Determine whether observed discriminatory features reflect causal mechanisms or spurious correlations

Establish whether the associations between textual, visual, and structural advertisement features and forced-labour outcomes reflect causal mechanisms of deceptive recruitment rather than spurious correlations or resource-related design differences.

Background

The empirical analysis identifies strong correlations between advertisement features and exploitation outcomes but does not demonstrate that those features cause or directly explain deceptive recruitment. Simpler vocabulary, poorer imagery, and other quality deficiencies may reflect resource constraints or design choices shared by legitimate low-resource employers.

The paper therefore leaves unresolved whether the model's most important signals are genuine indicators of deception or artefacts of the sampled populations and data-collection process. Domain-expert qualitative error analysis is proposed as a way to distinguish authentic indicators from artefacts and strengthen causal understanding.

References

The analysis identifies strong correlations between features and exploitation outcomes but does not claim causal mechanisms; observed patterns may reflect resource constraints or design choices rather than deception per se. Similarly, YOLO object detections contributed minimally to model performance (5.0\% SHAP), and specific COCO classes like boats and backpacks showed no significant univariate differences, and their inclusion in prior qualitative characterisations was not supported empirically.

— Detecting Deceptive Recruitment: A Signal-theoretic Machine Learning Framework for Early Identification of Labour Exploitation  (2609.20336 - Siraj et al., 17 Sep 2026) in Section 5.4, Limitations: Causality and spurious correlations; Section 5.7, Conclusion