What the study analyzed
Federal workforce responses to the 2025 return-to-office mandate, drawn from three Federal News Network pulse surveys of 16,526 unique respondents. As the authors note, manual coding alone of a corpus this large would have been impractical, so computational support was required to analyze the data systematically while maintaining interpretive rigor.
The open-access article traces how frustration and anger rose from 11% of comments in the first wave to a third of comments after the mandate, alongside growing concerns about performance and autonomy. The authors are Geneviève Morin, Carey Doberstein, and Étienne Charbonneau, Canada Research Chair in Comparative Public Management.
How Evidano fit the workflow
A documented, human-in-the-loop workflow
- 1Import
All 39,788 comments were imported into Evidano for AI-assisted coding.
- 2Code against your codebook
The tool coded comments against the researchers’ predefined thematic grid of three domains, subdomains, and subthemes, clustering passages by meaning. It produced a coding matrix with exemplar excerpts and text-level justifications for each decision.
- 3Human validation
The team reviewed classifications and manually verified a random 15% sample. No major misclassifications remained after refinement, and no themes outside the framework emerged.
- 4Interpret
Researchers interpreted the validated themes within their conceptual framework. The tool organized material; the analysis stayed with the team.
“Rather than relying on keyword matching, AILYZE identifies conceptual patterns based on contextual meaning, clustering related passages even when they do not share identical vocabulary.”
“This design ensured that AI augmented rather than replaced expert qualitative judgment, limiting automation bias while preserving analytical rigor.”
“The AI tool did not generate interpretations or theoretical claims; its function was restricted to organizing thematic material, which was subsequently interpreted through the conceptual framework and validated by the research team.”
“For scholars, the study demonstrates the value of AI-assisted qualitative analysis for examining large-scale administrative phenomena while retaining interpretive depth.”
From bottleneck to scalable analysis
Coding tens of thousands of comments by hand is impractical. Coder drift and fatigue build up at scale. Reproducing the exact process is difficult.
Codes the full corpus against your codebook. Deterministic outputs reduce drift and fatigue. Researchers review, validate, and stay in control.
Researchers reviewed and refined every classification.
Text-level justifications enabled auditing and contestation.
Deterministic, version-locked inference yields identical outputs.
Published under a CC BY licence, so the full protocol and validation log are there for anyone to check.