The Online Paradox: Why Software Practitioners Feel Less Safe Learning in Online Settings

This bundle contains supplementary materials for an upcoming academic publication, The Online Paradox: Why Software Practitioners Feel Less Safe Learning in Online Settings, by Christiaan Verwijs, Evelien Acun-Roos and Daniel Russo. Included in the bundle are the dataset, Structural Equation Models (SEM), Gioia schemas, and a codebook for the qualitative coding. This replication package is made available by C. Verwijs under a "Creative Commons Attribution Non-Commercial Share-Alike 4.0 International"-license (CC-BY-NC-SA 4.0). About the study As software engineering work shifts toward distributed and hybrid models, professional learning increasingly occurs through online, peer-based platforms alongside traditional face-to-face settings. While digital environments expand access to peers and expertise, their impact on the psychological conditions that support learning remains unclear. Psychological safety — the shared belief that one can speak up, ask questions, and disagree without fear of embarrassment or exclusion — is central to learning and innovation, yet underexplored across interaction modalities in software engineering. This mixed-methods study examines how 143 software practitioners experience psychological safety in online and face-to-face learning interactions. We use a mix of quantitative and qualitative data to investigate how psychological safety varies between modalities and demographic variables, and what contributes to low psychological safety in the experience of practitioners Results show that self-reported psychological safety is substantially lower in online than in face-to-face interactions. Practitioners describe exclusionary behaviors, negative interaction patterns, and hostility as primary threats, and typically respond by reducing contributions, avoiding interpersonal risks, and withdrawing altogether. Member checking was used to validate interpretations, and involve practitioners directly in the sense-making of results, surfacing explanatory mechanisms and implications grounded in professional experience. These findings suggest an online paradox: environments designed to broaden participation may simultaneously make interpersonal risk-taking feel less safe. The study offers an evidence-based characterization of how interaction modalities shape learning-related behaviors in contemporary software work. Psychological safety is positioned as a condition that enables, rather than suppresses, constructive disagreement, and the results inform practitioners and organizers designing hybrid and online learning settings in software engineering. About the dataset The dataset for Phase I (SPSS) contains anonymized response data from 143 participants in the Agile Software Engineering Community. Data was gathered between December 2023 and March 2024. We cleaned the individual response data from careless responses and removed all data that could potentially identify teams, individuals, or their parent organizations. Qualitative responses and response IDs from participants have been removed to protect anonymity and privacy. About the SPSS syntaxes The bundle includes the syntaxes we used to prepare the dataset from the raw import, as well as the syntax we used to generate descriptives. This is mostly there for other researchers to verify our procedure. About the codebook Thematic analysis was employed to iteratively identify, refine, and report themes in the thick response data, thereby inducing a conceptual model for the experience of psychological safety. To reinforce scientific rigor and increase replicability, we used the structured approach recommended by Gioia et. al. (2013) as an analytical framework. In the first stage, open coding was employed to analyze the responses to each essay question, identifying one or more first-order concepts that closely align with the language used by participants. In the second stage, axial coding was used to aggregate first-order concepts into related second-order themes, which were grouped into top-level aggregate dimensions in the third stage. The first and second stages are performed with as little theoretical knowledge as possible to prevent confirmation bias. A total of 11.415 words were analyzed and coded. To strengthen validity, multiple coders were involved in the coding process. The resulting codebook is included as an .odt file (generated from QualCoder). Coded cases are available in the file "data (for analysis of qualitative codes, N=130)" in directory "Phase 1".