Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

When Algorithms Manage Humans: A Double Machine Learning Approach to Estimating Nonlinear Effects of Algorithmic Control on Gig Worker Performance and Wellbeing

Arunkumar V, Nivethitha S, Sharan Srinivas, Gangadharan G. R

TL;DR

This study addresses how algorithmic management conditions the effectiveness of person-centered HR practices in gig work by estimating a nonlinear moderated mediation via CNIE using Double/Debiased Machine Learning. It finds a stable, positive indirect effect on wellbeing through the relational psychological contract across all levels of algorithmic control, while the indirect effect on performance follows a U-shaped pattern—strong at low and high control but weakest in the murky middle where clarity is lacking. The work demonstrates that conventional linear models can misestimate or even reverse such effects, highlighting the value of flexible causal inference in sociotechnical settings. Practically, it offers design guidance: either preserve autonomy or implement clear, transparent, and accountable oversight to sustain performance and protect wellbeing, with methodological implications for future research in algorithmic HRM.

Abstract

A central question for the future of work is whether person centered management can survive when algorithms take on managerial roles. Standard tools often miss what is happening because worker responses to algorithmic systems are rarely linear. We use a Double Machine Learning framework to estimate a moderated mediation model without imposing restrictive functional forms. Using survey data from 464 gig workers, we find a clear nonmonotonic pattern. Supportive HR practices improve worker wellbeing, but their link to performance weakens in a murky middle where algorithmic oversight is present yet hard to interpret. The relationship strengthens again when oversight is transparent and explainable. These results show why simple linear specifications can miss the pattern and sometimes suggest the opposite conclusion. For platform design, the message is practical: control that is only partly defined creates confusion, but clear rules and credible recourse can make strong oversight workable. Methodologically, the paper shows how Double Machine Learning can be used to estimate conditional indirect effects in organizational research without forcing the data into a linear shape.

When Algorithms Manage Humans: A Double Machine Learning Approach to Estimating Nonlinear Effects of Algorithmic Control on Gig Worker Performance and Wellbeing

TL;DR

This study addresses how algorithmic management conditions the effectiveness of person-centered HR practices in gig work by estimating a nonlinear moderated mediation via CNIE using Double/Debiased Machine Learning. It finds a stable, positive indirect effect on wellbeing through the relational psychological contract across all levels of algorithmic control, while the indirect effect on performance follows a U-shaped pattern—strong at low and high control but weakest in the murky middle where clarity is lacking. The work demonstrates that conventional linear models can misestimate or even reverse such effects, highlighting the value of flexible causal inference in sociotechnical settings. Practically, it offers design guidance: either preserve autonomy or implement clear, transparent, and accountable oversight to sustain performance and protect wellbeing, with methodological implications for future research in algorithmic HRM.

Abstract

A central question for the future of work is whether person centered management can survive when algorithms take on managerial roles. Standard tools often miss what is happening because worker responses to algorithmic systems are rarely linear. We use a Double Machine Learning framework to estimate a moderated mediation model without imposing restrictive functional forms. Using survey data from 464 gig workers, we find a clear nonmonotonic pattern. Supportive HR practices improve worker wellbeing, but their link to performance weakens in a murky middle where algorithmic oversight is present yet hard to interpret. The relationship strengthens again when oversight is transparent and explainable. These results show why simple linear specifications can miss the pattern and sometimes suggest the opposite conclusion. For platform design, the message is practical: control that is only partly defined creates confusion, but clear rules and credible recourse can make strong oversight workable. Methodologically, the paper shows how Double Machine Learning can be used to estimate conditional indirect effects in organizational research without forcing the data into a linear shape.
Paper Structure (40 sections, 2 equations, 8 figures, 7 tables, 1 algorithm)

This paper contains 40 sections, 2 equations, 8 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Theoretical Framework and Hypothesis Development
  3. HR Practices as Costly Signals and Job Resources
  4. The Psychological Contract as the Mediating Mechanism
  5. Algorithmic Control as a Nonlinear Moderator
  6. Algorithmic Control Intensity
  7. Algorithmic Control Clarity
  8. The Emergence of the "Murky Middle"
  9. A Conditional Process Model with Non-Monotonic Effects
  10. Addressing Endogeneity and Alternative Mechanisms
  11. Methodology
  12. Empirical Setting
  13. Data Collection Procedure
  14. Power Analysis
  15. Non response and Sampling Bias Checks
  16. ...and 25 more sections

Figures (8)

  • Figure 1: The Proposed Conditional Process Model. HR Practices are hypothesized to influence outcomes via the relational psychological contract, with the second stage of this mediation being non-linearly moderated by Algorithmic Control.
  • Figure 2: Visual Workflow of the Double Machine Learning (DML) Estimation Process. This diagram illustrates the key steps detailed in Algorithm 1, including the cross-fitting orchestration, nuisance model fitting, residualization, and final stage estimation.
  • Figure 3: Simulation Study: Method Comparison on Root Mean Squared Error (RMSE). The chart highlights the dramatic increase in estimation error for the OLS model when its linearity assumption is violated, while DML's performance remains robust.
  • Figure 4: Visual Comparison of DML and OLS Performance Across DGPs. DML (blue) accurately recovers the true function (black dashed line) in all scenarios. OLS (red) only succeeds in the linear case (A) and imposes a misleading linear fit in the non-linear U-shaped (B) and sine-wave (C) scenarios, leading to catastrophic estimation errors.
  • Figure 5: Double Machine Learning estimates of the Conditional Natural Indirect Effect (CNIE) by algorithmic control. Panel (a) shows the indirect effect of HRP on Performance, and panel (b) shows the indirect effect on Well Being, each mediated by the relational psychological contract. The pattern for Performance is U shaped, whereas the pattern for Well Being remains positive and approximately stable. Shaded bands show 95% confidence intervals.
  • ...and 3 more figures