Algorithmic Management Resource Model and Crowdworking Outcomes:

Mixed-methods Approach of Computational and Configurational Analysis

Mohammad Soltani Delgosha (PhD)
Associate Professor in Business Analytics, University of Birmingham, Birmingham, UK
[Link]@[Link]

Nastaran Hajiheydari (PhD)*

Senior lecturer in Digital Marketing & Analytics, Queen Mary University of London, London, UK
[Link]@[Link]
Address: Bancroft Building, 4th Floor, Queen Mary University of, Mile End Rd, London E1 4NS
Phone: +44 (0)114 215 7180

*Corresponding Author

1
Algorithmic Management Resource Model and Crowdworking Outcomes:
Mixed-methods Approach of Computational and Configurational Analysis

Abstract
The delegation of managerial functions such as job allocations, performance appraisals, and disciplining
working behaviors to automated, intelligent algorithms has transformed various aspects of workplace
dynamics. Despite the increasing prevalence of algorithmic management in today’s workplaces, its
implications for working outcomes remain underspecified. Given the contextual novelty of this
research, we adopted a mixed-methods approach to theorize an algorithmic management resource
model and investigate its configural relationships with crowdworkers’ engagement and burnout. This
was achieved by analyzing online crowdworker community narratives and subsequently developing
nuanced insights into the resources that algorithmic management offers or impedes. In phase 1, drawing
on Conservation of Resources (COR) theory tenets, we utilized computational text analysis to explore
resource gains and losses associated with algorithmic management. Then, using configurational
analysis over two studies (N=322), we identified and empirically examined the interrelationships among
resource passageways and working outcomes, specifically engagement and burnout. Our results support
a theoretical understanding of the algorithmic management resource model and shed greater light on
several configurations of algorithmic resource passageways, sufficiently explaining crowdworkers’
engagement and burnout in distributed, dispatched work settings such as online labor platforms.

Keywords: Algorithmic management; Crowdworkers; Engagement; Burnout; Computational text analysis;

fsQCA; Resource gains and losses.

2
 1. Introduction
The pervasive integration of algorithms into workplaces is radically reshaping the nature and
structure of work. In today's work environments, algorithms are not only replacing human
workers but also taking over managerial roles once exclusive to humans (Lee, 2018). Online
labor platforms (OLPs) are particularly reliant on this novel form of management to orchestrate
their day-to-day operations and manage their workforce efficiently (Möhlmann et al., 2021).
Without algorithmic management, digital platforms such as Uber, Lyft, Deliveroo, and others
would be unable to effectively manage their vast pool of crowdworkers and scale their
operations. Recent estimates show that in the US and UK alone, over 64 million people work
for OLPs (Statista, 2022).

Although automated self-learning algorithms in workforce management enhance efficiency,

scalability, and reach, they also significantly transform working dynamics, such as job design,
power structures, and work experiences (Cameron et al., 2023; Duggan et al., 2023; Gal et al.,
2020). The detached, distributed nature of OLP environments creates new working dynamics
in which automated algorithms not only mediate economic transactions but also play a central
role in governing social exchanges. From OLPs’ perspective, algorithmic management aims to
establish a highly transactional relationship with workers, compensating them for the quantity
of work undertaken rather than focusing on developing mutual trust and commitment (Duggan
et al., 2020). However, research suggests that many crowdworkers may perceive their working
relationships with OLPs as extending beyond pure economic exchange (van Zoonen et al.,
2024; Wood et al., 2019). Tensions over issues like supervision, work belonging, and
dependency (Möhlmann et al., 2021) suggest that there is more to crowdworking arrangements
than simply transactional exchanges (Ashford et al., 2018; Leicht-Deobald et al., 2019).

Extant research has investigated workers’ positive and negative experiences with algorithmic
workforce management (e.g., Benlian et al., 2022; Gal et al., 2020; Möhlmann et al., 2021).
Findings indicate that while some crowdworkers report high levels of frustration and anxiety
due to stressors such as pressure to follow app instructions, fear of negative ratings, penalties,
and lack of transparency (Cram et al., 2022; Koivusalo et al., 2024), others express satisfaction
or even preference for OLP work (Berger et al., 2019; Cropanzano et al., 2023). However, the
reasons why some crowdworkers perceive this algorithmically managed work setting
positively while others see it negatively remain ambiguous. This ambiguity is further
complicated by the high attrition rates observed among OLP workers, estimated at 12.5% per
month (MacDonald, 2019). Such mixed work outcomes underscore the need for a nuanced
understanding of the complex dynamics at play in algorithmically managed work
environments—the reasons some workers find it challenging, while others perceive it as
beneficial.

Scholars suggest that the variation in worker experiences might be attributed to the complicated
challenge of balancing freedom and control in the crowdwork setting (Cameron & Rahman,
2022). Due to the special nature of crowdworking, OLPs must respect the autonomy of
freelance workers while ensuring service quality and customer satisfaction. The multifaceted
duality of crowdwork raises questions as to whether prior theories are sufficient to understand
algorithmically managed work dynamics. In the absence of a conventional employer–employee
relationship, the OLP work setting challenges the traditional understandings of governance
(Gol et al., 2019), control (Wiener et al., 2023), and organizational support (Jabagi et al., 2019).
For instance, rather than selecting workers based on cultural fit and providing socialization and
training, OLPs simply require app downloads, registration, and basic background checks
(Rosenblat, 2018). Additionally, algorithmic job allocation reimagines the work assignment

3
process by both restricting and enabling crowdworker autonomy through selective information
provision and work suggestions (Hajiheydari & Delgosha, 2024). Crowdworkers are also
subject to constant surveillance, real-time control, and nudging (Cram et al., 2022). With its
primary focus being collecting and processing huge amounts of work behavior data,
algorithmic performance management introduces new configurations of evaluating, directing,
and disciplining workers.

By offering or limiting resources such as information or monetary rewards, algorithmic

management seeks to navigate the balance between freedom and control, which simultaneously
enables and restricts crowdworkers. For instance, while OLP algorithms may provide a steady
stream of work opportunities, they may also reduce crowdworkers’ control over task selection
and work pace. Similarly, algorithmic performance monitoring may offer clear feedback and
opportunities for improvement, but it may also increase stress and reduce perceived autonomy.
This intricate balance of enabling and restricting influences various aspects of crowdwork
conditions and outcomes. We argue that the interplay of resource provision and restriction
shapes crowdworkers’ experiences and influences work outcomes such as burnout and
engagement.

Burnout (a result of excessive resource losses) and low engagement (due to insufficient
resource gains) are well-established precursors to attrition in workplaces (Bakker et al., 2023;
Crawford et al., 2010). Research has demonstrated that burnout is a leading cause of employee
turnover, as it undermines job satisfaction and organizational commitment (Bakker &
Demerouti, 2017). Conversely, higher engagement levels have been associated with lower
attrition rates, as engaged workers are more likely to remain committed to their work and
organization (Schaufeli et al., 2006). Still, we have limited knowledge about the indicators
behind burnout and engagement and little understanding of which constructs shape these two
critical work outcomes in OLP contexts. Unpacking these outcomes is imperative for both
theory⎯by developing a model that identifies new constructs and explaining their
interdependencies and different pathways they shape burnout and engagement⎯and for
practice⎯by informing strategies to reduce attrition, improve worker well-being, and enhance
platform productivity.

Drawing on the conservation of resources (COR) theory (Halbesleben et al., 2014) and
leveraging observational and survey data, we developed and tested a theoretical framework of
an algorithmic management resource model. COR theory posits that environmental conditions
such as management practices affect resource gain and loss, with imbalances significantly
impacting work outcomes. Stress arises from resource losses or failures to gain resources. By
conceptualizing algorithmic management practices as potential sources of resource gain and
loss, we seek to investigate how these dynamics shape engagement and burnout.

Motivated by the theoretical and practical significance of the algorithmic management

phenomenon, our first research question (RQ1) asks: What are crowdworkers’ perceptions of
resource gains and losses in the context of algorithmic management? We address this question
using an inductive computational approach (Nelson, 2020) grounded in first-hand
crowdworker experiences (Study 1). We applied computational textual analysis techniques to
unpack the thematic structure of conversations between users on [Link], an online
community where crowdworkers share their work experiences, emotions, and opinions.
Observational data offer the most natural way to study social phenomena, providing an
aggregated picture of individuals’ perceptions and experiences (Humphreys & Wang, 2018).
By analyzing naturally occurring data (Müller et al., 2016), we aimed to identify representative

4
topics of algorithmic management resource gains and losses and ensure organic validity of our
findings.

Rather than focusing on single, isolated variables, COR theory suggests that resources travel
in packs or caravans (Hobfoll, 2011), which support, complement, or substitute each other
(Halbesleben et al., 2014). We thus adopted a deductive configurational approach to further
reveal insights into how the constellations of algorithmic management resource gains and
losses shape work outcomes. Studies 2 and 3 address our second research question (RQ2): “In
what ways does the interplay of perceived algorithmic management resource gains and losses
shape crowdworker burnout and engagement?” We used a configurational approach to
examine the interrelationships among algorithmic management work conditions and their
impacts on burnout and engagement. These studies suggest that crowdworkers perceive
resource caravans of algorithmic management that cannot be fully explained by linear
symmetrical models (Maier et al., 2021). Fuzzy-set qualitative comparative analysis (fsQCA),
as a case-oriented comparative analytical approach (Fiss, 2011; Park et al., 2017; Ragin, 2008),
was utilized across these two studies to uncover different configurational patterns.

This research contributes to the existing body of knowledge in two ways. First, it advances our
understanding of the nuanced ways algorithmic management can simultaneously enable and
restrict crowdworkers, delineating its dual impact on resource provision and depletion.
Leveraging a COR theoretical lens and computational methods, we provide granular,
empirically-grounded insights into crowdworkers’ perceptions of algorithmic management,
uncovering indicators of germane resource gain and loss constructs affecting burnout and
engagement of OLP workers. Second, we adapt these constructs into a survey and examine
their effects on burnout and engagement, developing a model of algorithmic management
resources that explains the reasoning behind such effects. Our configurational analysis explores
the complex interdependencies of resource gains and losses in algorithmically managed work
settings. Shifting the focus from single resources to “resource caravans,” our research sheds
light on how the interplay of resource gains and losses shape crowdworkers’ engagement and
burnout. This approach allows for more nuanced theorizing about the specific algorithmic
management resource caravans that shape crowdworker outcomes and demonstrates the
importance of resource constellations rather than single-resource analyses. Although “resource
caravans” have received some attention in IS research (Giddens et al., 2023), more is needed
to identify how resources combine or interact in digitally-mediated environments.

2. Theoretical Background
2.1. Algorithmic Management and Crowdworkers
The essence of management is to efficiently and effectively coordinate, allocate, and control
resources to achieve desired goals. As artificial intelligence advances and data infrastructures
become more robust, organizations are increasingly delegating managerial tasks⎯such as
resource allocation, performance management, or operational optimization⎯ to automated
self-learning algorithms. These intelligent agents are now capable of hiring staff, assigning
tasks, measuring productivity, appraising performance, and even selecting workers for layoffs.
This emergent type of management is reshaping conventional work settings (Duggan et al.,
2020). On the positive side, assigning managerial functions to algorithms increases decision-
making efficiency, optimizes resource distribution, and drives economic value (Brynjolfsson
& McAfee, 2014; Kellogg et al., 2020; Lindebaum et al., 2020). Algorithms powered by
computational big data analysis surpass human decision-making abilities, enhancing precision
in actions, predictions, and resource allocation. However, recent research suggests that

5
algorithmic decisions are sometimes perceived as inauthentic or even unethical (Jago, 2019),
which can lead to resistance among workers (Dietvorst et al., 2018). With the growing use of
algorithmic management, scholars are increasingly focused on examining its effects on
workplace dynamics, including working relationships, job design, and employee well-being
(Duggan et al., 2020; Kellogg et al., 2020; Parker & Grote, 2022). For instance, Curchod et al.
(2020) argue that algorithms are far from neutral machines; they are designed primarily to serve
the interests of business owners rather than to empower workers.

Algorithmic management is particularly prevalent in digital platforms, where automated

algorithms perform most managerial functions. This approach is crucial for OLPs due to the
distributed nature of crowdworking (Jabagi et al., 2019), large amounts of market data
(Möhlmann et al., 2021), and the need to effectively match millions of supplies and demands
(Delgosha & Hajiheydari, 2020). Algorithms embedded within OLP applications match
crowdworkers with consumers, assign jobs, monitor and evaluate performances, and regulate
transactions. By leveraging algorithmic management, OLPs can scale their operations,
coordinating activities of large, disaggregated crowdworkers and using big data to optimize
operations and desired outcomes. However, employing automated algorithms for workforce
management has reshaped various aspects of working arrangements, including rewards and
recognition, job attitudes, and power structures. Although crowdworkers are increasingly
attracted to OLPs for their flexibility and income potential (Fu et al., 2022; Rani et al., 2021),
labor activists, worker unions, and academics have raised concerns regarding the consequences
of algorithmic management. They caution that it may lead to unfair decisions (Fieseler et al.,
2019), lack of support (Kuhn & Maleki, 2017), reduced interpersonal communication (Anicich,
2022; Goods et al., 2019) and power imbalances (Cutolo & Kenney, 2021).

Because algorithmic management is a relatively new phenomenon, the IS literature on the

subject is limited but growing. As summarized in Appendix A, this limited research has
discussed various algorithmic mechanisms and how they are commercially effective or
essential. Scholars have noted that platforms are required to implement some directives which
may lead to information and power asymmetry (e.g., Cheng & Foley, 2019; Jarrahi &
Sutherland, 2019; Jarrahi et al., 2021), excessive governance and control mechanisms (e.g.,
Basukie et al., 2023; Gol, 2019; Huang, 2023; Möhlmann et al., 2021, Noponen et al., 2023),
as well as mechanical interactions (e.g., Stark & Pais, 2020). Other studies have examined how
workers make sense of and deal with algorithmic managerial functions (e.g., Gal et al., 2020;
Möhlmann et al., 2023). Still others have studied the potential negative health impacts of
algorithmic management on workers (e.g., Vignola et al., 2023).

While this research sheds light on different aspects of algorithms in the workplace, the primary
reasons for the variation in crowdworkers’ experiences under algorithmic management remain
underexplored. Specifically, the factors influencing why some workers perceive algorithmic
management positively—experiencing it as empowering—while others find it frustrating,
stressful, or alienating have not been thoroughly examined. Our research seeks to address this
gap by exploring how algorithmic management practices contribute to resource gains and
losses and how those, in turn, shape burnout and engagement among crowdworkers. By
applying COR theory, we aim to uncover the underlying mechanisms driving divergent work
experiences in algorithmically managed work environments.

2.2. Conservation of Resources Theory

COR theory offers an overarching framework to explore the effects of algorithmic management
on work outcomes such as engagement and burnout. This motivational theory suggests that

6
individuals are driven to acquire and conserve resources essential for their well-being and goal
attainment (Hobfoll et al., 2018). As per COR theory, individuals always seek to protect,
obtain, foster, and retain resources. Halbesleben et al. (2014) define resources as objects, states,
or conditions perceived as valuable for achieving one’s goals. For example, workers may seek
personal resources such as autonomy, social resources such as relationships, and material
resources such as financial rewards. COR theory suggests that resource loss can lead to stress
and burnout (Crawford et al., 2010; Menguc et al., 2013) and resource gain can enhance
employees’ well-being (Demerouti et al., 2017; Huettermann & Bruch, 2019). COR theory also
posits that situational conditions, working as passageways, can either facilitate or hinder
individuals’ abilities to gain or conserve resources (Hobfoll et al., 2018).

IS researchers have used COR theory to analyze the effects of new workplace technologies,
examining how they can accelerate working pace (Chen et al., 2009), increase job demands
(Ward & Steptoe-Warren, 2014), and create a risk of technology-induced overload (Harris et
al., 2015). From the COR perspective, the benefit and cost functions of algorithmic
management (i.e., those with positive and negative directionalities of influence on working
outcomes) can be captured as resource gains and losses. Here, resource gain refers to the
acquisition of resources through automated algorithms that may create value and help
crowdworkers achieve their goals. Conversely, algorithmic management may also lead
to resource loss, heightening cognitive load and depleting energy. For instance, crowdworkers
managed by algorithms might experience resource gains (autonomy and financial rewards) or
resources losses (unclear procedures or lack of power). Resource loss has primarily been
studied in the form of burnout, which refers to “a syndrome of chronic exhaustion, a cynical,
negative attitude regarding work, and reduced professional efficacy” (Bakker & Demerouti,
2017; p. 273).

Algorithmic management thus serves as an important passageway, either providing resources

to meet job demands or depleting them and creating job strain. Our study uses a COR
theoretical lens to capture both resource-gain and resource-loss passageways of algorithmic
management. This framework allows us to investigate crowdworkers’ working outcomes under
algorithmic management from both gain and loss perspectives. Rather than focusing narrowly
on specific consequences of algorithmic management, our work takes a bottom-up approach
by analyzing discussions within an online crowdworker community. This approach enables us
to capture a diverse range of opinions and concerns regarding this emerging form of
management, thereby providing a comprehensive view of algorithmic management’s impacts
on crowdworkers’ experiences.

3. Mixed-Methods Design
Given the contextual novelty of our research, we adopted a mixed-methods approach to explore
the implications of algorithmic management and develop rich insights into its impacts on
working outcomes. Although prior research has examined resource changes due to technology
and their effects on users’ well-being, the specific resources associated with algorithmic
management and its gain and loss passageways are still insufficiently explored.

Our mixed-methods approach comprised two primary strands, methodically designed to serve
a developmental purpose. Initially, we employed computational analysis to develop an
algorithmic management resource model, which was then subjected to further investigation
through configurational analysis. From a timing standpoint, this research design follows an
exploratory sequential mixed-method approach (Creswell & Clark, 2017), involving the
collection of quantitative data to empirically examine and elaborate on the findings uncovered

7
through our textual data analysis. In the first stage, an inductive computational exploration was
performed on digital trace data obtained directly from crowdworkers to reveal resource gains
and losses resulting from algorithmic management. The computational analysis strand relied
on narratives sourced from [Link], an online community originally launched for Uber
drivers to share their experiences, opinions, and feelings. With over 150 thousand members,
the forums contain contributions from Uber drivers, Lyft drivers, and delivery drivers working
with services such as DoorDash, Uber Eats, and Amazon Flex. We analyzed this rich source
of textual data, comprising more than 5 million posts, to identify resource gains and losses of
algorithmic management based on crowdworkers’ first-hand experience, perceptions, and
discussions. In the next stage of our empirical analysis, we designed two studies using
configurational analysis to explore and examine the interrelationships between algorithmic
management’s resource passageways and crowdworkers’ work outcomes, including burnout
and engagement.

Venkatesh et al. (2013; 2016) urge researchers to evaluate the appropriateness of mixed-
methods research design by considering the requirements of the research questions and the
purpose and context of the study. Our current understanding of crowdworking settings and,
more importantly, the implications of algorithmic management is still in its infancy (e.g.,
Möhlmann et al., 2021; Tarafdar et al., 2022). There is, therefore, a need to develop novel
theoretical perspectives by integrating findings from qualitative and quantitative methods,
triangulating data and overcoming the limitations of each perspective (Creswell & Clark, 2017;
Venkatesh et al., 2013). The employment of a mixed-methods approach enriches our
understanding of algorithmic management implications in crowdworking context. This
involves first exploring the firsthand experiences of workers predominantly managed by
automatic algorithms and, secondly, examining the implications of this emerging management
on work outcomes. Specifically, using opinion mining and topic modeling as computational
text analysis techniques, this research uncovers resource-loss and resource-gain passageways
tied to algorithmic management. Furthermore, utilizing fsQCA as a configurational analysis
method, this research seeks to gain insights into how these resource passageways drive
crowdworkers’ work outcomes.

The configurational approach employed in Studies 2 and 3 enables us to formulate new

theoretical propositions for explaining and analyzing nonlinear, bottom-up emergent relations.
Considering recent research—which recognizes that work conditions have synergistic, not
isolated, effects on worker outcomes (Parker et al., 2017)—applying configurational models,
rather than a linear approach, can better elucidate the mutual relations shaping crowdworker
burnout and engagement. In keeping with Hobfoll et al.’s (2018, p. 107) position that
“resources do not exist individually, but travel in packs, or caravans,” we adopt a
configurational approach to examine the interplay of algorithmic management-related resource
gains and losses. This approach allows us to thoroughly capture how specific configurations
(i.e., caravans) of work conditions lead to crowdworkers’ burnout or engagement. Figure 1
illustrates the research design of this study.

8
 Figure 1. Mixed-Methods Research Design

4. Study 1: Exploring the Algorithmic Management Resource Model

from Crowdworkers’ Narratives

4.1. Data Collection and Preprocessing

In the first stage, we based our computational analysis on textual narratives generated by
crowdworkers in their online community. We used a combination of opinion mining and topic
modeling to uncover the thematic structure of discussions. Our primary data source for the first
study was [Link], a well-known independent online community for crowdworkers with
more than 150K members. User-generated data scraped from online communities or forums
are increasingly used by scholars to examine various economic, social, and organizational
phenomena (Berente et al., 2019). The availability of large-scale digital trace data in online
forums provides an invaluable opportunity to access detailed behavioral insights with
unprecedented scale and depth.

In our context, crowdworkers on [Link] post content and interact with each other to
seek and offer advice. Our textual dataset was built using a web crawler to extract 5.8 million
forum posts from April 2014 (beginning of [Link]) to March 2020. Most of the posts
are from Uber and Lyft drivers or delivery riders who work for DoorDash, Uber Eats, or
Amazon Flex. This dataset represents a broad spectrum of crowdworkers’ perspectives, as
these users frequently share experiences and seek advice within this community. The organic
nature of these discussions makes it an ideal source for capturing the authentic first-hand
experience of working under algorithmic management on OLPs. By analyzing these
interactions, we gain valuable insights into the real-world impacts of algorithmic management
on crowdworkers.

4.2. Data Analysis: Opinion Mining and Topic Modeling

We applied two machine learning methods—XLNet (Yang et al., 2019) and Structural Topic
Modeling (STM) (Roberts et al., 2019)—to analyze the crowdworkers’ attitudes and opinions
toward algorithmic management. The sentiment-based topic extraction process allowed us to
explore resource gains and losses resulting from algorithmic management. This text-mining

9
process tapped into the crowdworkers’ conversations to understand their feelings and
experiences, which were expressed in unstructured textual data associated with algorithmic
management.

We first used XLNet, a deep learning technique, to capture crowdworkers’ attitudes as

expressed in their posts. XLNet is a contextualized language representation learning method
founded on a generalized permutation language modeling objective. With self-attention
mechanisms and bidirectional learning as its backbone model, XLNet produces more accurate
representations and achieves improved prediction accuracy (Yang et al., 2019), which makes
it an appropriate method for analyzing our textual data. Details on the implementation of
XLNet are provided in Appendix B.

Given our large dataset, we then employed topic modeling to identify the thematic content in
user posts and extract topics related to the resource passageways of algorithmic management.
Topic modeling is an unsupervised computational technique that generates semantic categories
based on meaning and commonalities among words in a large collection of textual data—a task
otherwise impractical with large empirical datasets (Brown et al., 2020). We selected STM for
its capability to incorporate metadata into textual analysis (Roberts et al., 2019), allowing us to
extract topics grounded in crowdworkers’ optimistic and pessimistic attitudes. Previous studies
have demonstrated STM’s superiority over the commonly employed latent Dirichlet allocation
method in terms of both predictive performance and qualitative interpretation (Karanović et
al., 2021; Roberts et al., 2019). Details on our topic modeling method and results are provided
in Appendix C.

4.3. Coding and Interpretation of Topics

To interpret the empirical themes of each topic, we identified the top words associated with
each topic using two weighting algorithms: highest probability (which weighs words using raw
per-topic probabilities) and FREX (which calculates word frequency and exclusivity). Two
authors and a research assistant independently reviewed the top 10 words from both weighting
methods, along with the top 20 posts most closely associated with each topic, to assign
descriptive labels representing the resource gains and losses of algorithmic management.
Labels were assigned to each topic with a name that both summarized and accounted for that
topic (Charmaz, 2014). Primarily, labels were based on top probability terms in each topic and,
secondly, the top related corpora/texts that represent that topic.

The labels applied during this initial coding phase encompassed various aspects of
crowdworkers’ jobs, such as “unclear routines,” “automated messaging,” “false accusation,”
and “traffic information,” all of which directly relate to algorithmic management. Whenever
possible, we used comparative methods, more specifically within-source comparisons
(Charmaz, 2014) comparing data (posts) from one part of a forum category and thread with
another one. Consolidating and comparing the individual results indicated that there was
generally high similarity in the labels assigned to each topic, with an intercoder agreement of
75 percent. Conflicts in labelling between researchers were managed by discussing
discrepancies to reach a consensus on topic labels. Topics that none of the researchers identified
as relevant to the research question (such as “tipping,” “car conditions,” and “insurance”) were
marked as not meaningful and excluded from the results. Appendix C provides a detailed
overview of extracted topics, including examples of posts related to each topic, highly
associated words, the topic labels, and their Rank-1 metric.

10
In the second step, we used an iterative process to interpret the extracted topics by extensively
examining the original data. Our interpretation focused on identifying an algorithmic
management resource model. We used a grounded theory coding scheme (Corbin & Strauss,
1990) and clustered first-order codes (topics) into axial second-order themes that were then
further abstracted via selective coding (aggregated dimensions). After identifying relevant
topics, we performed axial coding to develop theoretical themes by combining common topics
and abstracting them into sub-categories. During this step, we iterated a process to evaluate the
data and existing theory, combining topics into themes (Charmaz, 2014; Gioia et al., 2013).
We compared our emerging computational codes (topics) with extant literature, noting
similarities and distinctions between topics informed by research on COR theory and
algorithmic management. For example, we found that the topics of unpredictable fares and
unstable commission fees have connections. They can be clustered as reward uncertainty.
Similarly, intrusive rating system, scrutinizing cancelation rates, and nitpicking boss can be
grouped as invasive supervising. Finally, we conducted selective coding by further abstracting
second-order themes into aggregated theoretical dimensions.

Drawing on a COR theoretical lens, we clustered our themes to develop algorithmic

management passageways that generate resources for crowdworkers (i.e., algorithmic job
allocation and support) or deplete resources (i.e., algorithmic ambiguity, unfairness, constant
monitoring, and mechanistic communication). Following Gioia et al.’s (2013) template, Table
1 presents the results of our coding process, including first-order codes (topics), second-order
themes, and aggregated dimensions. For more details, please refer to Appendix D.

Table 1. Thematic Structure of Algorithmic Management Resource Model

First-order codes (topics) Second-order themes Aggregated dimensions +/-
Unclear routines
Procedural inconsistency
Unforeseen app changes Algorithmic ambiguity
Unpredictable fare
Reward uncertainty
Unstable commission fees
Unfair surge pricing
Unfair compensation
Unacceptable fare

Resource Loss Passageway

False accusation Algorithmic unfairness
Unfair punishment
Account deactivation/suspension
Long waiting time
Unfair workload
Job misallocation
Automated messaging
Ineffective communication Detached relationship Algorithmic miscommunication
Insufficient explanation
Neglecting drivers’ concerns
Lack of voice
Unresponsiveness
App monitoring
Overcontrolling Ongoing surveillance
Privacy concern Algorithmic constant monitoring
Intrusive rating system
Scrutinizing cancelation rate Invasive supervising
Nitpicking boss
Time flexibility
Work flexibility
Location flexibility Algorithmic assignment
Resource Gain
Passageway

Job availability
Providing gigs
Task allocation
Traffic information
Navigation advice Information provision
Performance feedback Algorithmic support
Extra pay Reward provision

11
 Income

4.4. Results
The topic modeling analysis indicated that 42.7 percent of crowdworkers’ posts primarily
addressed issues related to resource loss (labeled as pessimistic posts), while 23.7 percent
focused on resource gain (labeled as optimistic posts). Our analysis identified six key thematic
clusters that encapsulate these topics. These six aggregate dimensions effectively grouped
topics at a broader level, with four dimensions representing resource losses and two
representing resource gains resulting from algorithmic management (see Appendix D).

4.4.1. Resource Loss Passageway

Algorithmic ambiguity
Topics referring to algorithmic ambiguity contained two second-order themes—“Procedural
inconstancy” and “reward uncertainty.” These accounted for 8.6% of online community posts,
covering four out of 30 relevant topics in our analysis. In their narratives, crowdworkers
expressed concerns about unclear routines, emphasizing the uncertainty and opacity of the
algorithmic system. They expressed sentiments such as, “There probably is a whole lot that
goes into the algorithm. But there’s no way for us to know for sure exactly what goes into it
and what's currently being used.” Additionally, they highlighted dissatisfaction with sudden,
unexpected, and unnecessary changes in the application, stating, “My app was working just fine
until the latest update. Why do those tech geeks at Uber insist on fixing something that ain’t
broken?” This theme suggests that the procedures or routines dictated by the algorithmic
system are not stable or predictable. Workers may feel that the algorithmic processes and the
application guiding their tasks are subject to frequent changes, making it challenging to
understand or anticipate the specific steps or rules governing their work.

In addition to process inconsistency, crowdworkers expressed discontent with changes in

payment structure and fares, citing concerns about the transparency and predictability of the
algorithmic reward system. For example, one asked, “Since Uber seems to adjust fares without
notification to the driver, how does everyone monitor their statements?” This lack of
transparency and clarity in the reward system can lead to confusion and frustration among
workers, as they may find it challenging to predict the outcomes of their efforts or understand
the factors influencing their compensation. According to the role theory (Kahn et al., 1964),
experiencing ambiguity in responsibilities and procedures leads to confusion, stress, and job
dissatisfaction. When workers are uncertain about the procedures and what is expected of them,
they may consume more cognitive and emotional resources (Lee & Ashforth, 1996). This
increased effort can result in resource depletion, leaving workers with fewer resources to cope
with other work-related challenges or personal demands (e.g., O’Driscoll & Beehr, 1994;
Schmidt et al., 2014). In the context of algorithmic management, we, therefore, discuss that
uncertainty or a lack of clarity about how their income is calculated or how the system works
is a significant stressor with negative implications for crowdworkers’ cognitive resources.

Algorithmic unfairness
Topics related to algorithmic unfairness, comprising three second-order themes and six topics,
emerged as the predominant focus within the entire corpus, representing 16% of community
posts. We clustered discussions on unfair surge pricing and low earnings into the second-order
theme of “unfair compensation.” Conversations on false accusations and account deactivation/
suspension were grouped into the theme of “unfair punishment,” and topics addressing long
waiting times and job misallocation were labeled under the theme of “unfair workload.” Given
12
that algorithms widely govern compensation and other working conditions, crowdworkers
continuously “hold [a] magnifying glass up to justice issues” (Colquitt & Zipay, 2015, p. 82)
to assess the fairness of algorithmic decision making.

Unfair compensation was the topic of 8.3% of crowdworkers’ discussions in their online forum.
Unfair compensation is a crucial concern as it directly impacts the financial well-being of
crowdworkers, potentially leading to dissatisfaction and a sense of exploitation. They
complained, “Why does Uber make so much more than we do on these rides? Not at all fair
paymen[t] practice on Uber’s part!” A portion of community posts (4.9%) also argued that in
some cases the algorithm punishes them unfairly or accuses them falsely. For example, they
share bitter experiences of being punished, such as, “I was permanently deactivated from Uber,
apparently some customer complained to them that I tried to ask for sexual favors when I was
dropping them. (Before anyone asks no I didn’t do any such thing).” Finally, under the same
category of unfairness, 2.8% of posts are related to unfair workload, with complaints about
long waiting times or being sent to the wrong job or pick up point. For example, “As I write
this it’s been well over 3 hours and have yet to get a ping… I think Lyft does not like me.”
Others complained about the algorithm’s allocation: “Long pick up, no PAX no fee: Saturday
night I accept a long pick up. 22 min and about 16 miles. I get to the location, timer starts the
countdown ....”

From a social exchange perspective (Wayne et al., 1997), researchers maintain that fair
procedures and distribution signal that employees are valued, while unfairness communicates
devaluation that can be devastating for individuals (Howard & Cordes, 2010). In this context,
employees may internalize blame or perceive themselves as less worthy of their job. Coping
with perceived unfairness demands cognitive and emotional resources, leading to a gradual
depletion of energy for employees (Dishon-Berkovits, 2018). Similarly, in OLPs,
crowdworkers seeing no clear and fair relationship between their effort and the allocation of
rewards may initially invest time, effort, and resources to alleviate the situation. However, if
unfairness persists, a substantial amount of mental energy would be devoted to processing and
responding to perceived injustice, resulting in reduced feelings of personal accomplishment,
exhaustion, and energy depletion

Algorithmic miscommunication
Conversations on algorithmic miscommunication constituted 8.9% of community posts. We
grouped five topics into two second-order themes: “detached relationship” and “lack of voice.”
Through topic modeling, we uncovered that automated messaging, ineffective communication,
insufficient explanations, neglecting drivers’ concerns, and unsupportiveness were
communication issues that crowdworkers experienced in working with algorithms. In platform
work settings, crowdworkers typically lack the opportunity to build a personal relationship with
a supervisor. They may not have colleagues to socialize with or a team to be a part of. In
addition, unlike conventional work settings—where open communication exists between
workers and their supervisors, as well as among co-workers—OLPs primarily rely on
algorithmic determinations to optimize efficiency, resulting in predetermined interactions
(Walsh, 2020).

Pre-programmed and algorithm-generated messages lack personalization, human touches, or

nuanced responses. This is frustrating for crowdworkers. In their narratives, crowdworkers
criticized how the algorithm contributes to a sense of disconnection between the workers and
the platform, saying, “I didn't even receive their usual ‘request received’ robomail. I then sent
a test email, and I instantly got a robo response. Sent original once more, then nothing.”

13
Similarly, one worker noted that “The only complaint I actually do have is that it is very
frustrating that you can only communicate with Uber via email….”

Moreover, companies overlooking or not adequately addressing the issues and feedback raised
by crowdworkers, in other words, lack of assistance or encouragement from the platform, was
a significant topic of crowdworker discussion. They shared that “So twice now I have asked
that a supervisor/manager contact me to discuss this trip. And twice now I have received an
email from a first level support person asking me to call the Helpline to discuss the problem(s).
ARGH!!!” The perception of being undervalued leads to heightened stress or frustration,
leading in turn to decreased morale and job satisfaction.

Algorithmic constant monitoring

Six topics comprising 9.2% of the community posts were related to algorithmic constant
monitoring. We categorized app monitoring, overcontrolling, and privacy concerns under the
theme of “ongoing surveillance,” and grouped the topics of the intrusive rating system, the
scrutinization of cancelation rates, and nitpicking bosses into the “invasive supervising”
second-order theme. These topics highlight the frustrations expressed by crowdworkers in their
posts, particularly regarding how the algorithm constantly monitors them and their behavior.
“They threatened to deactivate me if my cancellation rate rises. The funny part is that I always
keep it below 15% over 2 years of driving. They want me to keep it below 2%, which is absolute
bullcrap!!” Similarly, they shared experiences on how tightly the app monitors the rating
scores; for example, one user mentioned that “Anyone [else] notice the newest bullshit about
rider feedback issues. ALMOST every day I drive now I get some kind of rider feedback issue...
Latest was ‘late arrival missed eta’….” The fact that the platform has access to their
information also raises privacy concerns, as exemplified in this sample post: “Where is the
privacy for the drivers, pax can know literally everything there is to know about us as uber give
them all the tools to do so.”

Unlike traditional close supervision, OLP algorithms largely monitor crowdworkers

continuously, intrusively, and without warning. By collecting real-time data on workers’
performances, algorithms seek to control working behaviors. This can be interpreted as an
electronic panopticon (Prassl, 2018) used by platforms to invasively monitor crowdworkers,
building working conditions like a permanent probation period (Gramano, 2020). Such
invasive supervision and monitoring undermines employees’ autonomy and control over their
work, depleting their psychological resources and contributing to stress (George, 1996).
Algorithmic constant monitoring also raises the possibility of resource loss related to feeling
uncertainty and fear about being punished for not working well enough.

4.4.2. Resource Gain Passageway

Algorithmic assignment
Four of the 30 relevant topics in our analysis refer to algorithmic assignment, accounting for
9.2% of online community posts. We grouped time flexibility and location flexibility topics
into the second-order theme of “work flexibility.” Job availability and task allocation were
clustered under the “providing gigs” label. One of the key advantages of algorithmic
management in contemporary work settings such as OLPs is the ability to afford crowdworkers
a high degree of flexibility in choosing where and when to work. For example, a sample post
points to the time flexibility that the algorithm grants to drivers: “Today is supposed to be my
day off, but the app is open and if I get a good ride I'll hop on it. Love the flexibility, love how
I can decide to make $50 or so today to take the pressure off of tomorrow as I'll be up watching

14
football tonight….” Posters also appreciated the availability of jobs and task allocation by the
algorithm in narratives such as, “Ever since my rating gone up to 4.73, I been getting lots of
lux ride for my pool. Today I just got off a new Lexus NX on pool.” By dynamically assigning
jobs based on real-time demand and worker availability, OLPs’ algorithms allow workers to
adjust their work hours and availability based on their preferences or changing circumstances.
Workers can choose to work more during busy periods or reduce their workload during low-
demand periods, providing flexibility in terms of time and location.

Algorithmic support
Making up 14.5% of the community narratives, five topics refer to algorithmic support. The
second-order theme “information provision” consists of traffic information, navigation advice,
and performance feedback. Extra pay and income topics have been grouped into the “reward
provision” theme, accounting for 7.4% of the posts. The algorithms offer real-time data on task
availability, demand patterns, and performance metrics that support crowdworkers in
effectively managing their work endeavors. Crowdworkers appreciate receiving live
information, saying “I finally created Excel worksheets with the info that Uber provides.” More
frequently, workers talked about their payment, reimbursement, and reward provision from the
platforms in various posts: “It’s real. Caught a few 1.5x and a couple 1.2x. With Lyft's earnings,
I walked away after about 5-6 hours with $140.” “I got a 1.8 with Uber today but when I
checked the passenger pays 12 I got $21 Uber paid the difference.” “I'm happy with my
earnings this week, but even happier I got my $235 for 55 rides! Best bonus yet.” Automated
algorithms manage rewards and incentives based on workers’ performances. By rewarding
crowdworkers, OLPs incentivize desirable behaviors, motivating workers to excel and
increasing their engagement.

4.5. Discussion
Study 1 explores how crowdworkers perceive the resource dynamics of algorithmic
management. By analyzing crowdworkers’ first-hand experiences as shared in an online
community, we revealed how algorithmic management both enables and constrains
crowdworkers by providing or limiting resources. On the one hand, by employing automated
algorithms, OLPs offer resource gains such as flexibility, informational and reward support,
and efficient task allocation. On the other hand, resource losses—including procedural
ambiguity, unfairness, miscommunication, and constant monitoring—are prominent
challenges. These findings uncover the underlying work conditions in OLP work settings. The
insights from this study informed the development of the algorithmic management resource
model and provided a foundation for exploring how these resource dynamics shape key work
outcomes like burnout and engagement.

5. Study 2: Configurations of Crowdworker Burnout

Study 2 addresses RQ2 by specifically focusing on how algorithmic management–related
resource gains and losses collectively affect crowdworkers’ experiences of burnout. Research
grounded in COR theory demonstrates that resource loss and lack of resource gain both directly
contribute to worker burnout (Crawford et al., 2010). When workers face significant resource
losses—such as job insecurity, financial setbacks, or lack of support—they experience
increased job strain. These resource losses deplete their overall resource pool and create a
deficit, making it challenging to meet job demands (Halbesleben et al., 2014). The imbalance
between resource gains and losses can lead to chronic exhaustion and eventual burnout
(Hobfoll et al., 2018). In examining how these algorithmic management passageways

15
contribute to burnout, we adopted a configurational approach (Misangyi et al., 2017; Park et
al., 2017) and employed fsQCA (Ragin, 2008).

5.1. Participants and Procedure

We chose a parallel sampling approach, in which “samples for the quantitative and qualitative
components of the study are different but drawn from the same underlying population”
(Venkatesh et al., 2016, p. 446). For Study 2, the primary research population consisted of
crowdworkers from various OLPs. Using Amazon’s Mechanical Turk (MTurk), we recruited
participants working for Uber, Deliveroo, Lyft, Uber Eats, Just Eat or any other ride-hailing or
delivery platforms. MTurk has been recognized for its ability to produce high-quality survey
data through the use of attention checks (Berinsky et al., 2012) and reputation scores (Peer et
al., 2014). To ensure data quality, we limited participation to individuals with an approval
rating above 95% and excluded those who failed an attention check (“Please select the
‘Disagree’ button if you believe your answers are correct”; 12 cases) or did not work on OLPs
(5 cases), resulting in a final sample size of 185 (Mage = 38.6, SDage=12.7; 59.4% male).

5.2. Survey Instrument and Measures

We identified and adapted appropriate scales from previous studies to develop our survey
instrument, which was designed to evaluate algorithmic management resource passageways.
Appendix E lists the final survey items. To further contextualize and validate our instrument,
we consulted five experts in job design (three academics and two industry professionals) and
conducted a pre-test with 17 crowdworkers. Feedback from both the experts and pre-test
respondents led to minor adjustments in terminology and item sequencing to improve clarity.

To measure algorithmic job allocation, we used the results from the topic modeling and
subsequent qualitative analysis (first– and second-order coding) in Study 1, considering the
perceived autonomy associated with managing work through algorithms. We adopted the four-
item perceived autonomy scale from Morgeson and Humphrey (2006) and modified it based
on the context of this study. Our topic modeling identified two primary themes for algorithmic
support: information provision and reward provision. For information provision, we used
information sharing scales by Mohr and Spekman (1994); to measure reward provision, we
used the financial benefit scale from Bock et al. (2005). To measure algorithmic ambiguity, we
used the procedural justice construct (Colquitt, 2001) and reversed its related five items.
Algorithmic unfairness was measured based on distributive justice scale (Colquitt, 2001). The
concept of perceived availability of social support in the workplace (Fernet et al., 2013) was
reversed and used to measure algorithmic miscommunication. A five-item scale by Jensen and
Raver (2012) on perceived surveillance was used to measure algorithmic constant monitoring.
Finally, to measure burnout, we applied a four-item scale from Little et al. (2007) on a seven-
point scale (never=1, always=7). Our assessment of the measurement model after testing
content, convergent, and discriminant validity indicates that the measures are reliable and valid
(see Appendix F). To mitigate the potential for common method bias (CMB), we followed the
recommendations outlined by Podsakoff et al. (2003) to apply certain procedural remedies and
explicitly tested for the presence of CMB, as detailed in Appendix G.

5.3. Configurational Analysis

fsQCA, a case-oriented method based on set theory, allows us to analyze within-case
relationships and explore complex interdependencies and asymmetric relationships between
conditions and outcomes. Using a configurational perspective, we aimed to identify equifinal
and asymmetric links among algorithmic management-elated resource losses and gains (which
served as our conditional variables) and work outcomes. Configurations are equifinal because

16
alternative combinations of resource passageways may be producing the same outcome, and
links are asymmetric because the presence of resource losses and gains, as well as their absence,
may create different configurations resulting in the same outcome (Furnari et al., 2021;
Misangyi et al., 2017). In Study 2, the conditions included resource-loss and resource-gain
passageways, with burnout as the outcome of interest.

When conducting fsQCA, we initially calibrated all interval scale values into fuzzy-set
membership scores. The fsQCA calibration process involves choosing threshold levels where
the respondent’s score is judged to be “fully in” membership or “fully out” of membership of
the set (Soltani Delgosha, et al., 2024). We calibrated the data by using a common rule for all
conditions, setting the fully-in and fully-out cutoff points at +/- one standard deviation (SD)
from the median, with the point of maximum ambiguity set at the median value of each variable
(Douglas et al., 2020). Following this process, the resulting fuzzy membership score is between
0 and 1, where 0 shows a fully-out membership, 1 shows fully-in membership, and 0.5
represents the most ambiguous situation (neither fully in nor fully out).

fsQCA analyzes the necessary and sufficient conditions to identify which configurations of
conditions lead to an outcome. The necessity analysis revealed that algorithmic unfairness is
essential for burnout, as its raw consistency score exceeds the 0.90 threshold (Greckhamer et
al., 2018). In addition, to avoid trivial necessary conditions (type 1 error), we ensured that the
coverage score (0.82) and the relevance of necessity score (0.73) of the necessity condition
were higher than 0.60. Next, sufficiency analysis was conducted using the fsQCA Quine-
McClusky truth table as a Boolean chart to detect all logically possible combinations of
conditions found in the dataset. Following the recommendations of Greckhamer et al. (2018),
we set the raw consistency cutoff at 0.80, the frequency cutoff at 2 (to retain more than 80% of
the cases), and proportional reduction in inconsistency (PRI) at 0.60.

5.4. Results
The configurations driving crowdworker burnout are shown in Table 2. The fsQCA results
yield five configurations associated with high burnout and one configuration for low burnout.

High burnout. The overall solution coverage of 0.78 and solution consistency of 0.93 reflects
high empirical relevance and consistency among the five high-burnout configurations. Each
configuration characterizes a class of crowdworkers experiencing significant exhaustion.
Among the configurations, HB5 holds the highest empirical relevance, with a row coverage of
0.38. This configuration depicts crowdworkers as perceiving algorithmic management as

17
straining, characterized by OLPs behaving unfairly, interacting mechanistically, constantly
monitoring, and enforcing ambiguous procedures. Crowdworkers in HB5 would experience
burnout regardless of resource gains.

HB1 represents crowdworkers facing challenges of algorithmic ambiguity and unfairness,

expressing dissatisfaction with job allocation and support provided by algorithmic
management. The HB2 configuration denotes crowdworkers facing relational challenges;
although they perceive a high level of autonomy, the lack of support within a mechanistic and
unfair relationship structure contributes to their burnout. Crowdworkers in the HB3
configuration feel exhausted, primarily due to algorithmic unfairness. These workers believe
that the algorithmic assignment and support fail to meet their expectations, while also feeling
that compensation, workloads, or penalties determined by algorithms are not fair.
Configuration HB4 is largely similar to HB2, reemphasizing that when algorithmic
management is not able to provide adequate support, mechanistic and unfair relationships can
lead crowdworkers to burnout.

Low burnout. The fsQCA results indicate only one configuration associated with low burnout,
suggesting an asymmetric relationship between algorithmic management-related resource
passageways and work outcomes. This finding implies that the adverse effects of resource loss
outweigh the benefits of resource gains within algorithmic management. Configuration LB1
demonstrates that when algorithmic management is not ambiguous, effective algorithmic job
allocation can mitigate the negative impacts of algorithmic miscommunication and constant
monitoring, ultimately reducing job strain. In other words, appropriate job assignment helps
crowdworkers to cope with the demands of overmonitoring and miscommunication, thereby
reducing the likelihood of burnout.

5.5. Discussion
The findings from Study 2 deepen our understanding of how algorithmic management resource
caravans drive burnout among crowdworkers. Using fsQCA, we identified five distinct
configurations of resource losses and gains that lead to burnout. Notably, algorithmic
unfairness emerged as a core condition present in all burnout pathways, underscoring its
salience as a stressor in OLP environments. This finding indicates that many crowdworkers
perceive algorithmic management as unfair and a tool to make them subservient to the
platforms. The analysis also reveals that the absence of algorithmic support amplifies the
negative effects of resource losses, indicating that support mechanisms can buffer against some
resource losses. However, the asymmetric nature of the relationships highlights that not all
resource gains can compensate for resource losses, particularly when unfairness,
miscommunication, and invasive supervision are perceived. The pathways illustrated in Figure
2 correspond to distinct algorithmic management resource gain and loss caravans lead to
crowdworker burnout.

18
 Figure 2. Configural Algorithmic Resource Model to Crowdworkers’ Burnout

Note: representing the condition presence in the configuration; indicating the

condition absence in the configuration.

6. Study 3: Pathways to Crowdworker Engagement

Organizational and psychological studies show that when work environments provide valuable
resources, they can motivate workers to dedicate their efforts to accomplish tasks and achieve
work goals (Bakker et al., 2017). When jobs provide resources that satisfy basic needs or
support goal achievement, workers have a positive work-related mindset that drives
engagement. Lin et al. (2019) suggest that work engagement is the function of resource gains
and losses. Gains such as recognition, rewards, or accomplishment enhance workers’ resource
pool, giving them a sense of progress and growth. Engaging in activities that lead to resource
gains can be highly motivating, as they contribute to the expansion and improvement of an
individual’s resource reservoir. Engaged workers thus proactively attempt to meet their job
requirements and take initiatives that go beyond what is expected or required of them. In
contrast, resource loss leads to a sense of helplessness, disappointment, and reduced control
over one’s circumstances, which in turn can undermine motivation and engagement.

Recent research recognizes that engagement and burnout are not simply opposing outcomes
(e.g., Demerouti et al., 2017; Maslach & Leiter, 2008). For instance, Ong and Johnson (2023)
found that configurations of job conditions shaping engagement and burnout are asymmetrical.
That is, the conditions that promote engagement are distinct from those that contribute to lack
of burnout. Study 3 aims to investigate how algorithmic management-related resource gains
and losses affect crowdworker engagement. In addressing this query, we used fsQCA to
identify configurations of resource passageways that produce engagement.

6.1. Participants and Procedure

For Study 3, we approached the community of contributors to [Link]. As noted above,
most members of this community work with the Uber, Lyft, DoorDash, Uber Eats, and Amazon
Flex platforms. In February 2021, we posted a temporary announcement under the community
category and stories threads. In this post, we provided information about our research and
explained that we would analyze anonymized personal descriptions of participants’ working
conditions from their own perspectives. We further clarified that their participation in this study
is voluntary, results will be used only for research purposes, and that our study was not related
to any business and not organized by platform firms. To motivate participation, we committed

19
to providing a £5 electronic gift card to those who answer the questionnaire completely and
accurately. After three weeks and three re-announcements of the post, we collected 175
questionnaires. We excluded the respondents that did not work for a platform (6 cases). We
also removed responses that indicated inappropriate response behavior, including unreasonable
time for completion, i.e., any handling time less than 5 minutes (11 cases) or failing at least
one attention-trap question (21 cases). Ultimately, for the third study we used 137 appropriately
completed responses for our configurational analysis. The final sample group ranged from 18
to 68 years of age (Mage = 41.7, SDage = 13.2) and was 62.04% male.

6.2. Survey Instrument and Measures

All resource-gain and resource-loss passageway conditions were measured using the same
scales as Study 2. To measure engagement, we used the nine-item scale from Schaufeli et al.
(2006). Similar to Study 2, we tested the measurement model and convergent and discriminant
validity of constructs. Table F.3 in Appendix F presents the measurement test results.

6.3. Results
High engagement. fsQCA analysis revealed three configurations associated with high
engagement among crowdworkers (see Table 3). Configuration HE1, which has the highest
coverage (0.29), represents the most prevalent group of engaged crowdworkers. They display
vigor, absorption, and dedication when all resource loss conditions are absent and resource
gains are at high levels. In HE2, crowdworkers are satisfied with the algorithmic support and
perceive algorithmic management as neither unfair nor overly mechanistic. While they express
discontent with constant monitoring and overcontrolling, the support provided by the platform
enhances their engagement. In the third configuration (HE3), despite opaque procedures and
miscommunications from OLPs, the crowdworkers do not view algorithmic management as
unfair. They benefit from algorithmic job assignment and support, resulting in an enhanced
sense of engagement.

All three of the configurations leading to crowdworker engagement demonstrate consistency

scores above the threshold of 0.8, assuring that each identified path reliably results in our
outcome of interest. The overall solution consistency (0.91) guarantees that these three
configurations are consistently linked to the engagement. However, the overall solution
coverage for engagement configurations (0.42) is notably lower than that for burnout (0.78),
corroborating our findings from Study 1. This implies that crowdworkers perceive algorithmic
management as more straining than engaging. Furthermore, the presence of algorithmic support

20
across all engagement-promoting configurations underlines its salient role in fostering
crowdworker engagement.

Low engagement. fsQCA identifies two configurations associated with low engagement. Both
configurations show high satisfaction with the flexibility and autonomy offered by algorithmic
management. However, in Configuration LE1, miscommunication poses a barrier to
engagement, limiting the effectiveness and frequency of workers’ interactions with OLPs.
Configuration LE2 indicates that a lack of adequate support—despite the flexibility provided
by algorithmic management—diminishes crowdworkers’ engagement.

6.4. Discussion
Study 3 extends our exploration of how algorithmic management resource gains and losses
shape crowdworker engagement. Figure 3 shows three algorithmic management resource
caravans that enhance engagement, with high levels of algorithmic support and fair
management practices emerging as pivotal enablers. These findings indicate the importance of
providing reward resources and information, making fair decisions, as well as offering flexible
and efficient job allocation in order to foster engagement. Interestingly, even in the presence
of resource losses like constant monitoring, ambiguous procedures, or miscommunication,
engagement can be sustained when resource gains, especially algorithmic support, are present.
This study highlights the dynamic interdependencies of resource gains and losses in promoting
engagement and suggests that platform firms should prioritize supportive and fair practices to
promote positive worker experiences. Together with the findings from Study 2, this study
illuminates the various pathways through which algorithmic management influences burnout
and engagement.

Figure 3. Configural Algorithmic Resource Model of Crowdworkers’ Engagement

Note: representing the condition presence in the configuration; indicating the

condition absence in the configuration.

7. General Discussion and Post-hoc Analysis of Findings

7.1. Discussion
We used the findings from Studies 1-3 to draw inferences, which were then integrated into the
development of meta-inferences for the algorithmic management resource model (see Table 4
for a summary of inferences and meta-inferences). Based on these meta-inferences, we
conclude with six propositions to explain and predict the impacts of algorithmic management
on work outcomes.

In OLPs, algorithms function as the primary decision-makers, playing a pivotal role millions
of times a day by determining job assignments, pay rates, penalties, and more. Conceivably,

21
crowdworkers constantly assess the fairness of these algorithmic decision-making practices,
particularly concerning outcomes that directly impact them. Analysis of crowdworkers’
narratives reveals that 16% of community posts note perceived injustices in compensation (e.g.,
low fares and surge pricing issues), penalty enforcement (e.g., false accusations and account
deactivation/suspension), and workload allocation (e.g., long waiting times and job
misallocations). Mounting evidence suggests that perceived unfairness is a significant job
stressor and a potential driver of job strain. As per the relational model of authority, individuals
care deeply about how authorities treat them and continuously assess the neutrality, integrity,
and trustworthiness of those authorities’ decisions (Blader & Tyler, 2015). When
organizational procedures and interpersonal treatment are perceived as unjust, employees are
more likely to experience frustration and stress. For crowdworkers, unfair decisions also
amplify uncertainty about future earnings and job security. Due to their inability to appeal or
influence these decisions, (Choudary, 2018), crowdworkers often feel constrained by
algorithmic directives (Deng et al., 2016), which could undermine their sense of dignity and
professional identity (Jabagi et al., 2024). Fieseler et al. (2019) argue that the lack of bargaining
power and low social standing contribute to inadequate application of justice principles in
algorithmic decisions. This perception of unfairness can lead to decreased job satisfaction and
increased attrition rates. Our fsQCA results further underscore the critical role of algorithmic
unfairness, which is present in all burnout configurations but absent in those leading to
engagement. We therefore suggest the following:

Proposition 1. Perceived unfairness of algorithmic management practices regarding the

compensation system, punishment mechanisms, and workload allocation is a salient
work condition contributing to burnout.

Consistent with the “coping hypothesis” (Van Woerkom et al., 2016), we found that specific
resource gains promote coping with resource depletion. Our findings reveal that algorithmic
support and positively perceived algorithmic job assignment can help offset the negative
impacts of ambiguity and miscommunication. Specifically, when algorithms support
crowdworkers by providing rewards, information, and flexible job allocations, crowdworkers
may better cope with ambiguous or inconsistent procedures, helping them stay engaged. This
occurs because algorithmic support can act as one of the compensatory resources that help
workers navigate uncertainty and maintain their sense of control. When algorithms provide
rewards and information, they create predictable pathways for success, reducing the cognitive
load and anxiety associated with ambiguous procedures (Gonzalez-Mulé et al., 2021). Further,
flexible job allocations give workers agency in managing their work, which research shows
can buffer against stress from procedural uncertainty (Maslach & Leiter, 2008). Job autonomy
is primarily a cognitive resource, providing discretion to the individuals to process and apply
information in their work (de Jonge & Dormann, 2006). As a job resource, autonomy allows
workers to prioritize tasks and align their responsibilities with their skills and preferences,
fostering a sense of control and reducing job strain. Likewise, our findings reveal that high
levels of perceived autonomy and support allow crowdworkers to tolerate the detached
relationships and lack of voice. We thus propose the following:

Proposition 2. Algorithmic support and assignment may facilitate coping with

algorithmic ambiguity and miscommunications, thereby enhancing crowdworker
engagement.

Approximately 9.2% of community posts include crowdworkers venting about being

micromanaged by algorithms. They express feeling pressured by the constant, real-time

22
tracking of their performance and behavior. Crowdworkers share stories of dealing with
ongoing surveillance, including app tracking, overcontrolling, and invasions of their privacy.
Many criticize this invasive supervision style, describing algorithms as nitpicking bosses that
scrutinize even minor actions and impose strict penalties for low rating scores and cancellation
rates. Such feelings of being under constant monitoring diminish engagement, lower morale,
and damage trust (Carlson et al., 2017; Holland et al., 2015). The demanding nature of
algorithmic control restricts crowdworkers’ autonomy and increases work intensity through a
combination of behavioral nudges and constant surveillance (Duggan et al., 2020; Wood et al.,
2019). Further, 8.9% of forum posts highlight crowdworkers’ frustrations with ineffective and
unsupportive communication from OLPs. Many report that their concerns are overlooked and
that they have little opportunity to voice their opinions or receive meaningful responses. This
lack of responsiveness may create a sense of disconnection between workers and the platform,
reinforcing the perception that they are merely managed by impersonal algorithms rather than
being recognized as valued contributors. Ultimately, this absence of meaningful
communication heightened stress and exacerbates job strain. However, our configurational
analysis revealed that the relationships between algorithmic constant monitoring, algorithmic
miscommunication and burnout is not symmetrical; the effects of overmonitoring and
miscommunication on burnout depend on interactions with other conditions. Specifically,
algorithmic constant monitoring, when coupled with algorithmic unfairness and
miscommunications, will produce crowdworker burnout. Similarly, algorithmic
miscommunication, in conjunction with perceived unfairness, may also lead to burnout. These
findings suggest that the impacts of these two conditions are contingent upon the broader
resource environment in which crowdworkers operate. Therefore, we propose the following:

Proposition 3. The impacts of algorithmic constant monitoring and miscommunication

on crowdwork burnout is asymmetrical, with burnout emerging not from overmonitoring
or miscommunication alone, but from its interaction with other conditions.

In line with the “buffering hypothesis,” our results suggest that algorithmic assignment can,
by providing flexibility and autonomy, alleviate the impact of some resource losses on
exhaustion (Bakker et al., 2005). Prior research has demonstrated that the associations
between resource depletion and burnout are less pronounced when resource gains function
as buffers against resource losses (Bakker & Demerouti, 2017). As a job resource, perceived
autonomy promotes a sense of control over working conditions and fosters intrinsic
motivation (Deci & Ryan, 2004). Providing employees with control is an important job
resource that can buffer against the adverse effects of some limitations, such as workload
(Bakker et al., 2005; Brauchli et al., 2014). In the context of OLPs, crowdworkers can better
tolerate close supervision and interactions with non-human agents when the algorithm
provides work flexibility and guarantees the availability of jobs. Based on the fsQCA results
and established psychological theories, we propose that algorithmic assignment can serve
as a buffer. This perspective offers a practical understanding of how positive aspects of
algorithmic management can mitigate potential negative effects. Therefore, we suggest the
following:

Proposition 4. Algorithmic assignment can buffer the impacts of algorithmic constant

monitoring and algorithmic miscommunication, thereby reducing crowdworker burnout.

23
Table 4. Inferences from Studies 1-3 and Development of Meta-inferences

Study 1 inferences Study 2 inferences Study 3 inferences Meta-inferences

Unclear routines + Unforeseen app changes + Present in burnout (HB1, Absent in HE1, present in Algorithmic ambiguity is a salient resource-loss condition for burnout
Unpredictable fees + Unstable commission HB5), but not in lack of HE3 for engagement, but outcome.
fees = Resource losses, perceived as burnout “don’t care” in lack of
“algorithmic ambiguity” engagement Algorithmic support and assignment can jointly mitigate the negative
impact of algorithmic ambiguity on crowdwork engagement.
Unfair surge pricing + Unacceptable fares + A necessary condition for Absent in engagement (HE1- Algorithmic unfairness is an essential resource-loss condition for burnout
False accusations + Unreasonable account burnout (HB1-HB5), but HE3) but “don’t care” in lack and absent in all configurations for engagement.
deactivations + Long waiting times + Job “don’t care” in lack of of engagement.
Findings indicate critical importance of ensuring fairness in compensation,
misallocations = Resource losses, perceived burnout
as “algorithmic unfairness” workload distribution, and penalty assessment
Automated messaging + ineffective Present in burnout (HB2, Absent in engagement (HE1, Algorithmic miscommunication has asymmetrical relationship with
communication + insufficient explanations + HB4, HB5) and present in the HE2) and present in lack of burnout, functions as a supplementary condition, and its impact contingent
neglecting workers’ concerns + lack of burnout engagement (LE1, LE2) on the other conditions.
unresponsiveness = Resource losses,
Both algorithmic assignment and support are required for crowdwork
perceived as “algorithmic
miscommunication” engagement when algorithmic miscommunication is present.
App monitoring + Overcontrolling + Present in burnout (HB2, Absent in engagement (HE1) Algorithmic constant monitoring has asymmetrical relationship with
Privacy concerns + Intrusive rating system HB5) and absent in lack of and present lack of burnout, functions as a supplementary condition, and its impact contingent
Scrutinizing cancelation rates + Nitpicking = burnout engagement (LE2) on the other conditions.
Resource losses, perceived as “algorithmic
Algorithmic support may facilitate coping with algorithmic constant
constant monitoring”
monitoring and promote engagement.
Time flexibility + Location flexibility + Absent in (HB1, HB3), Present in engagement (HE1, Algorithmic assignment may mitigate the expected associations between
Job availability + Task allocation = present in (HB2, HB4), and HE3) and present in lack of burnout and algorithmic constant monitoring as well as algorithmic
Resource gains, perceived as “algorithmic don’t care condition in HB5 engagement (LE1, LE2) miscommunication.
assignment” for burnout, present in lack of
Findings reveal that algorithmic assignment is not sufficient condition for
burnout
engagement.
Traffic information + Navigation advice + Absent in burnout (HB1- Present in engagement (HE1- Algorithmic support’s presence is a critical condition for engagement, and
Performance feedback + Extra pay + Income HB4) and ‘don’t care’ in lack HE3) and ‘don’t care’ for lack its absence can largely contribute to burnout.
= Resource gains, perceived as “algorithmic of burnout of engagement
support” Algorithmic support alone can mitigate the adverse effects of algorithmic
constant monitoring.
Note: “don't care" conditions refer to situations where certain conditions in a configuration are considered irrelevant to the outcome.

24
Our findings further reveal that although algorithmic micromanaging can increase job strain,
providing informational support (e.g., performance feedback) and reward support (e.g., income)
can facilitate coping with the negativity associated with micromanagement. Prior research has
found that excessive monitoring signals to workers that managers want to restrict their
autonomy (Long & Sitkin, 2018). They perceive supervisor monitoring as something that limits
their flexibility and ability to act on their own decisions (George & Zhou, 2001). Excessive and
invasive monitoring can be seen as a manifestation of micromanagement that takes away
autonomy from workers (Alvesson & Sveningsson, 2003), which has been shown to be
associated with increased stress levels; workers may feel pressured to always meet their
manager’s expectations. Nevertheless, a number of studies have demonstrated that
organizational support facilitates coping with job stressors (e.g., Ganster et al., 1986;
Demerouti et al., 2001). Organizational support, by offering tangible and cognitive resources
such as information or financial rewards, helps employees address specific stressors. As per
the organizational support theory (Kurtessis et al., 2017), workers with high perceived
organizational support engage in increased job-related effort, resulting in enhanced in-role job
performance. Accordingly, we suggest the following:

Proposition 5. When algorithmic management constantly monitors crowdworkers,

algorithmic support may facilitate coping with this resource loss and enhance
engagement.

There has been an expectation in past research that resource gains might buffer against resource
losses (Bakker & Demerouti, 2017). Conversely, our findings suggest that when crowdworkers
experience unfair treatment in terms of reimbursement, job allocation, or punishment, along
with a lack of voice (i.e., OLPs neglecting their concerns) and a detached relationship
(characterized by automated messaging, ineffective communication, and insufficient
explanations), work flexibility alone is not enough to prevent frustration. Analyzing fsQCA
paths to burnout indicates that the autonomy resulting from algorithmic job assignment does
not shield crowdworkers from the negative impacts of unfairness and mechanistic interactions.
While algorithmic assignment creates a sense of flexibility, it cannot compensate for deeper
relational and procedural injustices. This occurs because crowdworkers’ psychological needs
for fairness, respect, and meaningful communication supersede instrumental flexibility. When
algorithmic management fails to address these fundamental needs, workers might experience
high level of stress and frustration that flexibility cannot mitigate. Our results indicate that
supportive mechanisms (information and reward provision) are essential to preventing
crowdworker exhaustion. We thus propose the following:

Proposition 6. When algorithmic management is unfair and unsupportive and

communications are mechanistic, crowdworkers are likely to experience burnout, despite
the flexibility provided by such algorithms.

7.2. Conclusion, Implications, and Limitations

Across three studies, we extended our understanding of algorithmic management implications

and offered insights into the realities of OLPs’ work settings. Utilizing a COR theoretical lens
and employing computational and configurational analysis, our findings unearth both which
resource gains and losses emerge from algorithmic management and how they jointly influence
work outcomes.

25
Our work responds to scholarly calls for further empirical investigation into the conditions
under which algorithmic management becomes either more enabling or more restraining
(Meijerink & Bondarouk, 2023). Taking a bottom-up computational approach, we gained fine-
grained insights into crowdworkers’ opinions and perceptions of algorithmic management and
its associated resource gains and losses. While prior research has noted the complex balance of
freedom and control in OLPs (Cameron & Rahman, 2022)—where algorithmic management
simultaneously enables and restrains crowdworkers—our research took a step toward a better
understanding of how this dual complexity of algorithmic management reshapes work
conditions and outcomes. Leveraging a COR theoretical lens, this work theorizes and
operationalizes an algorithmic management resource model to explore the resource dynamics
emerging from this form of management.

One contribution of this model is to explain why some crowdworkers have more positive work
experiences and continued engagement, compared to others who experience high levels of
stress and frustration. Specifically, Study 1 offers valuable insights into how algorithmic
management influences work conditions by uncovering the resource gains and losses that result
from this emerging form of management. Our findings indicate that algorithmic management
can cause frustration by generating various recourse losses, including procedural
inconsistencies; reward uncertainties; unfair compensation, workload, and punishment;
detached relationships; lack of voice; ongoing surveillance; and invasive supervising. At the
same time, it can enable crowdworkers by providing resource gains such as flexible work
schedules and locations, efficient gig allocations, and informational and reward support.

In Studies 2 and 3, we adapted identified resource gain and loss constructs and examined their
mutual impacts on key work outcomes, including engagement and burnout. Utilizing a
configurational approach, we explored the algorithmic management resource caravans that
shape crowdworkers’ engagement and burnout in OLP work settings. Shifting attention from
linear, net-effect variables to configurations of variables, we contributed to algorithmic
management scholarship by theorizing and empirically examining complex relationships
between algorithmic management and work outcomes. fsQCA analysis revealed five distinct
configurations that explain/predict how resource gains and losses contribute to crowdworkers’
burnout—and three configurations that lead to their engagement. The extracted configurations
resemble resource caravans: patterns of interdependencies between resource losses and gains
in terms of contingencies, complementarities, and substitutions. These findings contribute to
COR theory by demonstrating how algorithmic management creates resource caravans where
gains and losses are interlinked. Our findings show that unlike traditional work environments,
where resource caravans develop primarily through human interactions, algorithmic
management can simultaneously trigger multiple resource gains or losses by providing or
restricting resources. This advances our understanding of how resource caravans form and
propagate in algorithmically managed work environments.

Third, our results confirm “the primacy of resource loss” principle of COR theory in the OLP
work setting. We found that the presence of all four resource losses increases crowdworker
burnout regardless of resource gains. This conclusion is supported by the high raw coverage of
burnout configurations, indicating that high levels of algorithmic management resource losses
lead to burnout. Additionally, in line with organizational justice theory (Colquitt & Zipay,
2015), our work shows that perceiving algorithmic decisions as unfair might lead to
crowdworker burnout. However, this mechanism is contingent on the level of support offered
by algorithmic management. fsQCA results indicate a notable absence of algorithmic support
within burnout configurations, suggesting that algorithmic support may facilitate coping with

26
unfairness. Additionally, by comparing configurations that lead to burnout or lack of burnout,
we found that the flexibility and autonomy offered by algorithmic job allocation is an effective
buffer against the adverse effects of algorithmic micromanagement and miscommunications,
thereby inhibiting exhaustion. Nonetheless, it is interesting to note that when crowdworkers
perceive algorithmic management as unfair, especially in conjunction with miscommunication
or overmonitoring, the resource gains derived from algorithmic job assignment may not
sufficiently alleviate the resulting psychological costs, leading crowdworkers to experience
burnout. These findings are important because while prior research in conventional work
settings suggests that job autonomy can buffer the adverse effects of resource loss and mitigate
job strain (Bakker et al., 2005; Brauchli et al., 2014), our results indicate that in algorithmic
management, the autonomy and flexibility alone appear insufficient to cope with resource
losses.

fsQCA analysis further shows that high level of algorithmic support coupled with fair
management are salient conditions for enhancing crowdworkers’ engagement. Importantly,
this finding underscores once more the interdependence between algorithmic management
support and unfairness conditions in shaping work outcomes. We also found that, even in the
face of micromanagement and invasive supervision, crowdworkers can still experience
engagement when algorithmic management provides a high level of support. However, the
presence of both resource gains, i.e., algorithmic assignment and support, is imperative to
effectively address the challenges associated with the ambiguity of algorithmic procedures and
miscommunications. Next, our findings establish complex asymmetrical relationships between
algorithmic management resource passageways and work outcomes by showing that some
conditions are present in one configuration and absent in another. Notably, we observed an
asymmetrical relationship between algorithmic constant monitoring and burnout; this resource
loss may be present in one configuration driving crowdworker burnout but absent in another.
Our fsQCA results confirm that burnout and engagement are not opposite outcomes (Ong &
Johnson, 2023). More importantly, they reveal that algorithmic management resource loss
passageways hold greater significance than resource gain passageways because the path to
crowdworker burnout is wider. Several different configurations can shape exhaustion similarly;
its overall solution coverage is far greater than engagement, 0.78 vs. 0.42 respectively.

From a practical perspective, our findings highlight key implications pertaining to algorithmic
management that platform firms should take into account. Transitioning from human managers
to algorithmic bosses entails particular types of resource provision and depletion that impact
the crowdworking experience. Our findings indicate that optimizing the balance between
resource gains and losses plays a fundamental role in determining important work outcomes,
including burnout or engagement. Because crowdworkers are particularly susceptible to
frustration, platforms should strive to recognize when crowdworkers are feeling exhausted and
develop strategies to alleviate such perceptions, potentially by reinforcing algorithmic
management resource gain passageways. By ensuring that algorithmic decision-making
processes and criteria for job allocation, performance monitoring, and remuneration are clearly
communicated to crowdworkers, OLPs can promote transparency and prevent confusion.
Human-in-the-loop design, especially when dealing with high-stakes scenarios such as
suspending or deactivating crowdworkers, can be effective in mitigating workers’ perceptions
that the algorithm errs and lacks consideration (Burton et al., 2020). It would also be beneficial
to establish human-centered interactions with crowdworkers in order to address their sense of
isolation. Furthermore, platform firms can establish open and transparent lines of
communication, encouraging two-way interactions, active listening, and feedback mechanisms
to ensure that crowdworkers’ concerns and ideas are heard and addressed appropriately. They

27
can also facilitate access to counseling services or coaching programs (e.g., professional
guidance on how to resolve issues with job allocation, compensation, performance appraisal,
and more). By doing so, managers can create a more supportive and effective work
environment, ultimately benefiting both the crowdworkers and the platform itself.

Finally, some limitations of our research warrant consideration. First, while our analysis draws
on both naturally occurring data from an online community and study survey responses, these
sources may not capture the full diversity of crowdworker experiences across different
platforms or regions, potentially limiting the generalizability of our findings. Second, the mixed
outcomes observed in crowdworking suggest the potential presence of individual
idiosyncrasies and/or contextual moderators that shape the relationship between algorithmic
management and worker experiences. Further research should examine the person-centered
and contextual factors contributing to crowdworker exhaustion, and potential interventions or
safeguards for algorithmic management challenges. Third, the focus in the current study on
specific work outcomes, such as burnout and engagement, may overlook other important
variables, such as job satisfaction or turnover intentions, which could provide additional
insights into the nuanced impacts of algorithmic management. Lastly, although our
configurational approach offers a deeper understanding of the interplay between resource gains
and losses, it may not fully capture the temporal dynamics or long-term impacts of algorithmic
management practices, as our data were cross-sectional rather than longitudinal. Future
research might benefit from longitudinal studies and cross-platform analyses to further validate
and expand upon our findings.

References
Alshahrani, A., Ghaffari, M., Amirizirtol, K., & Liu, X. (2020, July). Identifying optimism and pessimism in
twitter messages using xlnet and deep consensus. In 2020 International Joint Conference on Neural
Networks (IJCNN) (pp. 1-8). IEEE.
Alvesson, M., & Sveningsson, S. (2003). Good visions, bad micro-management and ugly ambiguity:
Contradictions of (non-) leadership in a knowledge-intensive organization. Organization studies, 24(6),
961-988.
Anicich, E. M. (2022). Flexing and floundering in the on-demand economy: Narrative identity construction under
algorithmic management. Organizational Behavior and Human Decision Processes, 169, 104138.
Ashford, S. J., Caza, B. B., & Reid, E. M. (2018). From surviving to thriving in the gig economy: A research
agenda for individuals in the new world of work. Research in Organizational Behavior, 38, 23-41.
Avey, J. B., Reichard, R. J., Luthans, F., & Mhatre, K. H. (2011). Meta‐analysis of the impact of positive
psychological capital on employee attitudes, behaviors, and performance. Human resource development
quarterly, 22(2), 127-152.]
Bakker, A. B., Demerouti, E., & Euwema, M. C. (2005). Job resources buffer the impact of job demands on
burnout. Journal of occupational health psychology, 10(2), 170.
Bakker, A. B., & Demerouti, E. (2007). The job demands-resources model: State of the art. Journal of managerial
psychology, 22(3), 309-328.
Bakker, A. B., & Demerouti, E. (2017). Job demands–resources theory: Taking stock and looking
forward. Journal of occupational health psychology, 22(3), 273.
Bakker, A. B., Demerouti, E., & Sanz-Vergel, A. (2023). Job demands–resources theory: Ten years later. Annual
Review of Organizational Psychology and Organizational Behavior, 10, 25-53.
Benlian, A., Wiener, M., Cram, W. A., Krasnova, H., Maedche, A., Möhlmann, M., ... & Remus, U. (2022).
Algorithmic management: bright and dark sides, practical implications, and research opportunities. Business
& Information Systems Engineering, 64(6), 825-839.
Berente, N., Seidel, S., & Safadi, H. (2019). Research commentary—data-driven computationally intensive theory
development. Information systems research, 30(1), 50-64.
Berinsky, A. J., Huber, G. A., & Lenz, G. S. (2012). Evaluating online labor markets for experimental research:
Amazon. com's Mechanical Turk. Political analysis, 20(3), 351-368.
Blader, Steven L., and Tom R. Tyler, 'Relational Models of Procedural Justice', in Russell S. Cropanzano, and
Maureen L. Ambrose (eds), The Oxford Handbook of Justice in the Workplace, Oxford University Press, 2015, p.
351–370.

28
Brauchli, R., Bauer, G. F., & Hämmig, O. (2014). Job autonomy buffers the impact of work–life conflict on
organizational outcomes. Swiss Journal of Psychology.
Brown, N. C., Crowley, R. M., & Elliott, W. B. (2020). What are you saying? Using topic to detect financial
misreporting. Journal of Accounting Research, 58(1), 237-291.
Brynjolfsson, E., & McAfee, A. (2014). The second machine age: Work, progress, and prosperity in a time of
brilliant technologies. WW Norton & Company.
Burton, J. W., Stein, M. K., & Jensen, T. B. (2020). A systematic review of algorithm aversion in augmented
decision making. Journal of behavioral decision making, 33(2), 220-239.
Cameron, L., Lamers, L., Leicht-Deobald, U., Lutz, C., Meijerink, J., & Möhlmann, M. (2023). Algorithmic
management: Its implications for information systems research. Communications of the Association for
Information Systems, 52, 556-574.
Cameron, L., & Rahman, H. (2022). Expanding the locus of resistance: Understanding the co-constitution of
control and resistance in the gig economy. Organization Science, 33(1), 38-58.
Carlson, J. R., Carlson, D. S., Zivnuska, S., Harris, R. B., & Harris, K. J. (2017). Applying the job demands
resources model to understand technology as a predictor of turnover intentions. Computers in Human
Behavior, 77, 317-325.
Cennamo, C. (2021). Competing in digital markets: A platform-based perspective. Academy of Management
Perspectives, 35(2), 265-291.
Charmaz, K. (2014). Constructing grounded theory. Sage
Chen, S., Westman, M., & Eden, D. (2009). Impact of enhanced resources on anticipatory stress and adjustment
to new information technology: a field-experimental test of conservation of resources theory. Journal of
Occupational Health Psychology, 14(3), 219.
Choudary, S. P. (2018). The architecture of digital labour platforms: Policy recommendations on platform design
for worker well-being. ILO Future of work Research paper series, 3.
Colquitt, J. A. (2001). On the dimensionality of organizational justice: a construct validation of a measure. Journal
of applied psychology, 86(3), 386.
Colquitt, J. A., & Zipay, K. P. (2015). Justice, fairness, and employee reactions. Annu. Rev. Organ. Psychol.
Organ. Behav., 2(1), 75-99.
Cram, W. A., Wiener, M., Tarafdar, M., & Benlian, A. (2022). Examining the impact of algorithmic control on
Uber drivers’ technostress. Journal of management information systems, 39(2), 426-453.
Crawford, E. R., LePine, J. A., & Rich, B. L. (2010). Linking job demands and resources to employee engagement
and burnout: a theoretical extension and meta-analytic test. Journal of applied psychology, 95(5), 834.
Creswell, J. W., & Clark, V. L. P. (2017). Designing and conducting mixed methods research. Sage publications.
Croidieu, G., & Kim, P. H. (2018). Labor of love: Amateurs and lay-expertise legitimation in the early US radio
field. Administrative Science Quarterly, 63(1), 1-42.
Cropanzano, R., Keplinger, K., Lambert, B. K., Caza, B., & Ashford, S. J. (2023). The organizational psychology
of gig work: An integrative conceptual review. Journal of Applied Psychology, 108(3), 492.
Curchod, C., Patriotta, G., Cohen, L., & Neysen, N. (2020). Working for an algorithm: Power asymmetries and
agency in online work settings. Administrative Science Quarterly, 65(3), 644-676.
Cutolo, D., & Kenney, M. (2021). Platform-dependent entrepreneurs: Power asymmetries, risks, and strategies in
the platform economy. Academy of Management Perspectives, 35(4), 584-605.
Deci, E. L., & Ryan, R. M. (Eds.). (2004). Handbook of self-determination research. University Rochester Press.
De Jonge, J., & Dormann, C. (2006). Stressors, resources, and strain at work: a longitudinal test of the triple-
match principle. Journal of applied psychology, 91(6), 1359.
Delgosha, M. S., & Hajiheydari, N. (2020). On-demand service platforms pro/anti adoption cognition: Examining
the context-specific reasons. Journal of Business Research, 121, 180-194
Demerouti, E., Bakker, A. B., Nachreiner, F., & Schaufeli, W. B. (2001). The job demands-resources model of
burnout. Journal of Applied psychology, 86(3), 499.
Demerouti, E., Van den Heuvel, M., Xanthopoulou, D., Dubbelt, L., & Gordon, H. J. (2017). Job resources as
contributors to wellbeing. In The Routledge companion to wellbeing at work (pp. 269-283). Routledge.
Deng, X. N., & Joshi, K. D. (2016). Why individuals participate in micro-task crowdsourcing work environment:
Revealing crowdworkers' perceptions. Journal of the Association for Information Systems, 17(10), 648.
Deng, X., Joshi, K. D., & Galliers, R. D. (2016). The Duality of empowerment and marginalization in microtask
crowdsourcing. MIS quarterly, 40(2), 279-302.
Dietvorst, B. J., Simmons, J. P., & Massey, C. (2018). Overcoming algorithm aversion: People will use imperfect
algorithms if they can (even slightly) modify them. Management Science, 64(3), 1155-1170.
Dishon-Berkovits, M. (2018). The role of organizational justice and stress in predicting job burnout. Journal of
Career Development, 45(5), 411-424.
Douglas, E. J., Shepherd, D. A., & Prentice, C. (2020). Using fuzzy-set qualitative comparative analysis for a
finer-grained understanding of entrepreneurship. Journal of Business Venturing, 35(1), 105970.

29
Duggan, J., Carbery, R., McDonnell, A., & Sherman, U. (2023). Algorithmic HRM control in the gig economy:
The app‐worker perspective. Human Resource Management, 62(6), 883-899.
Duggan, J., Sherman, U., Carbery, R., & McDonnell, A. (2020). Algorithmic management and app‐work in the
gig economy: A research agenda for employment relations and HRM. Human Resource Management
Journal, 30(1), 144-132.
Durward, D., Blohm, I., & Leimeister, J. M. (2020). The nature of crowd work and its effects on individuals’ work
perception. Journal of Management Information Systems, 37(1), 66-95.
Fernet, C., Austin, S., Trépanier, S. G., & Dussault, M. (2013). How do job characteristics contribute to burnout?
Exploring the distinct mediating roles of perceived autonomy, competence, and relatedness. European
Journal of Work and Organizational Psychology, 22(2), 123-137.
Fieseler, C., Bucher, E., & Hoffmann, C. P. (2019). Unfairness by design? The perceived fairness of digital labor
on crowdworking platforms. Journal of Business Ethics, 156(4), 987-1005.
Fu, Y., Feng, J., & Ye, Q. (2022). Skill Spanning in the Online Labor Market: A Double-Edged Sword?. Journal
of the Association for Information Systems, 23(3), 750-778.
Furnari, S., Crilly, D., Misangyi, V. F., Greckhamer, T., Fiss, P. C., & Aguilera, R. V. (2021). Capturing causal
complexity: Heuristics for configurational theorizing. Academy of Management Review, 46(4), 778-799.
Gal, U., Jensen, T. B., & Stein, M. K. (2020). Breaking the vicious cycle of algorithmic management: A virtue
ethics approach to people analytics. Information and Organization, 30(2), 100301.
Ganster, D. C., Fusilier, M. R., & Mayes, B. T. (1986). Role of social support in the experience of stress at work.
Journal of applied psychology, 71(1), 102.
George, J. F. (1996). Computer-based monitoring: Common perceptions and empirical results. Mis Quarterly,
459-480.
George, J. M., & Zhou, J. (2001). When openness to experience and conscientiousness are related to creative
behavior: an interactional approach. Journal of applied psychology, 86(3), 513.
Gegenhuber, T., Ellmer, M., & Schüßler, E. (2021). Microphones, not megaphones: Functional crowdworker
voice regimes on digital work platforms. Human Relations, 74(9), 1473-1503.
Giddens, L., Petter, S., & Fullilove, M. H. (2023). Information technology as a resource to counter domestic sex
trafficking in the United States. Information Systems Journal, 33(1), 8-33.
Gioia, D. A., Corley, K. G., & Hamilton, A. L. (2013). Seeking qualitative rigor in inductive research: Notes on
the Gioia methodology. Organizational research methods, 16(1), 15-31.
Gonzalez-Mulé, E., Kim, M. M., & Ryu, J. W. (2021). A meta-analytic test of multiplicative and additive models
of job demands, resources, and stress. Journal of Applied Psychology, 106(9), 1391.
Goods, C., Veen, A., & Barratt, T. (2019). “Is your gig any good?” Analysing job quality in the Australian
platform-based food-delivery sector. Journal of Industrial Relations, 61(4), 502-527.
Gramano, E. (2020). Digitalisation and work: challenges from the platform-economy. Contemporary Social
Science, 15(4), 476-488.
Greckhamer, T., Furnari, S., Fiss, P. C., & Aguilera, R. V. (2018). Studying configurations with qualitative
comparative analysis: Best practices in strategy and organization research. Strategic Organization, 16(4),
482-495.
Hajiheydari, N., & Delgosha, M. S. (2024). Investigating engagement and burnout of gig-workers in the age of
algorithms: an empirical study in digital labor platforms. Information Technology & People, 37(7), 2489-
2522.
Halbesleben, J. R., Neveu, J. P., Paustian-Underdahl, S. C., & Westman, M. (2014). Getting to the “COR”
understanding the role of resources in conservation of resources theory. Journal of management, 40(5),
1334-1364.
Humphreys, A., & Wang, R. J. H. (2018). Automated text analysis for consumer research. Journal of Consumer
Research, 44(6), 1274-1306.
Harris, K. J., Harris, R. B., Carlson, J. R., & Carlson, D. S. (2015). Resource loss from technology overload and
its impact on work-family conflict: Can leaders help?. Computers in Human Behavior, 50, 411-417.
Hickman, L., Thapa, S., Tay, L., Cao, M., & Srinivasan, P. (2022). Text preprocessing for text mining in
organizational research: Review and recommendations. Organizational Research Methods, 25(1), 114-146.
Hobfoll, S. E. (2011). Conservation of resource caravans and engaged settings. Journal of occupational and
organizational psychology, 84(1), 116-122.
Hobfoll, S. E., Halbesleben, J., Neveu, J. P., & Westman, M. (2018). Conservation of resources in the
organizational context: The reality of resources and their consequences. Annual review of organizational
psychology and organizational behavior, 5, 103-128.
Holland, P. J., Cooper, B., & Hecker, R. (2015). Electronic monitoring and surveillance in the workplace: The
effects on trust in management, and the moderating role of occupational type. Personnel Review.
Howard, L. W., & Cordes, C. L. (2010). Flight from unfairness: Effects of perceived injustice on emotional
exhaustion and employee withdrawal. Journal of Business and Psychology, 25, 409-428.

30
Huettermann, H., & Bruch, H. (2019). Mutual gains? Health‐related HRM, collective well‐being and
organizational performance. Journal of Management Studies, 56(6), 1045-1072.
Jabagi, N., Croteau, A. M., Audebrand, L. K., & Marsan, J. (2019). Gig-workers’ motivation: Thinking beyond
carrots and sticks. Journal of Managerial Psychology.
Jabagi, N., Croteau, A. M., Audebrand, L., & Marsan, J. (2024). Fairness in Algorithmic Management: Bringing
Platform-Workers into the Fold. Proceedings of the 57th Hawaii international conference on system
sciences (2024).
Jago, A. S. (2019). Algorithms and authenticity. Academy of Management Discoveries, 5(1), 38-56.
Jarrahi, M. H., Newlands, G., Lee, M. K., Wolf, C. T., Kinder, E., & Sutherland, W. (2021). Algorithmic
management in a work context. Big Data & Society, 8(2), 20539517211020332.
Jarrahi, M. H., & Sutherland, W. (2019). Algorithmic management and algorithmic competencies: Understanding
and appropriating algorithms in gig work. In Information in Contemporary Society: 14th International
Conference, iConference 2019, Washington, DC, USA, March 31–April 3, 2019, Proceedings 14 (pp. 578-
589). Springer International Publishing.
Jensen, J. M., & Raver, J. L. (2012). When self-management and surveillance collide: Consequences for
employees’ organizational citizenship and counterproductive work behaviors. Group & Organization
Management, 37(3), 308-346.
Kahn, W. A. (1990). Psychological conditions of personal engagement and disengagement at work. Academy of
Management Journal, 33(4), 692-724.
Kahn, R. L., Wolfe, D. M., Quinn, R. P., Snoek, J. D., & Rosenthal, R. A. (1964). Organizational stress: Studies
in role conflict and ambiguity. Wiley.
Karanović, J., Berends, H., & Engel, Y. (2021). Regulated dependence: Platform workers’ responses to new forms
of organizing. Journal of Management Studies, 58(4), 1070-1106.
Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of
control. Academy of Management Annals, 14(1), 366-410.
Koivusalo, M., Svynarenko, A., Mbare, B., & Perkiö, M. (2024). Globalization, platform work, and wellbeing—
a comparative study of Uber drivers in three cities: London, Helsinki, and St Petersburg. Globalization and
Health, 20(1), 18.
Kuhn, K. M., & Maleki, A. (2017). Micro-entrepreneurs, dependent contractors, and instaserfs: Understanding
online labor platform workforces. Academy of Management Perspectives, 31(3), 183-200.
Kurtessis, J. N., Eisenberger, R., Ford, M. T., Buffardi, L. C., Stewart, K. A., & Adis, C. S. (2017). Perceived
organizational support: A meta-analytic evaluation of organizational support theory. Journal of
management, 43(6), 1854-1884.
Lee, M. K. (2018). Understanding perception of algorithmic decisions: Fairness, trust, and emotion in response to
algorithmic management. Big Data & Society, 5(1), 2053951718756684.
Lee, M. K., Kusbit, D., Metsky, E., & Dabbish, L. (2015, April). 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 (pp. 1603-1612).
Lee, R. T., & Ashforth, B. E. (1996). A meta-analytic examination of the correlates of the three dimensions of job
burnout. Journal of applied Psychology, 81(2), 123.
Leicht-Deobald, U., Busch, T., Schank, C., Weibel, A., Schafheitle, S., Wildhaber, I., & Kasper, G. (2022). The
challenges of algorithm-based HR decision-making for personal integrity. In Business and the Ethical
Implications of Technology (pp. 71-86). Cham: Springer Nature Switzerland.
Lin, M., Miao, L., Wei, W., & Moon, H. (2019). Peer Engagement Behaviors: Conceptualization and Research
Directions. Journal of Service Research, 22(4), 388-403.
Lindebaum, D., Vesa, M., & Den Hond, F. (2020). Insights from “the machine stops” to better understand rational
assumptions in algorithmic decision making and its implications for organizations. Academy of
Management Review, 45(1), 247-263.
Little, L. M., Simmons, B. L., & Nelson, D. L. (2007). Health among leaders: Positive and negative affect,
engagement and burnout, forgiveness and revenge. Journal of management studies, 44(2), 243-260.
Long, C. P., & Sitkin, S. B. (2018). Control–trust dynamics in organizations: Identifying shared perspectives and
charting conceptual fault lines. Academy of Management Annals, 12(2), 725-751
Luthans, F., Avolio, B. J., Avey, J. B., & Norman, S. M. (2007). Positive psychological capital: Measurement and
relationship with performance and satisfaction. Personnel psychology, 60(3), 541-572.
Luthans, F., Norman, S. M., Avolio, B. J., & Avey, J. B. (2008). The mediating role of psychological capital in
the supportive organizational climate—employee performance relationship. Journal of Organizational
Behavior: The International Journal of Industrial, Occupational and Organizational Psychology and
Behavior, 29(2), 219-238.
Luthans, F., & Youssef-Morgan, C. M. (2017). Psychological capital: An evidence-based positive
approach. Annual review of organizational psychology and organizational behavior, 4, 339-366.

31
Maslach, C., & Leiter, M. P. (2008). Early predictors of job burnout and engagement. Journal of applied
psychology, 93(3), 498.
Meijerink, J., & Bondarouk, T. (2023). The duality of algorithmic management: Toward a research agenda on
HRM algorithms, autonomy and value creation. Human Resource Management Review, 33(1), 100876.
Menguc, B., Auh, S., Fisher, M., & Haddad, A. (2013). To be engaged or not to be engaged: The antecedents and
consequences of service employee engagement. Journal of business research, 66(11), 2163-2170.
Misangyi, V. F., Greckhamer, T., Furnari, S., Fiss, P. C., Crilly, D., & Aguilera, R. (2017). Embracing causal
complexity: The emergence of a neo-configurational perspective. Journal of management, 43(1), 255-282.
Morgan, R. M., & Hunt, S. D. (1994). The commitment-trust theory of relationship marketing. Journal of
marketing, 58(3), 20-38.
Morgeson, F. P., & Humphrey, S. E. (2006). The Work Design Questionnaire (WDQ): developing and validating
a comprehensive measure for assessing job design and the nature of work. Journal of applied
psychology, 91(6), 1321.
Möhlmann, M., Alves de Lima Salge, C., & Marabelli, M. (2023). Algorithm sensemaking: how platform workers
make sense of algorithmic management. Journal of the Association for Information Systems, 24(1), 35-64.
Möhlmann, M., Zalmanson, L., Henfridsson, O., & Gregory, R. W. (2021). ALGORITHMIC MANAGEMENT
OF WORK ON ONLINE LABOR PLATFORMS: WHEN MATCHING MEETS CONTROL. MIS
Quarterly, 45(4).
Mohr, J., & Spekman, R. (1994). Characteristics of partnership success: partnership attributes, communication
behavior, and conflict resolution techniques. Strategic Management Journal, 15(2), 135-152.
Müller, O., Junglas, I., Brocke, J. V., & Debortoli, S. (2016). Utilizing big data analytics for information systems
research: challenges, promises and guidelines. European Journal of Information Systems, 25, 289-302.
Nelson, L. K. (2020). Computational grounded theory: A methodological framework. Sociological Methods &
Research, 49(1), 3-42.
O'Driscoll, M. P., & Beehr, T. A. (1994). Supervisor behaviors, role stressors and uncertainty as predictors of
personal outcomes for subordinates. Journal of organizational Behavior, 15(2), 141-155.
Ong, W. J., & Johnson, M. D. (2023). Toward a configural theory of job demands and resources. Academy of
Management Journal, 66(1), 195-221.
Parent-Rocheleau, X., & Parker, S. K. (2022). Algorithms as work designers: How algorithmic management
influences the design of jobs. Human Resource Management Review, 32(3), 100838.
Park, Y., El Sawy, O. A., & Fiss, P. (2017). The role of business intelligence and communication technologies in
organizational agility: A configurational approach. Journal of the association for information
systems, 18(9), 1.
Parker, S. K., & Grote, G. (2022). Automation, algorithms, and beyond: Why work design matters more than ever
in a digital world. Applied Psychology, 71(4), 1171-1204.
Parker, S. K., Morgeson, F. P., & Johns, G. (2017). One hundred years of work design research: Looking back
and looking forward. Journal of applied psychology, 102(3), 403.
Peer, E., Vosgerau, J., & Acquisti, A. (2014). Reputation as a sufficient condition for data quality on Amazon
Mechanical Turk. Behavior research methods, 46(4), 1023-1031.
Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003). Common method biases in behavioral
research: a critical review of the literature and recommended remedies. Journal of applied
psychology, 88(5), 879.
Prassl, J. (2018). Humans as a service: The promise and perils of work in the gig economy. Oxford University
Press.
Rani, U., Kumar Dhir, R., Furrer, M., Gőbel, N., Moraiti, A., & Cooney, S. (2021). World employment and social
outlook: the role of digital labour platforms in transforming the world of work. Geneva: International
Labour Organisation.
Rich, B. L., Lepine, J. A., & Crawford, E. R. (2010). Job engagement: Antecedents and effects on job
performance. Academy of Management Journal, 53(3), 617-635.
Roberts, M. E., Stewart, B. M., & Tingley, D. (2019). Stm: An R package for structural topic models. Journal of
Statistical Software, 91, 1-40.
Roche, M., Haar, J. M., & Luthans, F. (2014). The role of mindfulness and psychological capital on the well-
being of leaders. Journal of occupational health psychology, 19(4), 476.
Rosenblat, A. (2018). Uberland: How algorithms are rewriting the rules of work. University of California Press.
[Link]
Schaufeli, W. B., Bakker, A. B., & Salanova, M. (2006). The measurement of work engagement with a short
questionnaire: A cross-national study. Educational and psychological measurement, 66(4), 701-716.
Schmidt, S., Roesler, U., Kusserow, T., & Rau, R. (2014). Uncertainty in the workplace: Examining role
ambiguity and role conflict, and their link to depression—a meta-analysis. European Journal of Work and
Organizational Psychology, 23(1), 91-106.

32
Soltani Delgosha, M., Hajiheydari, N., & Olya, H. (2024). A person‐centred view of citizen participation in civic
crowdfunding platforms: A mixed‐methods study of civic backers. Information Systems Journal, 34(5),
1626-1663.
Statista (2022) Gig economy in the U.S. - Statistics & Facts (Accessed 01/10/22 from
[Link]
Tarafdar, M., Page, X., & Marabelli, M. (2022). Algorithms as co‐workers: Human algorithm role interactions in
algorithmic work. Information Systems Journal.
Van Woerkom, M., Bakker, A. B., & Nishii, L. H. (2016). Accumulative job demands and support for strength
use: Fine-tuning the job demands-resources model using conservation of resources theory. Journal of
Applied Psychology, 101(1), 141.
van Zoonen, W., Sivunen, A. E., & Treem, J. W. (2024). Algorithmic management of crowdworkers: Implications
for workers’ identity, belonging, and meaningfulness of work. Computers in Human Behavior, 152, 108089.
Venkatesh, V., Brown, S. A., & Bala, H. (2013). Bridging the qualitative-quantitative divide: Guidelines for
conducting mixed methods research in information systems. MIS quarterly, 21-54.
Venkatesh, V., Brown, S. A., & Sullivan, Y. W. (2016). Guidelines for conducting mixed-methods research: An
extension and illustration. Journal of the Association for Information systems, 17(7), 2.
Walsh, M. (2020). Algorithms Are Making Economic Inequality Worse. Harvard Business Review, 32.
Ward, S., & Steptoe-Warren, G. (2014). A conservation of resources approach to BlackBerry use, work-family
conflict and well-being: Job control and psychological detachment from work as potential
mediators. Engineering Management Research, 3(1), 8.
Wayne, S. J., Shore, L. M., & Liden, R. C. (1997). Perceived organizational support and leader-member exchange:
A social exchange perspective. Academy of Management journal, 40(1), 82-111.
Wiener, M., Cram, W. A., & Benlian, A. (2023). Algorithmic control and gig workers: a legitimacy perspective
of Uber drivers. European Journal of Information Systems, 32(3), 485-507.
Wood, A. J., Graham, M., Lehdonvirta, V., & Hjorth, I. (2019). Good gig, bad gig: autonomy and algorithmic
control in the global gig economy. Work, Employment and Society, 33(1), 56-75.
Yang, Z., Dai, Z., Yang, Y., Carbonell, J., Salakhutdinov, R. R., & Le, Q. V. (2019). Xlnet: Generalized
autoregressive pretraining for language understanding. Advances in neural information processing
systems, 32.

33
 Appendix A- Review of Algorithmic Management Literature

Authors Context Research Design Major Findings

Cheng & Foley Airbnb hosts Thematic analyses of online community • The asymmetry of algorithmic information can enhance Airbnb's
(2019) forum posts influence and control over hosts' practices.
• This information asymmetry can substantially impede hosts' sense of
control.
Gol et al (2019) Crowdworking platforms Theoretical review strategy • For effective crowdwork platform governance, control and coordination
mechanisms are crucial.
• Governance, particularly control, plays a more significant role in
centralized platforms, while coordination is vital for decentralized ones.
• The emphasis on effective governance is particularly pronounced in non-
routine creative crowdwork.
Jarrahi & Digital labor platforms, the case Qualitative study based on interviews • Workers are not passive recipients of algorithmic management and
Sutherland of Upwork users with Upwork workers control but they develop “algorithmic competencies” as a source of
(2019) personal competitive advantage in platform work.
• Workers may decide to circumvent or manipulate the algorithms to their
advantage.
• The gig workers on Upwork often lack a complete understanding of the
inner workings of many algorithms, creating information asymmetry that
can impede their ability to exert control over their work.
Basukie et al Go-Jek, an Indonesian ride- Qualitative research design, case study • Legal and ethical concerns are the main challenges of Go-Jek in
(2020) sharing platform emerging markets.
• Go-Jek is appreciated in this emerging economy since it offers job
opportunities and income generation.
Duggan et al Employment relations and HRM Conceptual paper • App-workers perceive their working relationship with the platform
(2020) for App-work beyond merely economic or transactional exchanges.
• The working relationship in app-work raises concerns around power
imbalances, fairness, and worker well-being.
• Algorithms mainly undertake the roles in work assignment and
performance management that were traditionally the preserve of HR
professionals.
Gal et al (2020) General people analytics Conceptual paper • Three ethical consequences of people analytics are discussed: opacity,
applications of Algorithmic datafication of the workplace, and nudging.
Management • The adverse effects of these challenges on members' ability to cultivate
their virtues are examined.

34
Authors Context Research Design Major Findings
• Three mitigation strategies are suggested including: reframing PA,
adopting new roles, and applying alternative design principles.
Galiere (2020) Food-delivery platform Qualitative case study • Algorithmic rational control, though fallible, is strengthened by
(Deliveroo) subjectification techniques.
• Various platform features, including pay-per-delivery and shift picker
systems, actively mobilize workers.
Stark & Pais Platform economy Conceptual paper • Unlike traditional markets, hierarchies, and networks, platforms draw
(2020) upon external assets, resources, and activities.
• Platforms manage value creation activities on their platforms without
traditional managerial authority, reshaping the managerial class in the
21st century.
• Users’ and providers' actions are translated into rankings and calculations
through complex feedback loops, leading to a unique form of cybernetic
control where accountability can be elusive.
Bucher et al Upwork workers Systematic content analysis of comments • Workers employ anticipatory compliance practices to pacify algorithms
(2021) from Reddit and avoid scrutiny or punishment.
• These practices involve direct and indirect strategies, such as avoiding
triggering words or undervaluing work for better ratings.
Jarrahi et al Standard and non-standard work Conceptual paper • Algorithmic management influences power dynamics at work by
(2021) settings increasing the power of managers over workers and simultaneously
decreasing managerial authority.
• Algorithmic management shapes organizational roles through algorithm
aversion and cognitive complacency.
• Algorithmic management impacts knowledge and information exchange
within an organization.
Möhlmann et al Online labor platforms- The Case study and grounded theory • OLPs play dual roles of matching and control: using algorithms, they
(2021) case of Uber optimize matching and accommodate the market, while simultaneously,
monitoring and controlling platform work.
• Platform workers experience autonomy, uncertainty and self-identity
tensions relating to work execution, compensation, and belonging.
• In managing the experienced tension, workers present market-like
(bypassing & switching) or organization-like (striking & embracing)
responses.
Wood (2021) Examples of platform works and Conceptual paper • Algorithmic management expedites the growth of precarious, fragmented
conventional employment employment arrangements (such as outsourcing, franchising, temporary
settings employment agencies, labor intermediaries, and digital labor platforms).

35
Authors Context Research Design Major Findings
• Algorithmic management exacerbates working conditions by increasing
standardization and reducing opportunities for using discretion and
intrinsic skills.
• Evidence from platform work and logistics highlights the danger of
algorithmic management intensifying work effort, creating new sources
of algorithmic insecurity and provoking workplace resistance.
Anicich (2022) App-based, food delivery Qualitative, autoethnography • Due to independent contracting, technological constraints, and having no
platform coworkers, app-work results in high depersonalization and low
interpersonal accountability for workers.
• To overcome this, app-workers rely on narrative structuring, fantasizing,
rationalizing, and connecting through storytelling in online forums.
Bujold et al Truck drivers in Eastern Canada Questionnaire-based survey from truck • Algorithmic surveillance transparency is linked to procedural justice, and
(2022) drivers algorithmic performance management transparency is tied to distributive
justice.
• Procedural justice mediates the negative link between algorithmic
surveillance transparency and intention to quit.
Gagné et al Conceptual effects of Literature review based on self- • Algorithmic management has predominantly negative impacts, frustrating
(2022) algorithmic management on determination theory (SDT). key psychological needs such as competence, autonomy, and relatedness.
work motivation • Some features of algorithmic management systems such as transparency,
reliability, fairness of the system, and level of human influence can
moderate these effects.
Kinowska & Workplace well-being in case of Analyzing the 2019 ECS data (European • The results affirm a moderate, direct impact of algorithmic management
Sienkiewicz using data analytics for some human resource managers and practices on workplace well-being.
(2022) organizational functions employees) • There is an indirect influence, primarily through a negative impact on job
autonomy and total rewards practices.
• The level of influence varies significantly, particularly in larger
organizations, where the impacts of algorithmic management on well-
being and job autonomy decrease.
Parent- Conceptual work design and Review/conceptual paper • Algorithms are currently able to perform six managerial functions
Rocheleau & HRM including: monitoring, goal setting, performance management,
Parker (2022) scheduling, compensation, and job termination.
• Algorithmic management can affect job resources (e.g., job autonomy,
job complexity) and job demands (e.g., workload, physical demands).
• Transparency, fairness and human influence can moderate the link
between algorithmic management on work design.

36
Authors Context Research Design Major Findings
Tomprou & Lee The impact of organizational Online experiments • The agent type (either human or algorithm) has no impact on
(2022) agent type—algorithmic versus psychological contracts regarding transactional inducements but does for
human—on psychological relational inducements, particularly in recruiting and low-inducement
contract delivery scenarios.
• Algorithmic agents signal reduced employer commitment to relational
inducements during recruiting, while the use of human agents results in
greater perceived breaches when the delivery of relational inducements is
low.
Felix et al Brazilian gig-workers Grounded theory based on interviews • Workers and gig-work organizations prefer relationships with more
(2023) autonomy or security.
• Wellbeing increases when preferences align, but divergent preferences
lead to violations and reduced well-being.
• Not everything is a matter of fit: when both individuals and organizations
have the same preference (for example, for more autonomy and less
security), worker well-being may be vulnerable to abuse such as in the
case of unsustainable workloads.
Huang (2023) food‐delivery platform in China Ethnographic study • There are two primary mechanisms for organizing labor under
algorithmic management: virtual organization of labor and algorithm-
driven labor process control.
• Initially, platform drivers are restructured into an outsourced labor force,
operating within an application-based virtual-networked production
framework
• Additionally, food-delivery platforms employ four algorithm-driven
control techniques—smart machinery control, information monopoly,
management by multi-stakeholders, and ‘carrots and sticks’.
Meijerink & Human resource management Conceptual paper • Algorithmic management exhibits a dual nature: while it restricts job
Bondarouk (HRM) algorithms autonomy and value in some aspects, it simultaneously enhances job
(2023) autonomy and value for workers.
• Algorithmic management is both influenced by and influential in terms of
job autonomy and value for workers.
Möhlmannn et Uber platform Empirical case study • It builds upon Weick’s enactment theory to conceptualize how platform
al (2023) workers make sense of the algorithms managing them.
• Algorithm sensemaking is a sophisticated, strategic, multi-step process
enabling platform workers to systematically comprehend and follow
algorithmic activity.

37
Authors Context Research Design Major Findings
• Algorithm sensemaking serves as a mediating process, connecting
workers' perceptions of tensions in their work environment (antecedents)
to their behavioral responses (consequences).
Noponen et al Reviewing algorithmic systems Systematic literature review • Organizations are using algorithms to manage workers in four different
(2023) in management literature ways: Algorithm as a manager, mentor, controller, and servant.
• Despite the potential for enabling workers, companies predominantly
utilize algorithmic systems for control, neglecting their organizing and
enabling potentials.
Vignola et al Workers’ health in platform- Review • Algorithmic management's impact on job quality likely extends to worker
(2023) based food and grocery delivery health and well-being.
companies • The design and implementation of algorithms can be balanced to meet
both organizational and worker needs.
• Policy changes in labor and social welfare laws are potential intervention
points, contingent on a better understanding of algorithmic management's
effects on workers' health.

1. Anicich, E. M. (2022). Flexing and floundering in the on-demand economy: Narrative identity construction under algorithmic management. Organizational
Behavior and Human Decision Processes, 169, 104138.
2. Basukie, J., Wang, Y., & Li, S. (2020). Big data governance and algorithmic management in sharing economy platforms: A case of ridesharing in emerging
markets. Technological Forecasting and Social Change, 161, 120310.
3. Bucher, E. L., Schou, P. K., & Waldkirch, M. (2021). Pacifying the algorithm–Anticipatory compliance in the face of algorithmic management in the gig
economy. Organization, 28(1), 44-67.
4. Bujold, A., Parent-Rocheleau, X., & Gaudet, M. C. (2022). Opacity behind the wheel: The relationship between transparency of algorithmic management, justice
perception, and intention to quit among truck drivers. Computers in Human Behavior Reports, 8, 100245.
5. Cheng, M., & Foley, C. (2019). Algorithmic management: The case of Airbnb. International Journal of Hospitality Management, 83, 33-36.
6. Duggan, J., Sherman, U., Carbery, R., & McDonnell, A. (2020). Algorithmic management and app‐work in the gig economy: A research agenda for employment
relations and HRM. Human Resource Management Journal, 30(1), 114-132.
7. Felix, B., Dourado, D., & Nossa, V. (2023). Algorithmic management, preferences for autonomy/security and gig-workers' wellbeing: A matter of fit?. Frontiers in
Psychology, 14, 1088183.
8. Gal, U., Jensen, T. B., & Stein, M. K. (2020). Breaking the vicious cycle of algorithmic management: A virtue ethics approach to people analytics. Information and
Organization, 30(2), 100301.
9. Galiere, S. (2020). When food‐delivery platform workers consent to algorithmic management: a Foucauldian perspective. New Technology, Work and
Employment, 35(3), 357-370.
10. Gagné, M., Parent-Rocheleau, X., Bujold, A., Gaudet, M. C., & Lirio, P. (2022). How algorithmic management influences worker motivation: A self-determination
theory perspective. Canadian Psychology/Psychologie canadienne, 63(2), 247.

38
11. Gol, E. S., Stein, M. K., & Avital, M. (2019). Crowdwork platform governance toward organizational value creation. The Journal of Strategic Information
Systems, 28(2), 175-195.
12. Huang, H. (2023). Algorithmic management in food‐delivery platform economy in China. New Technology, Work and Employment, 38(2), 185-205.
13. Jarrahi, M. H., Newlands, G., Lee, M. K., Wolf, C. T., Kinder, E., & Sutherland, W. (2021). Algorithmic management in a work context. Big Data & Society, 8(2),
20539517211020332.
14. Jarrahi, M. H., & Sutherland, W. (2019). Algorithmic management and algorithmic competencies: Understanding and appropriating algorithms in gig work.
In Information in Contemporary Society: 14th International Conference, iConference 2019, Washington, DC, USA, March 31–April 3, 2019, Proceedings 14 (pp.
578-589). Springer International Publishing.
15. Kinowska, H., & Sienkiewicz, Ł. J. (2022). Influence of algorithmic management practices on workplace well-being–evidence from European
organisations. Information Technology & People, 36(8), 21-42.
16. Meijerink, J., & Bondarouk, T. (2023). The duality of algorithmic management: Toward a research agenda on HRM algorithms, autonomy and value
creation. Human Resource Management Review, 33(1), 100876.
17. Möhlmannn, M., Alves de Lima Salge, C., & Marabelli, M. (2023). Algorithm sensemaking: How platform workers make sense of algorithmic
management. Journal of the Association for Information Systems, 24(1), 35-64.
18. Möhlmann, M., Zalmanson, L., Henfridsson, O., & Gregory, R. W. (2021). ALGORITHMIC MANAGEMENT OF WORK ON ONLINE LABOR PLATFORMS:
WHEN MATCHING MEETS CONTROL. MIS quarterly, 45(4).
19. Noponen, N., Feshchenko, P., Auvinen, T., Luoma-aho, V., & Abrahamsson, P. (2023). Taylorism on steroids or enabling autonomy? A systematic review of
algorithmic management. Management Review Quarterly, 1-27.
20. Parent-Rocheleau, X., & Parker, S. K. (2022). Algorithms as work designers: How algorithmic management influences the design of jobs. Human resource
management review, 32(3), 100838.
21. Stark, D., & Pais, I. (2020). Algorithmic management in the platform economy. Sociologica, 14(3), 47-72.
22. Tomprou, M., & Lee, M. K. (2022). Employment relationships in algorithmic management: A psychological contract perspective. Computers in Human
Behavior, 126, 106997.
23. Vignola, E. F., Baron, S., Abreu Plasencia, E., Hussein, M., & Cohen, N. (2023). Workers’ Health under Algorithmic Management: Emerging Findings and Urgent
Research Questions. International Journal of Environmental Research and Public Health, 20(2), 1239.
24. Wood, A. J. (2021). Algorithmic management consequences for work organisation and working conditions (No. 2021/07). JRC Working Papers Series on Labour,
Education and Technology.

39
Appendix B- Opinion Mining using XLNet

XLNet was architected as a multilayer bidirectional transformer encoder pretrained on a large

corpus, including the Toronto Book Corpus and English Wikipedia. To adapt XLNet to a
specific text mining task, a proper fine-tuning procedure is needed. Fine-tuning involves adding
an extra output layer on top of the pre-trained XLNet structure, to align with a specific task.
This is a method of transfer learning, where the knowledge acquired by the large neural network
on a vast corpus, such as Wikipedia, is applied to a specific task such as opinion mining (Dong
et al., 2019). As a result, fine-tuning XLNet requires comparatively smaller training data to
achieve satisfactory results, making it an advantageous technique for organizational and
management studies, where obtaining labels for text data, like valence, can be expensive and
time-consuming. To fine-tune XLNet, we used an optimism/pessimism labeled dataset that was
made available by Ruan et al. (2016). The dataset contains textual data that was manually
annotated by human raters as either optimistic or pessimistic, which aligns perfectly with the
classification tasks in our study. We followed the approach proposed by Alshahrani et al.
(2020) and considered (1/-1) threshold, that is posts with an average score of 1 or greater were
labeled as optimistic, whereas those with an average score of -1 or lower were labeled as
pessimistic.

We partitioned the dataset into an 80% training set and a 20% testing set. To train the model,
XLNet’s tokenizer was first used to convert the input sequence into tokens that corresponded
to XLNet’s vocabulary. These tokens were then fed into the model. The architecture of the
XLNet base model used for training consisted of 12 layers, 768 hidden neurons, 12 heads, and
a total of 110 million parameters. Our XLNet model was topped with an untrained classifier
layer which was fine-tuned in addition to tuning the transformers. We set a conservative base
learning rate of 2e-4 and batch size of 16 for 10 epochs (Fyffe et al., 2023).

As presented in Table B.1, our fine-tuned XLNet model has achieved an outstanding overall
accuracy of 93%, surpassing the Linguistic Inquiry and Word Count (LIWC)-based SVM by
14% and the LIWC-based logistic regression by 19% as base models. This higher performance
may be attributed to XLNet special capability to not only understand the relations between
words but also to compute contextualized representations of textual data.

Table B.1. Model accuracy comparison with benchmark models on the hold-out test dataset
Baseline 1: SVM Baseline 2: Logistic
based on LIWC regression based on
Opinions Fine-tuned XLNet features LIWC features
Optimism 90 73 68
Pessimism 96 85 80
Overall accuracy 93 79 74

Dong, L., Yang, N., Wang, W., Wei, F., Liu, X., Wang, Y., ... & Hon, H. W. (2019). Unified language model pre-
training for natural language understanding and generation. Advances in neural information processing
systems, 32.
Fyffe, S., Lee, P., & Kaplan, S. (2023). “Transforming” Personality Scale Development: Illustrating the Potential
of State-of-the-Art Natural Language Processing. Organizational Research Methods, 10944281231155771.
Ruan, X., Wilson, S., & Mihalcea, R. (2016, August). Finding optimists and pessimists on twitter. In Proceedings
of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) (pp.
320-325).

40
Appendix C- Topic Modeling Method and Results

C1- Identification of Topics

The topic modeling algorithms such as STM leverage statistical generative procedures to
uncover topics representing the latent, underlying structure in the text. Simply put, topic
modeling imitates the process of generating a document, by decomposing the documents into
two matrices: words per topic, and topics per document. Topic modeling assumes that each
document has its own topic distribution, and each topic has its own word distribution. The
algorithm outputs a list of the highest weighted words per topic suggesting the content of the
topics, as well as the words’ probabilistic relation with each topic. In addition, each document
can be annotated by a distribution over topics, which enables detecting of thematic patterns
across the documents. As a result, this computational approach makes inductive content
analysis more efficient, valid, transparent, and reproducible, enabling us to thematically
classify and discover hidden patterns in massive amounts of user-generated data.

To prepare the corpus for topic modeling analysis, we followed Hickman et al.’s (2022)
recommendations and performed several preprocessing to clean and parse the textual data. All
forum posts were converted to lowercase, and nonalphabetic characters including punctuations,
symbols, numbers, and whitespaces were removed from the final textual corpus. The cleaned
corpus was converted into a sparse token count matrix with documents (i.e., posts) in rows,
tokens in columns, and cell values showing the number of times each token appeared in each
post. To mitigate the likelihood of extracting uninformative topics and increase model
convergence, words with fewer than three letters or extremely sparse that repeated fewer than
five times (that are usually typos or proper nouns) or appeared in fewer than 500 posts were
removed. Any forum posts that no longer contained any tokens were also eliminated from the
corpus. We used stm package (Roberts et al., 2019) in the programming language R to estimate
topic models. As recommended by Roberts et al. (2019), the topic model was estimated using
Spectral algorithm as it outperforms other methods for modeling a large corpus. In addition,
we followed Schmiedel et al. (2019) guidelines to algorithmically find the right number of
topics. We fitted a structural topic model to investigate different topics discussed among
crowdworkers’ posts within pessimistic and optimistic posts. The final model identified 40
topics totally that were both semantically coherent and highly exclusive (see Table C.1 for
details of topic modeling and list of extracted topics).

To determine which topic model was best for our corpus, we were required to specify the
number of topics to be discovered from the document collection. Following Schmiedel et al.
(2018), we iteratively ran different topic models and analyzed the results to find the appropriate
number of topics which were not overloaded and overlapping. We used semantic coherence
and exclusivity metrics to find an appropriate number of topics in our corpus. Semantic
coherence is a measure of the internal coherence of topics and is maximized when the most
probable words in a given topic frequently co-occur together (Roberts et al., 2016). Exclusivity
is gauged by assessing the distinctiveness of topics through the comparison of word
distributions associated with each topic (Schmiedel et al., 2018).

Given the size of our corpus, we ran different topic models ranging from 10 to 100 topics and
assessed the average semantic coherence and exclusivity for each topic model. The Figure C.1
shows the scores for exclusivity at the top plot and scores for semantic coherence at the bottom
plot. Results indicate that a model consisting of 40 topics has better semantic coherence and
exclusivity values.

41
 Figure C.1. Exclusivity and semantic coherence scores of topic models

Hickman, L., Thapa, S., Tay, L., Cao, M., & Srinivasan, P. (2022). Text preprocessing for text mining in
organizational research: Review and recommendations. Organizational Research Methods, 25(1), 114-146.
Roberts, M. E., Stewart, B. M., & Tingley, D. (2019). Stm: An R package for structural topic models. Journal of
Statistical Software, 91, 1-40.
Schmiedel, T., Müller, O., & Vom Brocke, J. (2019). Topic modeling as a strategy of inquiry in organizational
research: A tutorial with an application example on organizational culture. Organizational Research
Methods, 22(4), 941-968.

42
C2- Topic Modelling Results

To complement the quantitative analysis and select the relevant topics, we qualitatively examined the interpretability of the different models. Our
analysis led to exclusion of 10 topics due to irrelevance (discussions included general feelings, questions, and different issues which were not
specifically related to our research question). As a result, the final model used for the analysis consisted of 30 topics and we excluded topics 1
(general feelings), 10 (tipping), 11 (riders’ behaviors), 19 (general forum posts), 20 (tenure), 29 (interactions with riders), 33 (car condition), 35
(types of riders), 37 (general questions), and 39 (insurance).

Topic ID Top Probability Words Exemplary Forum Posts Topic Label Rank1
1 good, thing, point, idea, best, sound, bad, pretty, luck, damn Good luck to everyone. Im not giving any rides untill General feelings
Monday. My stress levels csnt handle that traffic. 300,875
Keep safe everyone.
2 rate, low, acceptance, increase, count, current, high, result, My acceptance rate was nearly perfect then Uber Job misallocation
bad, gave started sending me requests that were 8-18 miles
away. Has anyone else experienced this? No way I’m 73,014
doing a pick up so far away when I live in a resort
area (Virginia Beach) that’s busy.
3 fee, cancel, show, far, charge, take, pickup, get, return, Why are they playing games with the cancellation Scrutinizing
location rate? So I've noticed this over the last few months cancellation rate
and I think the cancellation rate is being
83,326
manipulated. Take for example last night... When I
started the night, I was at a 7% cancellation rate, then
I got it down to 5%. …
4 time, long, waste, wait, happen worse, reject, gas, period, Is waiting on the lot a complete waste nowadays? I Long waiting time
spend waited almost 1 hour and half for a 15 dollar ride... I
would have made at least 30 by just driving and not 80,806
waiting on the lot. Anyone has any tips. I drive xl
and for both Uber and Lyft 4.9 rating.
5 job, meet, work, gig, benefit, drive, make, living, regular, If you can run twenty pools in a day where you get Task allocation
enjoy two/shuffle two, you are getting more money at the
end of the day than would a cab driver who ran his
meter on each of those same trips. The thing is that
they come fast and furious enough that you could 103,633
knock out those twenty is six hours on a good day
where almost everything fell just right, with a little
room for an aberrant trip, or two. Only on the very
best of days could a cab driver run twenty trips in six

43
Topic ID Top Probability Words Exemplary Forum Posts Topic Label Rank1
hours, and EVERYTHING would have to come and
fall JUST EXACTLY RIGHT, with no room for
anything aberrant.
6 mile, cent, true, cost, gas, maintenance, value, profit, drive, Apparently Uber is experimenting with the rate Unpredictable fare
math structure. Raising the per-minute rate and lowering
the per-mile rate. Rolling out in Dallas, Houston,
Sacramento, Salt Lake City and Minneapolis. They
claim that driver earnings will stay the same (So why
bother?). 176,337
They are raising the per-minute rate by 50-82% and
reducing the per-mile rate by 10-19%
They recently raised overall rates in Chicago by a
paltry amount which didn't come close to covering
gas increases.
7 service, feedback, provide, income, complaint, notice, Been driving for only four months, got this email: Nitpicking boss
order, act, terms, correct "I'm following up on feedback about your behavior
during a recent ride. The feedback alleged that you
56,732
asked your Passenger personal questions or
requested contact information. As a reminder, Lyft is
an inclusive community…
8 check, state, case, video, safety, record, seek, know, When downloading the app it obviously asks to Privacy
privacy, data allow access to other content and areas of your
phone. Is this just like Facebook in a sense, having
88,646
all of your data in the palm of their hands? Is this
moral or an invasion of privacy, especially if
information gathered is abused?
9 sure, doesnt, sense, feel, wonder, know, difference, change, someone knows how the ping system works? I am Unclear routines
clear, sound still trying to figure out this system.
Receiving trips surcharge for trips on uber. No idea 50,512
how it works. You guys have more information
about it ?
10 tip, sign, expect, cheap, delivery, cash, dollar, add, nice, I do real good with the tips. I'm pretty sure the signs Tipping
option help. I get a lot of people asking about the included
tip. I explain to them that not only is a tip not
119,189
included, but they take 25% of every single fare, and
that they provide nothing but passengers. I provide
the car, insurance, maintenance, tires plus everything

44
Topic ID Top Probability Words Exemplary Forum Posts Topic Label Rank1
in the car like Wi-Fi and Satellite Radio just to make
your ride more enjoyable.
11 seat, back, front, door, left, put, passenger, open, behind, I purposely kept my "stuff" on the front seat so Riders’ behaviors
window people wouldn't sit there. I don't know why, but most
of the riders I pick up want to sit in the front seat.
106,517
After reading post on this thread, I would prefer
them riding in the back as well. I pick up a pax today
who opened the front door …
12 night, day, home, morning, late, weekend, stay, early slow, I think I'm doing alright as a part-time driver. I've Time flexibility
free been driving for three months and I put in about 20
hours a week, mostly on Fridays and Saturdays. I
have a full-time job with benefits and a solid income.
Thus Uber, gives me flexibility to take on some
105,444
additional projects with my home. Also, I itemize my
taxes; so, I believe the IRS business deductions help
to make Uber more profitable. I guess I'll know after
I've been doing it for awhile. Does anyone know how
long the average Uber partner remains a partner?
13 make, agree, fact, reason, understand, follow, ask, talk, I get this message from Lyft support. The problem is, Insufficient
unfortunately, response I don't pull those tricks to get out of a short run like explanation
some drivers do. Apparently the Lyft nanny bots are
sniffing out what they THINK are tell-tale signs of
me doing that. After an experience the other morning
I found out what's really happening, and why. … I
called Lyft to ask if the customer had cancelled or
was it a Lyft bot. He said a Lyft bot did it, since I 69,320
wasn't making fast enough progress to the pickup
and another driver with a shorter ETA became
available, so the ride got switched to the other driver.
… The worst part? Unless you get on the phone and
beg, like I did, you don't even get a cancellation fee
in that case, since they consider YOU to be the
problem.
14 account, email, support, info, send, message, late, help, call, Uber (and Lyft) Driver "Support" is outsourced. Ineffective
contact Most of the e-Mail "support" is handled in Bangalore communication
120,713
and Pondicherry. Most of the telephone support is
handled in Quezon City. The average employee has a

45
Topic ID Top Probability Words Exemplary Forum Posts Topic Label Rank1
command of the English Language that approaches
that of a third grader. You send an e-Mail to
"support". The employee opens it. The computer
program highlights certain words and phrases, then
suggests several replies, in order of what the program
interprets as "relevant". The employee than picks one
and the program either sends that templated e-Mail
or, it takes the cookie cutter phrases and composes
an e-Mail. This is often why the response that you
receive has less than nothing to do with your
problem. If you are going to deal with Driver
"Support", you must be prepared to send several e-
Mails and make the tone of each one more and more
obnoxious
15 cant, believe, find, anymore, wont, without, able, speak, No one listens when I complain. I pour my heart out Neglecting
stand, trust in an Uber and get secretly recorded and I'm the bad drivers’ concerns 120,334
guy? …
16 airport, lot, around, traffic, street, town, city, area, road, Drivers now have their own personal navigator to Traffic
avoid help steer them to their destination using the best information
available, least congested route.
175,025
A new mobile phone app called ACT Traffic
provides audio alerts and other information about
congestion on the road network in real time.
17 rating, star, ridder, system, give, worry, driver, change, I believe the rating system is ridiculous. It starts at 5 Intrusive rating
attitude, five and has no where to go but down. Psychologically a system
terrible plan! Constantly worrying and fighting to
maintain a perfect record. No wonder the 6th star
plan shows behavior by drivers that is impossible to
172,269
sustain and benefits only the company's bottom line.
Forcing Uber drivers to try to live up to an
impossible standard, with failure resulting in
termination, is truly a stupid idea. Incredible how
this rating concept even started….
18 surge, driver, demand, market, price, many, huge, change, I haven’t seen surge pricing or any promotions in the Unfair surge
fare, available Harrisburg area in over a month. Anyone know why? pricing
154,823
I am getting frequent requests for over 16 miles away
for 3-5 min trips so obviously no Uber’s in area, isn’t

46
Topic ID Top Probability Words Exemplary Forum Posts Topic Label Rank1
this definition of supply and demand? Uber support
had no answers (as usual)
19 post, youre, read, forum, thread, reading, saying, right, I guess there are more people participating in this General forum
talking, welcome Forum.....ants or not, UP is growing....welcome posts 145,754
everybody... ask for the Billboard
20 year, age, old, month, started, day, new, couple, week, past I started driving Uber in the IE over 3 years ago. Tenure
Since then I moved to LBC then Vegas and now I'm
back In the IE. I was here when Uber opened shop in
the IE and Lyft was just making a name for
220,112
themselves in LA only. When I came out to drive in
the early mornings on the weekend there would be an
average of 3-5 cars max online from San Berdo to
Pomona. …
21 stop, report, police, rule, state, law, suck, mistake, charge, it is inevitable. After a certain amount of time False accusation
claim especially in Philly you will get a false report, and
you will get falsely deactivated. I have sent dash cam
footage to Uber to no avail. Uber is always going to 100,963
protect themselves before anything. I will never send
them dahs cam footage again, as it did not effect my
outcome of what i wanted done.
22 help, info, great, well, advice, receive, got, thought, set, I finally created excel worksheets with the info that Performance
message Uber provides. I searched on how are the promotions feedback
147,905
that we sign up for paid and it says that it will be
either a Miscellaneous or Other payment.
23 ever, time, question, heard, wait, happen, since, notice, late, Tried contacting other driver's insurance company, Unresponsiveness
take and miraculously, it looks like they are taking
responsibility and will arrange repairs! Keeping my 131,942
fingers crossed! BTW, still have not had any
response from Uber.
24 app, free, know, try, use, plan, time, control, work, watch Uber controls your jobs and they want to control Overcontrolling
your phone too! Cant even multi task without it
putting you offline without noticing. I know uber 39,317
constantly pings the app and put you offline and then
online if the app is open.
25 trip, fare, minimum, distance, long, take, base, short, Since Uber seems to adjust fares without notification Unstable
change, add to the driver , how does everyone monitor their commission fees 105,367
statements ? Seems for a driver that does any more

47
Topic ID Top Probability Words Exemplary Forum Posts Topic Label Rank1
than a few trips a week it would be too hard to
remember trips.
26 contract, rule, driver, deactivate, public, legal, account, I worked for Flex for 5 months at their West Drayton Account
wait, talk, ban London depot. I was unemployed so I thought ok, it's deactivation/suspe
reasonable money. … I recvd their email last week nsion 172,409
saying I'm being deactivated cos I dont pick up all
the parcels on my route. That is 200% wrong...I have
NEVER refused to pick up parcels at all.
27 pay, statement, amount, daily, bill, get, enough, attention, On Thursday, the company announced Uber Pro, a Extra pay
agree, app driver loyalty program that offers perks ranging from
extra pay and free online education to dent repair and
gas discounts. 91,970
The program may help retain drivers, who often quit
within a few months, research shows. Uber's drivers
are independent contractors so they do not receive
benefits such as health care or tuition reimbursement.
28 ride, accept, pool, request, pick, ping, destination, select, New-and-improved driver experience includes Location
area, around upgrades to upfront pay, plus features offering flexibility
drivers even more control, choice and transparency
… The “stay within area” location filter quickly
185,839
became a driver favorite upon introducing it earlier
this year, enabling drivers to stay closer to home
with shorter rides, avoid traffic, combine with streak
zones to maximize earnings …
29 Pax, water, talk, music, want, play, hear, prefer, game, So I have this couple on a select ride,. They were Interactions with
conversation talking to each other and when they got out the riders
woman says "be safe tonight, stay hydrated, I 100,368
brought this bottle of water for you!" That was a first
for me. …
30 app, phone, address, screen, google, show, gps, location, I use the in app Uber and Lyft navigation along with Navigation advice
update, route my knowledge of the area I am driving. My Uber
Navigation has a map and voice directions as well.
Wish Uber would add a quick touch mute button like 67,987
Lyft has. For now if I want to silence the Uber voice
directions I just turn on one of my Bluetooth ear
buds to redirect the sound to there.

48
Topic ID Top Probability Words Exemplary Forum Posts Topic Label Rank1
31 money, make, lose, save, decent, fare, profit, need, spend, Both Uber & Lyft have increased numbers of riders, Unacceptable fare
care especially in our markets. However, they do not pay
drivers fare or decent wages. The costs of driving
301,699
rideshare is hidden in our depreciating and
deteriorating cars. … I have been driving less and
less...its just not worth the gas, wear and tear, and
risk at these measly earnings.
32 different, thing, app, try, exactly, mind, didnt, turn, issue, my app version suddenly changed on me: -my profile Unforeseen App
stop is gone -option to share location is gone -rider changes
complements are gone -gps voice is changed - 143,319
connection to gobank gone etc etc did anyone
experience this problem before? do you know how to
change back to previous version? I had this happen..
33 car, clean, wear, eat, smell, leave, inside, food, put, keep Dear Paxholes: My car is not an eating place ... Car condition
Please!!! No food, no drinks, no smoking, no
hairspray. Cleaning up spills and working to get rid 95,472
of food, smoke, odors, and dirty windows, costs me
time and lost income. Please eat, drink, and apply
your makeup somewhere else …
34 income, earning, hour, guarantee, gross, number, dollar, Postmates 65 rides for $550 guarantee, just finished. Income
total, profit, average Anyone else? Decided to try and see how long it
would take, and what the tips would be like on
Postmates. They do not give you a bonus, they
guarantee your income after completing 65 trips and 311,769
maintaining above a 4.5 star rating. It took me 6 days
to hit the required number of rides. I found that tips
are better and you do stay busy. At the end I had
$480, so they will wait for 3 days to see about tips
coming in, then cash me out at $550
35 people, black, man, kid, woman, life, family, world, school, This afternoon I had a ride with a nice young family. Types of riders
white 2 young kids and one on the way. It was a minimum
310,475
fare to Walmart and they were getting a few things
for the baby. …
36 wait, minute, start, ready, soon, get, ride, ping, within, quick I'm new to Uber after delivering for Postmates for a Job availability
few months. I run both platforms, but in my opinion,
110,669
Uber does a better job assigning jobs for bicycle
deliveries. For example, Postmates will ping for a

49
 Topic ID Top Probability Words Exemplary Forum Posts Topic Label Rank1
delivery from Market Square to the other side of
Oakland. This cuts down on the amount of deliveries
I can make because of the distance traveled. Uber, on
the other hand, seems to assign shorter deliveries,
which is great when delivering by bicycle. For
example, I get a ping for a delivery from Market
Square to Smallman Street. I can take more
deliveries per evening, than I can with Postmates.
37 something, anything, someone, know, else, anyone, say, Anyone know any car auction places near the city of General questions
need, somewhere, maybe Boston that is open to the public to bid on? Any
410,780
advice for a first timer going? If you could, provide a
link or website so I can check it out. Thanks!
38 ask, tell, question, answer, call, explain, straight, let, need, if you get a human on support your lucky, normally Automated
bot you get a bot reply for the first couple of messages messaging
which picks up keywords, …
After four attempts, Uber still hasn't answered this
42,768
email (below). I didn't even receive their usual
"request received" robomail. I then sen a test email,
and I instantly got a robo response. Sent original
once more, then nothing.
39 insurance, personal, accident, policy, risk, cover, rideshare, I received the notice that my insurance policy was Insurance
vehicle, claim, fault expiring in <30 days, and to upload my new
information. No problem, I do that, and that's when 52,502
everything went weird. They rejected it, saying that
the policy has not reached it's start date! …
40 see, mile, check, speed, think, monitor, far, know, system, Uber is monitoring us even with the app off and even App monitoring
course turns itself back on in the background heres proof. I
shut the app off and put it to sleep the battery
70,011
monitor shows when I start moving my phone
around and then the red shows when I start driving
…

Appendix D- Thematic Structure of Algorithmic Management Resource Loss and Gain Passageways

50
51
if you get a human on support your lucky, normally you
get a bot reply for the first couple of messages which Automated messaging
picks up keywords, …

We work for an algorithm. Since the day I signed up for Uber Ineffective Detached relationship
there hasn't been a single face to face interaction. communication

Apparently the Lyft nanny bots are sniffing out what they THINK Algorithmic
are tell-tale signs of me doing that. After an experience the other Insufficient explanation
morning I found out what's really happening, and why. miscommunication
…so the next time an issue arises and problem solving needs
Neglecting drivers’
to occur, you use your own smarts & logic rather than
involving a system that is clueless and essentially useless. concerns

Resource Loss Passageway

Lack of voice
I was able to get through after calling for an hour straight.
According to who I spoke to, there is an issue preventing all Unresponsiveness
drivers in North America from going online

Uber is monitoring us even with the app off and even turns App monitoring
itself back on in the background heres proof. ..

Uber controls your jobs and they want to control your phone
too! Cant even multi task without it putting you offline without Overcontrolling Ongoing surveillance
noticing.

Personally, I don't think they have any right to my personal

contact information, or any other information that Uber has on Privacy concern
me (a lot). Algorithmic
constant
Ubers and Lyfts rating systems are inherently flawed. monitoring
They are however huge cost savings for both Intrusive rating system
companies….

Why are they playing games with the cancellation rate? So I've
noticed this over the last few months and I think the Scrutinising cancelation
rate
Invasive supervising
cancellation rate is being manipulated.

Been driving for only four months, got this email: "I'm
following up on feedback about your behavior during a recent 52
Nitpicking boss
ride. The feedback alleged that you asked your Passenger
personal questions…
 Exemplary Forum Posts First-order Themes Second-order Themes Aggregate Dimensions

Uber gives me the flexibility to do the important things. …. Time flexibility

Work flexibility
…The “stay within area” location filter quickly became a
driver favorite upon introducing it earlier this year … Location flexibility
Algorithmic
…. Money is earnt when demands are there. Patterns are assignment
obvious. So to make any money it's more beneficial to follow Job availability
the demands/patterns. …

Resource Gain Passageway

Providing gig
At 5:20PM today, I received 6 consecutive POOL pings from
15+ minutes away (max was 26 minutes). Task allocation

Does Uber navigation take into account traffic? I noticed on a recent

trip from Everett to downtown Seattle when traffic was bad…..take a Traffic information
side road then back on I5 before I hit downtown. It did seem like a
quicker route because of traffic

I've used Uber navigation since day one without needing the
voice ( which is easy to turn on and off), and it works Navigation advice Information provision
perfectly.

I finally created excel worksheets with the info that Uber

provides. I searched on how are the promotions that we sign Algorithmic
Performance feedback
up for paid and it says that it will be either a Miscellaneous or
Other payment.
support
True to their word bonus has been paid out on the 27 th.
Enjoy everyone do something useful with it. My $140 will be Benefit/ Extra pay
going towards a meal out tonight .
Reward provision
Postmates 65 rides for $550 guarantee, just finished. Anyone
else? Decided to try and see how long it would take, and what Income
the tips would be like on Postmates.

53
 Appendix E- Survey Instrument

Construct Items
Algorithmic assignment [Platform] allows me to make my own decisions about how to schedule my
work.
[Platform] gives me a considerable opportunity for independence and
freedom in how I do the work.
[Platform] allows me to plan how I do my work.
[Platform] gives me the opportunity to select my tasks.
Algorithmic Information [Platform] shares information about my performance with me.
support provision I am informed in advance of changing needs of clients.
In [platform], any information which might help me will be provided.
The platform keeps me informed about the events or changes that may affect
my job.
Reward I earn money when I work in [platform].
provision Working in [platform] benefits me financially.
Working in [platform] can improve my economic situation.
algorithmic ambiguity [Platform] procedures in assigning my tasks are not consistent.
[Platform] procedures in evaluating my tasks are not clear.
The task assignment and evaluation procedures of [platform] are biased.
[Platform] task assignment and evaluation procedures are based on accurate
information ®
I cannot understand how [platform] assigns jobs to me.
Algorithmic unfairness The reward I get from [platform] does not reflect the effort I have put into
my work.
Usually, the reward I get from [platform] is not appropriate for the work I
have completed.
The jobs that [platform] assigns to me are not fair.
Considering my performance, the reward [platform] gives me is unfair.
Algorithmic miscommunication [Platform] cannot help me understand and resolve my problems.
I cannot count on [platform] if I need advice to make a decision or to cope
with a problem.
I cannot ask [platform] for advice, because it does not understand me.
Algorithmic constant monitoring [Platform] watches me carefully to control my task completion.
[Platform] closely monitors me while I’m trying to do my work.
[Platform] inspects my work closely to see that it measures up to applicable
standards.
I feel that I am constantly being watched to see that I obey all rules of
[platform].
[Platform] ensures that our behaviors are being closely monitored at all times
Burnout I feel emotionally drained from working for [platform].
I feel used up at the end of the day.
I dread getting up in the morning and having to face another day of working
for [platform].
I feel I’m working too hard on my job at [platform].
My work really puts a lot of stress on me.
Engagement Vigor At my work in [platform], I feel bursting with energy.
At my job in [platform], I feel strong and vigorous.
When I get up in the morning, I feel like going to work for [platform]
Dedication I am enthusiastic about my job in [platform].
My job inspires me.
I am proud of the work that I do at [platform]
Absorption I am immersed in my work.
I get carried away when I’m working.
I feel happy when I am working intensely.

54
Appendix F- Measurement Model

We assessed the measurement model by testing content, convergent and discriminant validity.
Content validity was examined by ensuring alignment between the measurement items and the
established literature. This evaluation was conducted during the questionnaire design phase.
We asked the experts to provide feedback on whether the items are relevant and representative
of the constructs (Kerlinger & Lee, 2000). We assessed the convergent validity to check that
the measures of the same construct are more correlated with each other than with the measures
of another construct (Petter et al. 2007; Schwab, 2005). As presented in Table F.1, our factor
analysis confirms convergent validity. Factor loading values (shaded) show that there is a
strong correlation between each of the items and their corresponding construct. We further
verified the convergent validity using four criteria: internal consistency of constructs (>.7),
composite reliability (CR>.7), Cronbach's alpha (CA>.7), and average variance extracted
(AVE>.5). Table F.2 presents the mean, standard deviation, CA, CR, AVE for the constructs
which values confirm satisfying convergent validity.

As recommended by Fornell and Larcker (1981), we assessed the discriminant validity by

evaluating the cross-loadings in factor analysis as well as examining square root values of the
AVEs. The cross-loadings of the items on other constructs (Table F.1) exhibit low values,
thereby confirming discriminant validity. Additionally, the values of the square root of the
AVE, as presented on the diagonal in Table F.2, surpass the interconstruct correlations (off-
diagonal entries in Table F.2) underscoring the satisfactory discriminant validity of
measurement model.

Table F.1. Factor Analysis

HOC Algorithmic Support
Factors AAs IP RP APA AU AM ACM BUR
AAs1 0.77 0.05 0.13 0.07 0.05 0.02 -0.13 0.06
AAs2 0.78 0.07 0.06 0.05 -0.09 0 -0.01 0.12
AAs3 0.80 -0.04 0.09 0.03 -0.02 -0.03 0.07 0.01
AAs4 0.85 0.15 0.11 -0.06 -0.03 0.05 -0.04 -0.07
IP1 -0.02 0.83 0.03 0.03 -0.13 -0.01 -0.01 -0.01
IP2 0.11 0.84 0.17 0.12 -0.02 0 0.03 0.09
IP3 0.12 0.91 0.15 0.09 -0.06 0.03 0,05 0,02
IP4 0.08 0.87 0.03 0.04 -0.03 0.12 0.17 -0.03
RP1 -0.01 -0.06 0.92 0.05 0.08 0.05 0.1 -0.01
RP2 0.15 0.05 0.90 0.05 0.01 0.05 0.01 0.04
RP3 0.18 0.21 0.88 0.16 0.05 -0.08 -0.03 0.06
AA1 0.11 0.16 -0.03 0.78 0.08 0.05 0.04 0.13
AA2 -0.03 0.06 0.05 0.88 0.06 0.03 0.09 0.05
AA3 0.01 0.09 0.09 0.75 0.02 0.05 0.11 0.08
AA4 0 -0.03 0.06 0.77 0.03 0.01 0 -0.09
AA5 -0.05 0.05 0.03 0.83 0.17 0.04 0.07 0.06
AU1 0.14 0.13 0.08 0.01 0.92 0.15 -0.03 0.03
AU2 0.09 0.07 -0.03 0.11 0.85 0.04 0.01 0
AU3 -0.03 0.18 0.06 -0.06 0.90 0.09 0.03 -0.06
AU4 0.07 0.05 -0.09 0.05 0.86 0.01 0.13 0.02
AM1 0.04 0.09 0.17 0.01 0.09 0.75 0 -0.01
AM2 -0.05 0 0.04 -0.05 0.06 0.80 0.03 0.11
AM3 0.09 -0.03 0.08 0.02 0.03 0.77 0.01 0
AO1 0.17 0.02 -0.14 -0.05 0.02 0.04 0.74 -0.04
AO2 0.02 0.06 0.08 0.03 -0.05 0.15 0.81 0.14
AO3 0.03 0.09 -0.06 0.07 0.12 0.06 0.76 0.07
AO4 0.06 0.05 0.02 -0.02 -0.02 -0.02 0.83 0

55
 AO5 -0.06 -0.01 0.12 -0.01 -0.05 0.05 0.82 0.02
BUR1 0.04 0.13 -0.01 0.17 0.04 0.01 -0.04 0.85
BUR2 -0.07 -0.11 -0.11 -0.03 0.07 0.01 -0.01 0.93
BUR3 0.05 0.01 -0.02 0.06 0.02 0.21 0.09 0.92
BUR4 0.09 -0.05 0.07 -0.02 0.14 0.03 0.12 0.88
BUR5 0.1 0.02 -0.02 0 0.06 -0.07 0.16 0.87

Note: HOC= Higher Order Construct; AAs= Algorithmic Assignment; IP= Information Provision; RP= Reward Provision; AA=
Algorithmic Ambiguity; AU= Algorithmic Unfairness; AM= Algorithmic Miscommunication; ACM= Algorithmic constant monitoring;
BUR= Burnout

Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and
measurement error. Journal of marketing research, 18(1), 39-50.
Kerlinger, F. N., & Lee, H. B. 2000. Foundations of behavioral research. 4th Edition Worth, TX: Harcourt.
Petter, S., Straub, D., & Rai, A. (2007). Specifying formative constructs in information systems research. MIS
quarterly, 623-656.
Schwab, D. P. (2005). Research methods for organizational studies. Mahwah, NJ: Erlbaum.

Appendix G- Results of Assessing Common Method Bias

As mentioned above, we followed the guidelines of Podsakoff et al. (2003) to apply some
procedural remedies for decreasing potential CMB. For instance, on the questionnaire’s cover
page, we emphasized that data would remain confidential with the results of this survey, used
only for academic purposes and the respondents’ information kept anonymous. We further
designed the questionnaire using some reverse-coded items and diversity in item anchors.

In order to assess CMB, we used Harman’s one-factor test (Podsakoff et al., 2003) and marker-
variable technique (Lindell & Whitney, 2001). This test assesses the threat of CMB by
examining if a single latent factor emerged from the factor analysis or if one general factor
accounted for the majority of the covariance in the independent and dependent variables. For
our sample, unrotated factor analysis indicated that the first factor accounted for 28.4% of the
total variance. Because a single factor did not emerge and one general factor did not account
for most of the variance, we gained preliminary evidence that the measurement model was

56
robust to CMB (Podsakoff et al. 2003). Due to possible limitations of Harman’s one-factor test,
we additionally employed Lindell and Whitney’s (2001) marker variable assessment technique.
The low correlation between the main constructs and the marker variable (here enjoying
holidays) confirms that the CMB was not an issue. Collectively, across these criteria, we can
conclude that CMB does not present a significant threat to the study.

Lindell, M. K., & Whitney, D. J. (2001). Accounting for common method variance in cross-sectional research
designs. Journal of applied psychology, 86(1), 114.
Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003). Common method biases in behavioral
research: a critical review of the literature and recommended remedies. Journal of applied
psychology, 88(5), 879.

57