Abstract
Formally, gig workers determine their work hours. Gig platforms, however, shape work schedules with many tools, including ratings, deadlines, pay, and competition. The platform dependence perspective suggests that gig workers who rely on platform income are especially vulnerable to manipulation and often fail to fulfill their scheduling preferences. We offer one of the first quantitative tests of this prediction while also examining if the theory applies to highly flexible online gig work. Using a U.S. sample of Amazon Mechanical Turk workers, we find that mismatches between preferred and actual schedules are common and increase with hours worked. As predicted, mismatches are especially prevalent among platform-dependent respondents, but group differences are only significant for the minority of respondents working more than 30 hours per week. More research is needed to understand schedule mismatches among low-hour gig workers and how “strategic neglect” may help generate them.
In theory, people who do gig work through platforms like Uber, DoorDash, or Amazon Mechanical Turk are free to set their own work schedules. Such platforms often promise high levels of flexibility that allow people to “be their own boss” and earn extra money through a “side hustle” (Baber 2024). Indeed, workers on these platforms do not have supervisors who dictate when or how many hours to work—they are officially free to organize their work on the platform as they wish. Gig work arrangements are thus free from the direct managerial control that shapes work in most conventional jobs.
In practice, however, gig platform workers do not always get the schedules they prefer. Gig work can involve unsocial, unstable, and undesirably long hours (Rahman 2021; Wood et al. 2019). The “algorithmic control” documented in many studies of gig work (Kellogg, Valentine, and Christin 2020) plays a role in generating these puzzling outcomes (Schor et al. 2020). Still, it is only part of the story. Workers struggle to control their schedules even on platforms that do not rely heavily on algorithms to shape worker behavior (Lehdonvirta 2018).
One understudied piece of this puzzle is how worker heterogeneity is related to the fit between workers’ preferred and actual schedules. Although many gig workers may strive to thwart platform control (Joyce and Stuart 2021; Kellogg et al. 2020), they vary in their capacity for resistance. Reliance on platform earnings, in particular, may influence whether formal control (i.e., the authority to set one’s own schedule) translates into a close fit between workers’ preferred and actual schedules.
In this article, we draw on the theory of platform dependence to examine variation in the fit between preferred and actual work schedules on a single gig platform. The theory suggests that financial dependence on platform income is a “disciplinary device” that encourages people to take any work they can find on a platform and cede control of work schedules to customer and platform demands, thus leaving workers less likely to work the schedules they prefer (Schor et al. 2020). The theory also suggests that this relationship holds across many gig platforms.
Our analysis uses data from a large sample of Amazon Mechanical Turk (MTurk) workers from across the United States gathered in February 2020. We use daily schedule data from an entire week to examine how workers’ actual schedules correspond to their preferred and expected schedules. The platform dependence perspective suggests that people who need the platform income will have a worse fit between their preferred and actual schedules than workers who are not dependent. Schor et al. (2020), however, developed the theory without examining experiences on gig platforms like MTurk, which offer a distinct type of “microtask” work and especially high levels of formal control over when, how long, and where people work. Our study helps test the scope of platform dependence theory and contributes to research on schedule control on gig platforms more generally. Our study is also timely: The demand for microtask workers is likely to grow rapidly given the essential role they play in training, monitoring, and supporting artificial intelligence (Altenried 2020; Gray and Suri 2019; Stephens 2023; Tubaro, Casilli, and Coville 2020).
Background
Heterogeneity of Gig Work and Gig Workers
Definitions of gig work vary, but many scholars emphasize two central characteristics: Gig work relies on digital platforms or apps, and it is organized around “gigs,” that is, short-term engagements between workers and employers/customers (Kalleberg and Dunn 2016). The digital component of gig work is relatively new, but short-term, informal engagements between workers and employers/customers were common before 1950 (Kalleberg and Dunn 2016). Gig work is thus a new version of an old employment arrangement.
Scholars have also called attention to the great heterogeneity of gig work and how it can shape worker outcomes. Different kinds of gig work require different skills and assets, offer different rewards, and expose workers to different risks. Gig work can require face-to-face or only remote interactions (Kalleberg and Dunn 2016). Gig work can also require little or much economic capital. MTurk workers, for instance, only need a computer or mobile device to do their work, but Uber drivers need driver’s licenses and cars, and Airbnb hosts need a room, apartment, or house to rent. Platforms requiring more economic capital tend to provide higher earnings than platforms requiring less capital (Schor et al. 2020). Also, platforms that require specialized training/qualifications, like Freelancer.com, tend to pay more than platforms with fewer skill requirements, like MTurk (Kalleberg and Dunn 2016; Vallas and Schor 2020). Due to this heterogeneity, it is difficult to make generalizations about gig work.
Furthermore, scholars increasingly recognize that even people doing the same kind of gig work on the same platform can have very different experiences because gig workers themselves are a heterogeneous population. In part, this is because gig platforms allow great diversity in work hours and can thus accommodate people looking for very different things from the work (Dunn 2020; Schor et al. 2020). Gig work can be attractive to people who want flexible, part-time work and are less concerned about job security or fringe benefits, for example, students, homemakers, and retirees (Berg 2016; Moss et al. 2020). It can also attract people looking for full-time work, but it is often a poor fit for those who need full-time work that provides a steady income and health insurance (Kalleberg 2018).
These insights regarding the heterogeneity of gig work and gig workers motivate our analysis. We build on existing research and highlight the heterogeneity of gig work by studying the unique situation of microtask workers. We also examine how variations in workers’ circumstances outside a platform can generate within-platform heterogeneity. Specifically, we offer one of the first quantitative evaluations of the hypothesized link between platform dependence and work schedules.
The Case of Microtask Work
Gig workers’ ability to fulfill their scheduling preferences is likely to vary across platforms because of differences in the types of work platforms offer, and in this respect, microtask work seems to have some advantages. Workers who transport people or goods (for platforms like Uber, Postmates, or DoorDash) and those who do household tasks (e.g., TaskRabbit) must meet customers at particular places and times. This can make it difficult for people doing those kinds of gig work to fulfill their scheduling preferences. Indeed, depending on the current traffic, weather, time of day, or family responsibilities, workers may not want to fulfill a request to meet a customer at a specified time and place. Microtask workers, in contrast, do not need to meet customers at particular times or leave their homes. Furthermore, microtask work can be done at any hour of the day. It is not closely synchronized with variations in customer demand that can indirectly pressure rideshare and food delivery workers to work during mealtimes, rush hour, on weekends, or at night. Microtask jobs can also typically be completed very quickly, making it easier to integrate with other activities than gig work that requires larger chunks of available time. 1 Finally, microtask workers are generally not assigned tasks—rather, when they want to work, they select from the pool of available tasks, which means they do not have to worry about being punished for declining assignments (Lascău et al. 2024). Due to all of these factors, microtask workers may have an advantage when pursuing their scheduling preferences.
The fit between gig workers’ preferred and actual schedules may also vary across platforms offering the same type of work (e.g., microtask work) due to variations in the strategies platforms use to control workers. Studies guided by labor process theory clarify that although gig platforms generally lack direct managerial oversight, they employ many different tools to shape worker behavior (Joyce and Stuart 2021; Kellogg et al. 2020). Platforms, for instance, can shape gig work schedules using algorithms that monitor and respond to worker activity (Alvarez de la Vega et al. 2023) and customer ratings (Rahman 2021). They can design platforms to prioritize customer demands (Duggan et al. 2023) and enforce task deadlines (Yin, Suri, and Gray 2018). They can also shape schedules through the distribution of work, the regulation of payments (e.g., surge pricing), and competition (Lascău et al. 2022). Baber (2024:729–730) concludes that although platforms portray the work they offer as flexible, platforms actually strive to “craft a particular type of worker through . . . customer ratings, algorithmic assignment of jobs and app based surveillance mechanisms.” Much of their power to engage in these manipulations stems from a business model in which workers are classified as nonemployees, thus allowing platforms to sidestep many labor regulations (Baber 2024; Joyce and Stuart 2021). The degree to which gig workers get the schedules they prefer thus depends on the specific strategies a platform uses to control workers. Notably, due to the design of the platform, workers on MTurk have more control over their schedules than workers on some other microtask platforms (Lascău et al. 2022). MTurk workers, for instance, are not required to complete one job at a time (e.g., they can multitask by claiming a batch of tasks before working on any of them), and they are not required to wait a certain amount of time between jobs.
All these considerations suggest that people on the MTurk platform may often work the hours they prefer, but this prediction may be too optimistic. Indeed, two studies indicate that MTurk workers can have great difficulty controlling their schedules (Lascău et al. 2024; Lehdonvirta 2018). One crucial question is whether the outcomes these studies highlight are typical—both focused on small samples of highly active MTurk workers. Rather than expecting uniformly good or bad schedule outcomes, perhaps we should expect variations in outcomes even among workers on the same platform.
The labor process perspective hints at potential variations in workers outcomes by emphasizing that workers resist platform control and use a variety of tactics to exert their agency (Joyce and Stuart 2021; Kellogg et al. 2020). The implication is that some workers will retain more control over their work and work schedules than others. More generally, it suggests that worker outcomes will vary both across and within platforms. Recent work supports this idea by documenting variations in collective action in response to algorithmic control (Cini 2023) and varied resistance strategies among workers on the same platform (Rahman 2021; Toxtli, Suri, and Savage 2021). This evidence provides a useful corrective to the idea that all workers on a platform will have the same experiences or that platform technology completely overrides workers’ agency (Joyce and Stuart 2021). Indeed, like all other technologies, the implementation and consequences of algorithmic control and platform design are not deterministic—they are shaped by social factors (Joyce et al. 2023).
Why, then, might some workers be especially vulnerable to platform control (i.e., working at unwanted days, times, and durations) while others on the same platform are better able to work the schedules they want? Platform dependence theory offers guidance on a key worker characteristic that might influence the fit between preferred and actual schedules.
Platform Dependence
Drawing inspiration from other qualitative studies, including Lehdonvirta (2018) and Ravenelle (2019), Schor et al. (2020) suggest that workers on the same platform will vary in their ability to fulfill their schedule preferences due to differences in their financial dependence on gig work. Schor et al. (2020:833) describe platform dependence as “the extent to which workers are dependent on platform income to pay basic expenses rather than working for supplemental income.”
To date, studies that use this perspective to guide quantitative analyses have operationalized the concept of platform dependence using information about gig workers’ relationship to the conventional economy. Some studies classify workers as platform dependent if platform work is their only job (e.g., Reynolds and Kincaid 2023). Other studies focus on whether workers consider platform work their “main job” (e.g., Glavin and Schieman 2022; Keith, Harms, and Tay 2019; Rahman 2021). These approaches echo the idea that “Dependent earners are those who are wholly or primarily dependent on the platform for their livelihoods.” (Schor et al. 2020:841).
The concept of platform dependence, however, can also be measured in a subjective way by focusing on whether respondents consider the money earned through a platform an essential or supplementary part of their income (Brawley Newlin 2023). This subjective approach is important because people doing platform work as their sole or main job are not always financially dependent on it. Those who are retired or supported by parents or partners may do platform work for fun or to earn “extra” money and thus not need platform income even though it is their only job. Furthermore, if platform workers have other jobs, that does not mean they could meet their basic living expenses without platform income. Indeed, many people with conventional jobs pursue additional employment like platform work precisely because the wages provided by their conventional jobs are inadequate (Piasna et al. 2021). Given these considerations, we identify respondents who have another job, but we focus on whether they report needing the income they earn from platform work as our key indicator of platform dependence.
We also argue that work schedule outcomes are central to the platform dependence perspective and deserve more attention. Quantitative studies using the theory have examined worker distress (Glavin and Schieman 2022), motivations for doing gig work, and life satisfaction (Keith et al. 2019; Reynolds, Aguilar, and Kincaid 2024). These are all relevant and important issues. Schor et al. (2020), however, also argue that workers who are financially dependent on platform income have less control over their schedules than workers who are not dependent. They write: “Those who are not dependent on the platforms have better experiences and more control over when and how they work. They are more discriminating about whom they accept as customers, the amount of time they work, their conditions of work, and their schedules” (Schor et al. 2020:841–42).
When describing platform-dependent Uber drivers and delivery drivers, Schor et al. (2020) describe the loss of control over work schedules in detail. They write, “While supplemental earners are able to turn off their apps when business is slow, traffic is bad, or weather is forbidding, dependents find themselves locked into undesirable situations.” (Schor et al. 2020:850). They also note that many platform-dependent workers also “feel compelled to work outside of conventional office hours, e.g., weekends and late evenings” (Schor et al. 2020:852). They caution, however, that the robustness of their findings should be tested in other subtypes of gig work. Indeed, traffic and weather are irrelevant for online gig work. On platforms like MTurk, work is also less likely to require effort at particular times because the tasks are not tied to the rhythms of daily life like they are on ridesharing and delivery platforms. Still, these potential advantages of online gig work may be outweighed by fierce competition, the unpredictable supply of work, and the ensuing pressure to grab tasks before someone else takes them (Lascău et al. 2024).
Our Contributions
We contribute to the development of the platform dependence perspective and the study of gig work more generally by examining how platform dependence is related to the work schedules of MTurk workers across the United States. Schor et al. (2020) broadened research on gig work by studying workers on seven different types of gig platforms. Their sample, however, had two limitations: It did not include microtask workers, and it relied on respondents from the Boston area. Our analysis answers the call for research that buttresses the theory by addressing both limitations (Kalleberg 2021). First, we examine if the theory explains work schedules among MTurk workers. Given the great formal freedom MTurk workers have in determining the duration, timing, and location of their work, platform dependence might have little impact on their ability to fulfill their work schedule preferences. Second, by surveying MTurk workers across the United States, we test the generalizability of the theory to gig workers beyond the Boston area.
Furthermore, we provide one of the first quantitative tests of the theory’s predictions about work schedules. Using workers’ expected, actual, and preferred microtask schedules for an entire week, we measure the extent to which their actual schedules deviate from their expected schedules (i.e., expectation mismatches) and preferred schedules (i.e., preference mismatches). We pay special attention to differences in the schedules of those who need the income and those who do not.
Hypotheses
We focus on four hypotheses. The theory suggests that dependent workers will be less able than nondependent workers to work the schedules they prefer. This should be evident in the number of mismatches between their preferred and actual schedules. Our first hypothesis is thus:
Hypothesis 1: Platform dependence will be associated with preference mismatches.
Platform dependence also suggests that a particular mechanism will help generate work schedule mismatches. Specifically, it implies that dependent workers are more likely to have schedules they do not prefer because ceding control over their schedules to platforms and customers makes their work schedules unpredictable (Schor et al. 2020). People who need the money tend to accept any work they can find—the problem is that there is no reliable way to know exactly when work will be available (Lascău et al. 2024). Working at unplanned times could have a negligible or even a positive effect on the fit between a worker’s actual and preferred schedule. Still, when workers cannot predict their schedules, they cannot coordinate their platform work with the rest of their activities, and we suspect that will increase the chances of a mismatch between their preferred and actual schedules. We thus hypothesize:
Hypothesis 2: Expectation mismatches will help mediate the relationship between platform dependence and preference mismatches.
Furthermore, although the different outcomes for dependent and nondependent workers may reflect the amount of time each group spends on a platform, platform dependence theory implies that the two groups experience each hour of work on platforms in fundamentally different ways. Platform dependence recognizes that platforms shape worker schedules using tools like algorithmic control. In this way, it suggests that as workers spend more time on a platform, they will also accumulate mismatches between their preferred and actual schedules. The theory, however, also predicts within-platform variations in worker outcomes by positing that because of their different financial resources, dependent and nondependent workers differ fundamentally in their ability to resist the pull of platforms and retain control over their work schedules. In this way, the theory also suggests that, on average, each hour of microtask work will generate more mismatches among platform-dependent workers than among nondependent workers. We thus make two additional hypotheses:
Hypothesis 3: Higher microtask work hours will be associated with more preference mismatches.
Hypothesis 4: The positive relationship between microtask hours and preference mismatches (Hypothesis 3) will be stronger among platform-dependent workers than among nondependent workers.
Data
We test the predictions derived from the platform dependence perspective using data from a large sample of U.S. MTurk workers gathered in February 2020. It is not possible to recruit a true probability sample of MTurk workers (Difallah, Filatova, and Ipeirotis 2018). Like other studies, we therefore utilized purposive sampling (Berg 2016; Koustas 2019; Schor et al. 2020), but we took several extra steps to ensure that our sample reflects the general MTurk population as closely as possible. First, we posted our main survey on MTurk across seven consecutive days, eight times each day, so workers could participate no matter what days or times they typically work on the platform (Berg et al. 2018). Second, we adjusted the number of surveys in each posting to reflect activity levels at different times of day (Hitlin 2016). Finally, although it is common to impose eligibility restrictions such as number of tasks completed or approval rates (Robinson et al. 2019), we made our survey available to all U.S workers to recruit a more diverse population. All these steps increase our sample’s ability to reflect the MTurk worker population at the time of data collection.
Our data are also especially suitable for examining predictions about work schedules because in addition to the main survey, we collected daily work schedule information in the week after the main survey. This combination of data allows us to analyze workers’ schedules in great detail. For instance, we can identify schedule mismatches by comparing the work schedules respondents actually worked to the schedules they expected or preferred to work.
To create our analytic sample, we started with the 1,786 respondents who completed the main survey (where respondents reported their expected schedules and demographics). We dropped 70 respondents who did not appear to be in the United States and another 371 who did not pass our data quality checks (see the Appendix for details). We dropped another 393 respondents because they did not complete all seven daily surveys, thus providing incomplete information about their weekly work schedule. Finally, we dropped 113 more respondents who were missing on one or more of the variables used in our analysis, leaving a final sample of 839. As shown in Appendix Table A2, imputing missing values leads to virtually the same patterns of statistical significance for the focal variables in our final regression model.
Measures
Platform Dependence
To distinguish financial dependence on microtask work from involvement in other paid work, we measure platform dependence by focusing on respondents’ explanations for the hours they spend on microtask work. We started by asking respondents, “Consider how working more or fewer hours per week in this type of work would change your situation at work and outside work. What would currently be best for you?” The answer choices included spending fewer, more, or the same hours on microtask work and “I don’t know.” No matter what answer they chose, we then asked respondents about the main reasons they were not actually working fewer hours. Answer choices included: “I need the money,” “I need to build experience,” “I enjoy this work,” “I cannot finish my work in fewer hours,” and items that are only relevant for other types of work, such as, “I would not get promoted.” Respondents were allowed to choose more than one answer. We classify respondents as platform dependent if they did not work fewer hours because they need the money.
Preference Mismatches and Expectation Mismatches
We collected data about expected work schedules in the background survey. We asked respondents to report all types of work they expected to do in the week after the background survey, where they would do it, and the start and stop time for each “work period” to the nearest 15-minute increment. The directions explained that, “A new ‘work period’ begins any time you switch the type of work you are doing or return from a break for lunch, rest, etc.” Respondents, for instance, might report that on Monday they expected to do microtasks at home from 7:00 a.m. to 8:00 a.m., go to work as a regular employee from 9:00 a.m. to 5:00 p.m., and then do more microtasks at home from 9:00 p.m. to 11:00 p.m.
In the seven daily surveys collected in the week following the background survey, we used the same approach to gather information about respondents’ actual and preferred schedules. The actual schedule was simply the schedule they worked in 15-minute increments. To measure their preferred schedules for each day, we provided the same response options used for the expected and actual work schedules and asked, “Given your current situation at work and outside of work, exactly what work schedule would have been best for you on [day of the week]?”
We constructed two variables for our analysis by comparing respondents’ actual microtask schedules with both their preferred and expected microtask schedules. Our dependent variable measures preference mismatches. This is the degree of mismatch between respondents’ actual and preferred microtask schedule, assessed by comparing each 15-minute time slot in the actual and preferred schedule for each day. For any given time slot, work schedule mismatches can take two forms: A person may be (a) not working but prefer to work or (b) working but prefer not to work. We examine this distinction later, but we focus primarily on the more general measure by summing the total number of mismatches between the actual and preferred state across the 672 fifteen-minute periods in the week. In sequence analysis parlance, this is the Hamming distance between the two sequences that represent the actual and preferred schedule (Studer and Ritschard 2015). To make the measure more intuitive, we convert it to hours of mismatches per week. We used the same procedure to measure our mediating variable, expectation mismatches: discrepancies between actual and expected schedules.
Controls
We also include some additional variables in our analysis to control for other factors that might be related to work schedule outcomes. We measured age and the months of experience that each respondent had with MTurk. Due to gendered expectations for women to spend more time on household work (Matteazzi and Scherer 2021), which can lead to fewer hours spent on microtask platforms (Lehdonvirta 2018), we controlled for gender (women = 1, men = 0). Given past research suggesting that people of color are more likely to be platform dependent (Glavin and Schieman 2022), we also classified respondents as White (non-Hispanic), Black (non-Hispanic), Hispanic, or other. We control for the presence and employment status of partners with a set of indicator variables identifying respondents as having no partner, a partner who does not work for pay, a partner who works part-time, or a partner who works full-time. We also control for the presence of resident children (1 = resident children, 0 = no resident children). We control for education (1 = bachelor’s degree or higher, 0 = no bachelor’s degree) because people with more education tend to have better opportunities in the conventional job market (and thus lower chances of being dependent on microtask work), which might increase their chances of working the schedules they prefer (Schor et al. 2020). Finally, we use an indicator variable to identify respondents who combine microtask work with nonmicrotask work. Appendix Table A1 compares dependent and nondependent workers on each of these control variables.
Analytic Plan
To test our hypotheses, we estimated a series of nested ordinary least squares (OLS) regressions (see Table 2). Model 1 regresses preference mismatches on platform dependence to test whether platform-dependent workers had more preference mismatches on average (Hypothesis 1). Model 2 adds control variables to examine whether differences between dependent and nondependent workers persist after adjusting for other factors (Hypothesis 1). Model 3 adds expectation mismatches to examine whether less predictable microtask work schedules also tend to be less preferable and whether this association helps explain part of the difference between dependent and nondependent workers’ hours of preference mismatches (Hypothesis 2). Model 4 adds number of microtask hours (during the week of data collection) to examine whether working more hours increases workers’ propensity for preference mismatches (Hypothesis 3). Finally, Model 5 adds an interaction term between platform dependence and weekly microtask hours to examine whether each hour spent on microtasks leads to more preference mismatches for dependent workers compared to nondependent workers (Hypothesis 4).
Results
Platform dependence is common among our respondents and associated with undesirable work schedule outcomes. Over three quarters of our respondents are dependent on the income they earn doing microtasks (Table 1). This is consistent with the suggestion that platforms with few asset requirements will have many dependent workers (Schor et al. 2020).
Mean Hours of Work and Schedule Mismatches by Platform Dependence.
p < .01. ***p < .001.
Furthermore, platform-dependent workers have more difficulty than nondependent workers predicting their schedules and fulfilling their schedule preferences. Dependent workers have 22.0 hours of mismatches between their expected and actual schedules, whereas nondependent workers have 17.9 (difference p < .001). Dependent workers are thus less able to predict their schedules. Dependent workers also have 9.8 hours of mismatches between their preferred and actual schedules, whereas nondependent workers have only 6.2 (difference p < .01). Dependent workers are thus less able to work the schedules they prefer. These patterns provide some initial support for Hypothesis 1. It is also notable that respondents have many hours of mismatches relative to the number of hours they spend doing microtasks, especially if they are platform dependent. Among workers who are not platform dependent, the ratio of preference mismatches to actual work hours is (6.2 / 11.0) = 0.56. Among platform dependent workers, the ratio is 0.64.
A comparison of expectation mismatches and preference mismatches is also instructive. Specifically, the fit between actual and preferred schedules is somewhat better than the fit between actual and expected schedules. This suggests that when respondents do not work the schedules they expected, they do not always object to the changes. This is consistent with the idea that some deviations from the expected schedule reflect the influence of workers’ own preferences—an ability to change their schedules as needed. However, deviations from expected schedules are still strongly associated with preference mismatches.
As shown in Figure 1, workers who have difficulty predicting their schedules have more trouble fulfilling their work schedule preferences. Specifically, expectation mismatches are correlated with preference mismatches (r = .41; p < .000). This implies that when people cannot predict when they will be working (or not working), they tend to have schedules they do not prefer. This is consistent with Hypothesis 2.
Expectation mismatches and preference mismatches.
OLS regressions predicting preference mismatches across the week provide additional support for these bivariate findings. Model 1 in Table 2 shows that, on average, platform-dependent workers experienced 3.675 more hours of preference mismatches per week than nondependent workers. After adjusting for demographic characteristics, dependent workers had just under 3 more hours of preference mismatches (Model 2). Both models provide support for Hypothesis 1.
Ordinary Least Squares Regressions Predicting Preference Mismatches.
p < .05. **p < .01. ***p < .001.
The multivariate analysis also supports the idea that dependent workers have difficulty fulfilling their work schedule preferences because their schedules are less predictable. Model 3 shows that preference mismatches are associated with expectation mismatches. Specifically, each additional hour of mismatch between a worker’s expected and actual schedule is associated with an additional 0.296 hours of mismatch between the preferred and actual schedules. This reinforces the idea that unexpected schedule changes also tend to be undesirable. Furthermore, after accounting for this relationship between the two types of mismatches, the coefficient for platform dependence shrinks (by about 43 percent) and loses statistical significance. The reduction in the coefficient for platform dependence suggests a mediating effect whereby dependence leads to unexpected work hours, which lead to a worse fit between a worker’s preferred and actual schedules. A formal mediation test with bootstrapped standard errors from 10,000 replications indicates that the indirect effect (3.994 × 0.296 = 1.18) is statistically significant (p = .010).
Furthermore, the more hours respondents spent on microtask work per week, the more preference mismatches they tended to have. When we treat the relationship as a linear function, we estimate that each hour spent on microtask is associated with an additional one-third of an hour of preference mismatches (Model 4). Model 5, however, which adds hours squared, suggests that the relationship is best modeled as nonlinear. The significant squared term indicates that the connection between microtask hours and preference mismatches grows even stronger as the number of microtask hours increases. This supports Hypothesis 3, and it is consistent with the idea that many MTurk workers have trouble working the schedules they prefer.
Our final regression model adds an interaction between platform dependence and work hours to test Hypothesis 4. The results indicate that the relationship between microtask hours and preference mismatches is different for dependent workers than for nondependent workers. The implications of the interactions are difficult to discern from the regression coefficients, so following Mize (2019), we present the predicted values of the dependent variable in Figure 2 and discuss average marginal effects. Figure 2 suggests that for dependent and nondependent workers who spend fewer than 30 hours per week on microtasks, work hour increases have a similar effect. However, among those working more than 30 hours per week, additional hours are associated with significantly more preference mismatches for dependent workers than for nondependent workers. Average marginal effects calculated from Model 6 support this impression. For instance, increasing the number of microtask hours from 5 to 15 has a similar effect on preference mismatches among nondependent workers (3.041; p = .002) and dependent workers (2.422; p < .000). The difference between the two effects (3.041 – 2.422 = 0.619) is not significant (p = .581). Working more hours, however, has different effects for the two groups if respondents work more than 30 hours per week. Among nondependent respondents, increasing microtask hours from 30 to 40 hours per week is predicted to generate a 1.985 hour decrease in preference mismatches, but the effect is not significant (p = .208). Among dependent workers, in contrast, working 40 rather than 30 hours a week is associated with 4.852 more hours of preference mismatches (p = .000). In this case, the difference between the effect for nondependent and dependent workers (–1.985 – 4.852 = –6.837) is significant (p < .001). These findings regarding the effect of platform dependence below and above 30 hours are important when assessing support for the theory because roughly 88 percent of dependent workers and 93 percent of nondependent workers in our sample work less than 30 hours per week.
Preference mismatches by work hours and platform dependence.
To assess how restricting our sample to respondents who completed all seven daily surveys and using listwise deletion may have influenced our results, we estimated several modified versions of the final regression model in Table 2 (see Appendix Table A2). We began by estimating the final model as a structural equation model without any other modifications. This reproduced the results shown in Table 2. Then we used the mvml option in Stata 17 to estimate three additional versions of the model that used full-information maximum likelihood (FIML) to recover cases that were not included in the original version of the model. The first additional model was restricted to respondents who completed all seven daily surveys, but it used FIML to recover cases lost due to listwise deletion (N = 1,012). The second model recovered cases lost to listwise deletion and respondents who completed at least one daily survey (N = 1,318). The final model recovered as many respondents as possible (N = 1,564). The patterns of significance for the focal variables were nearly the same in these supplemental models as in the original model. In fact, the results for the first interaction are slightly stronger. This suggests that missing data and attrition had little effect on the main regression results.
Because our analysis of preference mismatches only highlights one of the ways the work schedules of dependent and nondependent workers can differ, we also conducted a supplementary descriptive analysis using basic statistics and some sequence analysis measures (Ritschard 2021; Studer et al. 2011) to explore additional differences. As noted previously, platform-dependent respondents do about 4 more hours of microtask work per week than nondependent respondents. The supplementary calculations show that this difference in the number of hours worked reflects each group’s preferred weekly hours (Table 3). Indeed, the groups are very similar in the correspondence between the number of hours they prefer to work and the number of hours they actually work: Both work just under 1 hour less per week than they prefer (Table 3, row 3). As shown in the last four rows of the top panel of Table 3, platform-dependent workers also tend to have significantly more variation in their actual microtask schedules across the days of the week. These variations reflect differences in the number of hours worked per day and the timing and sequencing of hours.
Microtask Work Schedules by Platform Dependence.
p < .05. **p < .01. ***p < .001.
We measure schedule variation as the average discrepancy in a respondent’s daily schedules (see Studer et al. 2011).
Mismatches due to timing = total mismatches – mismatches due to duration (i.e., abs(preferred hours – actual hours).
Our primary analysis also indicated that platform-dependent respondents have around 3.67 more hours of preference mismatches than nondependent respondents. The supplementary calculations in the last panel of Table 3 provide more detail about the nature and causes of these preference mismatches. Platform-dependent workers are more likely to be working reluctantly (i.e., did not prefer to work but did) and more likely to experience unwanted disengagement from the platform (i.e., preferred to work but did not). This suggests that financial dependence on the platform makes workers more susceptible to the pull of the platform at inopportune times and also more likely to wish they were working when they are not. The nature of the mismatches, however, is similar for both groups. Mismatches are generated by discrepancies between the preferred and actual duration of work and to a somewhat lesser degree by discrepancies between the preferred and actual timing of work. Finally, as shown in the last row of Table 3, dependent workers have more volatile schedules than nondependent workers: On average, they have more transitions among preferred and unpreferred states (for a discussion of volatility, see Ritschard 2021). These additional statistics are consistent with the idea that dependent workers have work schedules that are driven by platform and customer demands (rather than their own preferences). This makes work schedules unstable and pulls people onto the platform when they would rather be doing something else while also making them wish they could work more (Lascău et al. 2024).
Conclusion
In theory, gig work offers high levels of schedule control compared to most conventional work by formally allowing workers to set their own hours. In practice, however, a variety of platform characteristics seem to prevent many gig workers from obtaining the work schedules they prefer. Motivated in part by the idea that different types of gig work might lead to different outcomes, we examined schedule control among microtask workers on the MTurk platform. We reasoned that schedule control might be high among people doing this type of gig work because microtasks can be done online and at any time of day and workers select the tasks they want to complete. Our analysis, however, was also guided by the platform dependence perspective, which predicts variations in workers’ experiences even on the same platform. Specifically, it suggests that gig workers who depend on the income they earn through a platform tend to have less control over their work schedules, less predictable schedules, and in turn, less desirable work schedules than other workers (Schor et al. 2020).
In many ways, our analysis of MTurk workers supports platform dependence theory. As predicted by the theory, we find that workers who are dependent on microtask income are less able than other workers to fulfill their scheduling preferences. Furthermore, our mediation tests highlight one mechanism that helps explain this connection. Financial dependence on microtask work encourages people to work at unexpected times, and these expectation mismatches often lead to preference mismatches. This finding corroborates two qualitative studies of highly active MTurk workers that suggested that for platform-dependent workers, unpredictable hours reduce schedule control. Specifically, they argued that the unpredictable supply of work, stiff competition, and lack of automatic task assignment encourages platform-dependent workers to remain on call: spending as much time as possible seeking and completing microtasks (Lascău et al. 2024; Lehdonvirta 2018). We also find that the timing of work varies in unexpected ways that often depart from workers’ preferences. Together, the evidence suggests that financial vulnerability leads to schedules that are hard to predict, hard to reconcile with other activities, and more likely to conflict with workers’ own preferences.
However, our analysis also adds a new wrinkle to the literature on schedule control among gig workers, and it highlights a possible scope condition for platform dependence theory. In our sample of MTurk workers, being platform dependent is only associated with significantly more preference mismatches among the minority of respondents who do 30 or more hours of microtasks a week. Among that group, preference mismatches accumulate with hours worked at an increasing rate among platform-dependent workers, but they cease to accumulate among nondependent workers. A detailed comparison of work schedules reveals several other differences between platform-dependent and nondependent workers that may be worth of examining in future studies. Platform-dependent workers have more variability in the number and timing of hours from one day to the next. They also report more hours when they did not prefer to work but did anyway and more hours when they preferred to be working but were not. Among respondents doing fewer than 30 hours of microtasks per week, platform dependence does not seem to be the driving force behind preference mismatches. As microtask hours increase among these workers, preference mismatches rise in a similar way regardless of whether workers are platform dependent.
These results have several theoretical implications. First, they underscore the importance of platform dependence. The literature on gig work often emphasizes the heterogeneity of gig work and the idea that worker outcomes will vary depending on the type of gig work they do (Hoang, Blank, and Quan-Haase 2020; Kalleberg and Dunn 2016). However, we find that despite the fairly unique features of microtask work, platform dependence still shapes gig worker outcomes. On average, people who need the money they earn from microtasks are less able than others to work the hours they prefer. This support for platform dependence is especially notable because our sample and analytic approach differ starkly from the analysis that led to the theory. Schor et al. (2020) developed platform dependence theory inductively using in-depth qualitative data from respondents who did different kinds of gig work but not microtask work. We build on this foundation using a quantitative analysis of survey data with detailed schedule information from MTurk workers across the United States. Our findings thus show that the theory explains variations in work schedule outcomes even in a type of gig work that is fully online and offers workers particularly high formal control over their work schedules.
Second, our analysis suggests that the theory of platform dependence may have some limits. We find that the distinction between dependent and nondependent workers was important for explaining differences in preference mismatches among MTurk workers doing more than 30 hours of microtasks per week. However, for the many respondents who spent less time doing microtasks, platform dependence was less important: Preference mismatches increased with work hours in a similar way regardless of whether respondents were platform dependent. Other studies that examined schedule control among microtask workers did not examine low-hour workers (Lascău et al. 2024; Lehdonvirta 2018). This raises an important question: Why does platform dependence not seem to matter among low-hour workers? Future studies should examine gig workers with a range of work hours to learn more about platform dependence and preference mismatches among people doing relatively few hours of gig work.
Third, our results highlight the need for more empirical and theoretical examination of the ways platforms shape worker behavior using various forms of what we will call “strategic neglect.” We found that low-hour MTurk workers who are not platform dependent also have preference mismatches, but it is not clear what mechanisms created those mismatches. In conventional jobs, high levels of schedule control can lead workers to increase work effort, creating a “flexibility paradox” that could explain schedule preference mismatches (Chung 2022; Lu, Wang, and Olsen 2023). In gig work, however, many of the factors thought to produce that paradox (e.g., pressure from bosses and coworkers, workplace culture, etc.) are absent. In gig work, “post-disciplinary control mechanisms” that rely on opportunities rather than sanctions are more relevant (Vieira 2023). Many studies of gig work, for instance, emphasize how gig platforms design and employ algorithms to shape work schedules. Platforms use algorithms to assign work and evaluate performance (Kellogg et al. 2020), set pay with “surge pricing” (van Doorn 2020), and suspend workers for not completing enough jobs (Lascău et al. 2024). These factors, however, are also unlikely to explain our findings. The MTurk platform does not use algorithms as extensively as other platforms. For instance, it relies on customers to evaluate workers and set pay, and it does not automatically suspend low-hour or inactive workers. Finally, there is some evidence that platforms can elicit work effort or “self-exploitation” in subtle ways by fostering financial precarity, an entrepreneurial spirit, and a gamified atmosphere (Vieira 2023). The MTurk system, however, is not heavily gamified, and we observed low schedule control even among workers who do not rely on platform income. Why, then, do MTurk workers still have so many preference mismatches?
MTurk seems to shape work schedules, in large part, by not taking action. The platform, for instance, does not assign workers jobs. This means workers do not have to worry about declining jobs, but it also encourages them to be constantly “on call” by leaving them to find jobs themselves (before someone else takes them) without algorithmic assistance (Lascău et al. 2024). MTurk’s lack of algorithmic intervention thus leaves workers to fend for themselves in a highly competitive, atomized labor market of MTurk’s own making. Also, the MTurk interface does not provide tools that could help workers find the best tasks quickly (Kaplan et al. 2018), track their effective hourly wages (Toxtli et al. 2021), evaluate requesters (Savage et al. 2020), or exchange useful information with each other (Yin et al. 2016). This lack of support could help explain why MTurk workers struggle to work the schedules they prefer even when they are not platform dependent. Unstructured time is inherently difficult to control (Aeon and Aguinis 2017; Flaherty 2011), and MTurk’s minimal involvement in the distribution, evaluation, and remuneration of work makes time management even harder (Lehdonvirta 2018). Some workers collaborate outside the platform to combat these problems (Kaplan et al. 2018). Still, without easy access to information about requesters or tasks, search tools, connections to other workers, or other structures to help them manage their time, many MTurk workers struggle to work the hours they prefer even if they are not platform dependent. Furthermore, although the ability to work remotely makes microtask work more flexible than in-person gig work, working from home may also increase the risk of schedule mismatches by blurring work-life boundaries (Clark 2000). In short, when workers are “set free” in this competitive, unstructured, hard-to-navigate environment, many do not work the schedules they prefer. Paradoxically, they often wish they could spend more time on a platform that they often want to leave while they are actually working. More research is needed to understand subtle, hands-off ways of shaping worker behaviors and their relative importance across platforms. Our work is consistent with the idea that the “freedom” gig workers have from regulations and employer interference can put them at greater risk of domination (Hickson 2024).
Our results are also relevant for policymakers. Gig work is known to have some serious drawbacks. It generally lacks the fringe benefits and workplace protections of conventional work arrangements (Zipperer et al. 2022). Furthermore, it is often precarious work with unstable hours and income, which can harm health and well-being (Schneider and Harknett 2019; Thomas et al. 2022; Wood and Lehdonvirta 2021). Our analysis shows that in addition, although platforms promise high levels of flexibility and schedule control (Bergvall-Kåreborn and Howcroft 2014; Lascău et al. 2022), many workers still struggle to align their work schedules with their preferences—even when doing microtask work, which entails fewer logistical constraints than in-person gig work. Our findings also indicate that preference mismatches are most common among those who both rely on the income that microtasks provide and spend many hours pursuing platform income. Broadly speaking, our results are thus consistent with the argument that gig platforms primarily serve the needs of platforms and customers rather than workers (Lascău et al. 2022). Policymakers who are considering the role platform work may play in labor markets of the future should consider these downsides of platform work and how they might exacerbate broader inequalities.
Policymakers should also consider why so many Americans do gig work despite all these flaws. The rise of gig work may reflect the efforts of gig platforms to create fertile ground for their own growth.) Baber (2024:724) argues that gig platforms engage in cultural, regulatory, and market manipulation to “extract value, exert control, and transfer employment risks and costs onto workers” and that “The guise of flexibility is marketed in specific ways while obscuring the necessary economic need to supplement low wages or lack of full employment in the primary labour market.” (Baber 2024:731). The rise of gig work may also reflect the lack of other job options available for financially disadvantaged people who need income that can be attained without specialized training or capital or without leaving the home or interfering with the demands of a conventional job. Indeed, to be effective, efforts to shape the gig economy must recognize that the appeal of gig work is rooted in the relative strengths and weaknesses of the conventional economy and the gig economy. As Schor et al. (2020) explain, the conventional economy does not provide enough jobs with high wages or flexible schedules. This creates a demand for gig jobs with (comparatively) flexible schedules that generate supplemental income. These gig jobs, however, typically lack the fringe benefits and job security that attract workers to conventional employers. Gig platforms thus essentially freeride on conventional employers who offer benefits and security (Schor et al. 2020). Meanwhile, conventional employers benefit from gig platforms that help workers make up for inadequate wages in their conventional jobs. Policymakers should thus question whether the growing popularity of gig work is rooted in the quality of the opportunities it provides. Many workers may be choosing it out of desperation.
As research on gig work schedules continues to develop, we suggest that future studies examine the relative importance of platform features that can affect workers’ ability to fulfill their schedule preferences. On any given platform, mismatches between actual and preferred schedules may reflect the nature of the work, platform design elements (e.g., tight submission deadlines; Yin et al. 2018), or various kinds of algorithmic control (Kellogg et al. 2020). They can also reflect competition among workers (Lascău et al. 2022) and the difficulties people have organizing their time when there are few constraints (Lehdonvirta 2018). The presence of such features varies even among microtask platforms (Lascău et al. 2022; Lehdonvirta 2018). Studying how variations in these features are related to work schedule mismatches would allow researchers and policymakers to better identify which reforms would best help workers get the schedules they want.
Footnotes
Appendix: Sample Composition and Data Quality
To restrict our analysis to U.S. respondents, we used MTurk location qualifications. We also used time stamps, IP addresses, and self-reported time to verify that respondents were in the United States. We used two strategies to guard against multiple submissions from a single person. MTurk prevents workers from completing HITs more than once, but because we posted our survey in a new HIT every three hours for an entire week, we had to take extra precautions. To prevent multiple responses from a single respondent, we used the Unique Turker website to generate a script that blocks existing respondents from completing our survey more than once. We also used the Qualtrics “ballot-stuffing” protections.
To protect against bots and assess data quality, we asked respondents to correctly identify the number of dots in an image and included an explicit attention check. We dropped respondents who failed either of those checks. We also dropped respondents who reported a suspiciously large and diverse profile of paid work activities. Finally, we asked respondents to report the minimum and maximum number of hours they worked in the last four weeks and their personal and family income. We dropped any cases where the minimum number of hours was greater than the maximum or personal income was greater than the family income.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Russell Sage Foundation (Grant No. 2105-32350). Publication of this article was funded in part by Purdue University Libraries Open Access Publishing Fund.
1
All these characteristics of microtask work may influence the demographics of the people who do it. In the United States, people of color are more likely than White people to do some kind of gig work (Anderson et al. 2021), and they may be especially likely to do gig transportation or delivery work (Hoang et al. 2020). The population of MTurk workers, however, seems to mirror the U.S. population fairly closely in terms of race, gender, and income (Moss et al., 2020).