Analysis code and data for the quality survey experiment in "Could AI Leapfrog the Web? Evidence from Teachers in Sierra Leone"

Code and data for the human quality comparison in "Could AI Leapfrog the Web? Evidence from Teachers in Sierra Leone." Teachers recruited on Prolific rated AI chatbot responses and web search results for the same queries, without knowing which was which. The repository reproduces Figure 8 and the Appendix D robustness checks (Figures S11 and S12) from anonymized survey responses. The queries submitted by teachers in Sierra Leone and the associated teacher surveys were promised to be confidential and are not included; they can be shared under a data sharing agreement available from Fab Inc. (Paul Atherton, [e-mail]). Python (position and length analyses) and R (Figure 8) code are included; see README.md. Version 1.1 adds the analysis code for the paper's other figures and tables (code/confidential_data_required/); the inputs for those scripts are not included. Version 1.2 restores the Figure 2 script (version 1.1 contained a copy of the Figure 3 script in its place), builds the Figure 2A denominators on the complete grid of months, corrects the rolling-window retention section and adds the Table S1 regression to attrition_regression.py, preserves missing answers in the Yes-or-Uncertain outcome of the Figure 8 script, renames the WhatsApp overhead script to whatsapp_overhead_regression.py, and adds a section on scope and known limitations to README.md.