Teacher-GenAI Collaboration - Data and Code Package
Description
# Teacher-GenAI Collaboration - Data and Code Package (version 2, revision R1) This package accompanies the manuscript "Teacher-GenAI collaboration: A mixed-methods structural analysis of perceived functionality, ethical-pedagogical concerns, and institutional expectations." ## Contents - `Data.csv` - anonymized survey data (N = 406 teachers, Türkiye, May 2025). To protect participant privacy, the submission timestamp has been replaced by a random participant ID and the open-ended free-text column has been emptied (`OpenEnded_Excluded`). All Likert items and demographic categories are included; column headers are the original Turkish survey questions. - `sentiment_codes.csv` - text-free data for the 66 open-ended respondents: `RowIndex` (row in the N = 403 analysis frame), `HumanSentiment` (blind human coding by a single coder - the primary sentiment data), `LLMSentiment` (output of the archived LLM pipeline, used as an auxiliary agreement check), and `WordCount` (response length). Codes: 1 = positive, 0 = neutral, -1 = negative. - `llm_sentiment_analyse.py` - the archived LLM coding pipeline with the full coding prompt. The reported LLM codes were produced on 21 July 2026 with the model identifier `gemini-flash-latest` (an alias resolving to Gemini 3.5 Flash at that time) and default sampling parameters. The script requires a GEMINI_API_KEY and the unredacted survey file and is distributed for transparency. - `Analyses.ipynb` - executable analysis notebook (Python 3.12; pandas, semopy, pingouin, factor_analyzer) reproducing the results, tables, and figures of the article, including Table 1, item documentation for all 25 items, model comparison with log-likelihood-based AIC/BIC, HTMT with bootstrap confidence intervals, and the sample, item-set, and model-specification sensitivity analyses. - `Sensitivity_lavaan.R` - estimator sensitivity analyses (MLR, WLSMV with ordered items, FIML) in R/lavaan. ## Reproduction Keep all files in one directory. (1) Run `Analyses.ipynb` top to bottom; it reads `Data.csv` and `sentiment_codes.csv` and writes tables, figures, and the prepared samples in `sensitivity_inputs/`. (2) Run `Rscript Sensitivity_lavaan.R`. Raw open-ended texts are withheld for privacy and are available from the corresponding author on reasonable request. ## Version history - Version 2 (revision R1): AIC/BIC computed from the ML log-likelihood (version 1 reported values derived from the semopy fitting objective); HTMT computed from absolute correlations with bootstrap CIs; workload categories split on `;` and subject entries harmonized so that Table 1 is reproduced directly; item documentation for all 25 items; data-quality screening and additional sensitivity analyses; LLM run details documented. ## License and citation Data are released for scholarly use. Please cite the article and this dataset (Mendeley Data).