York: JP Morgan Chase Institute.

17.

Fenner

Yeshe

Garland

Suzanne M.

Moore

Elyse E.

Jayasinghe

Yasmin

Fletcher

Ashley

Tabrizi

Sepher

Gunasekaran

Bharathy

Wark

John D.

2012. Web-based recruiting for health research using a social networking site: An exploratory study. Journal of Medical Internet Research 14(1): e20.

18.

Grandey

Alicia

Dickter

David

Sin

Hock-Peng

. 2004. The customer is not always right: Customer aggression and emotion regulation of service employees. Journal of Organizational Behavior 25(3): 397–418.

19.

Greenbaum

Rebecca

Quade

Matt

Mawritz

Mary

Kim

Joongseo

Crosby

Durand

. 2014. When the customer is unethical: The explanatory role of employee emotional exhaustion onto work-family conflict, relationship conflict with coworkers, and job neglect. Journal of Applied Psychology 99(6): 1188–203.

20.

Groth

Markus

. 2005. Customers as good soldiers: Examining citizenship behaviors in Internet service deliveries. Journal of Management 31(1): 7–27.

21.

Groth

Markus

Nguyen

Helena

Johnson

Anya

. 2019. The moment of truth: A review, synthesis, and research agenda for the customer service experience. Annual Review of Organizational Psychology and Organizational Behavior 6(1): 89–113.

22.

Guidroz

Ashley

Geimer

Jennifer

Clark

Olga

Schwetschenau

Heather

Jex

Steve

. 2010. The nursing incivility scale: Development and validation of an occupation-specific measure. Journal of Nursing Measurement 18(3): 176–200.

23.

Hall

Jonathan V.

Krueger

Alan B.

2017. An analysis of the labor market for Uber’s driver-partners in the United States. ILR Review 71(3): 705–32.

24.

Harris

Lloyd

Daunt

Kate

. 2003. The consequences of dysfunctional customer behavior. Journal of Service Research 6(2): 144–61.

25.

Harris

Lloyd

Daunt

Kate

. 2004. Jaycustomer behavior: An exploration of types and motives in the hospitality industry. Journal of Services Marketing 18(5): 339–57.

26.

Harris

Lloyd

Daunt

Kate

. 2011. Deviant customer behaviour: A study of techniques of neutralisation. Journal of Marketing Management 27(7-8): 834–53.

27.

Hausman

Jerry

. 1978. Specifications tests in econometrics. Econometrica 46(6): 1271–71.

28.

Hochschild

Arlie Russell

. 1983. The Managed Heart: Commercialization of Human Feeling. Berkeley: University of California Press.

29.

Holman

David

Chissick

Claire

Totterdell

Peter

. 2002. The effects of performance monitoring on emotional labor and well-being in call centers. Motivation and Emotion 26(1): 57–81.

30.

Johansen

Bendt

Wedderkopp

Niels

. 2010. Comparison between data obtained through real-time data capture by SMS and a retrospective telephone interview. Chiropractic & Osteopathy 18(10): 1–7.

31.

Koopmann

Jaclyn

Wang

Liu

Yihao

Song

Yifan

. 2015. Customer mistreatment: A review of conceptualizations and a multilevel theoretical model. Research in Occupational Stress and Well Being 13: 33–79.

32.

Lee

Jae Eun

Batt

Rosemary

Moynihan

Lisa M.

2019. Strategic dilemmas: How managers use HR practices to meet multiple goals. British Journal of Industrial Relations 57(3): 513–39.

33.

Lee

Min Kyung

Kusbit

Daniel

Metsky

Evan

Dabbish

Laura

. 2015. Working with machines: The impact of algorithmic and data-driven management on human workers. In: Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (CHI ‘15), pp. 1603–612. New York: Association for Computing Machinery. doi: https://doi.org/10.1145/2702123.2702548.

34.

Leidner

Robin

. 1993. Fast Food, Fast Talk: Service Work and the Routinization of Everyday Life. Berkeley: University of California Press.

35.

Lovelock

Christopher

. 1994. Product Plus: How Product + Service = Competitive Advantage. Ann Arbor, MI: McGraw-Hill.

36.

Podsakoff

Philip M.

MacKenzie

Scott B.

Lee

Jeong-Yeon

Podsakoff

Nathan P.

2003. Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology 88(5): 879–903.

37.

Prassl

Jeremias

. 2018. Humans as a Service: The Promise and Perils of Work in the Gig Economy. London: Oxford University Press.

38.

Ramo

Danielle E.

Prochaska

Judith J.

2012. Broad reach and targeted recruitment using Facebook for an online survey of young adult substance use. Journal of Medical Internet Research 14(1): e20.

39.

Rosenblat

Alex

. 2018. Uberland: How Algorithms Are Rewriting the Rules of Work. Oakland: University of California Press.

40.

Rosenblat

Alex

Stark

Luke

. 2016. Algorithmic labor and information asymmetries: A case study of Uber’s drivers. International Journal of Communication 10(27): 3758–84.

41.

Rosenblat

Alex

Levy

Karen E. C.

Barocas

Solon

Hwang

Tim

. 2017. Discriminating tastes: Uber’s customer ratings as vehicles for workplace discrimination. Policy & Internet 9(3): 256–79.

42.

Rosenthal

Patrice

. 2004. Management control as an employee resource: The case of front-line service workers. Journal of Management Studies 41(4): 601–22.

43.

Rupp

Deborah

Spencer

Sharmin

. 2006. When customers lash out: The effects of customer interactional injustice on emotional labor and the mediating role of discrete emotions. Journal of Applied Psychology 91(4): 971–78.

44.

Rysman

Marc

. 2009. The economics of two-sided markets. Journal of Economic Perspectives 23(3): 125–43.

45.

Schor

Juliet

. 2014. Debating the sharing economy. Great Transformation Initiative. Accessed at https://greattransition.org/publication/debating-the-sharing-economy.

46.

Shao

Ruodan

Skarlicki

Daniel P.

2014. Service employees’ reactions to mistreatment by customers: A comparison between North America and East Asia. Personnel Psychology 67(1): 23–59.

47.

Shiffman

Saul

Stone

Arthur A.

Hufford

Michael R.

2008. Ecological momentary assessment. Annual Review of Clinical Psychology 4(1): 1–32.

48.

Skarlicki

Daniel P.

Folger

Robert

. 1997. Retaliation in the workplace: The roles of distributive, procedural, and interactional justice. Journal of Applied Psychology 82(3): 434–43.

49.

Skarlicki

Daniel P.

van Jaarsveld

Danielle D.

Walker

David D.

2008. Getting even for customer mistreatment: The role of moral identity in the relationship between customer interpersonal injustice and employee sabotage. Journal of Applied Psychology 93(6): 1335–47.

50.

Sliter

Michael

Jones

Morgan

. 2015. A qualitative and quantitative examination of the antecedents of customer incivility. Journal of Occupational Health Psychology 21(2): 208–19.

51.

Sliter

Michael

Jex

Steve

Wolford

Katherine

McInnerney

Joanne

. 2010. How rude! Emotional labor as a mediator between customer incivility and employee outcomes. Journal of Occupational Health Psychology 15(4): 468–81.

52.

Solomon

Michael R.

Surprenant

Carol

Czepiel

John A.

Gutman

Evelyn G.

1985. A role theory perspective on dyadic interactions: The service encounter. Journal of Marketing 49(1): 99–111.

53.

Song

Peijian

Xue

Ling

Rai

Arun

Zhang

Cheng

. 2018. The ecosystem of software platform: A study of asymmetric cross-side network effects and platform governance. MIS Quarterly 42(1): 121–42, A1–A6.

54.

Spector

Paul

. 2006. Method variance in organizational research: Truth or urban legend? Organizational Research Methods 9(2): 221–32.

55.

Spencer

Sharmin

Rupp

Deborah

. 2009. Angry, guilty, and conflicted: Injustice toward coworkers heightens emotional labor through cognitive and emotional mechanisms. Journal of Applied Psychology 94(2): 429–44.

56.

Strauss

Anselm

Corbin

Juliet

. 2008. Basics of Qualitative Research: Techniques and Procedures for Developing Grounded Theory. Volume 3. Newbury Park, CA: SAGE Publishing.

57.

Tabachnick

Barbara G.

Fidell

Linda S.

2001. Using Multivariate Statistics. 4th edition. Needham Heights, MA: Allyn & Bacon.

58.

Ticona

Julia

Mateescu

Alexandra

Rosenblat

Alex

. 2018. Beyond disruption: How tech shapes labor across domestic work and ridehailing. Report, June 27. New York: Data & Society Research Institute.

59.

Tiwana

Amrit

Konsynski

Benn

Bush

Ashley

. 2010. Research commentary—Platform evolution: Coevolution of platform architecture, governance, and environmental dynamics. Information Systems Research 21(4): 675–87.

60.

Uber. 2019. Understanding acceptance rates. Uber Help. Accessed at https://help.uber.com/partners/article/understanding-acceptance-rates?nodeId=9a0cb1d3-b34d-4e97-937f-d1c5775b29eb.

61.

Van Dyne

Linn

Graham

Jill W.

Dienesch

Richard M.

1994. Organizational citizenship behavior: Construct redefinition, measurement, and validation. Academy of Management Journal 37(4):765–802.

62.

Van Jaarsveld

Danielle

Walker

David

Skarlicki

Daniel

. 2010. The role of job demands and emotional exhaustion in the relationship between customer and employee incivility. Journal of Management 36(6): 1486–504.

63.

Van Maanen

John

. 1999. The smile factory: Work at Disneyland. In Frost

Peter

(Ed.), Reimagining Organizational Culture, pp. 58–76. Newbury Park, CA: SAGE Publications.

64.

Wang

Liao

Hui

Zhan

Yujie

Shi

Junqi

. 2011. Daily customer mistreatment and employee sabotage against customers: Examining emotion and resource perspectives. Academy of Management Journal 54(2): 312–34.

65.

Winter

Susan

Berente

Nicholas

Howison

James

Butler

Brian

. 2014. Beyond the organizational “container”: Conceptualizing 21st century sociotechnical work. Information and Organization 24(4): 250–69.