**Overview**
These files enable the user to reproduce main analyses from the PNAS article, "Disrupted Routines Anticipate Musical Exploration".
User-level data are not available so that streamers cannot be identified by analysts. The study design was designated exempt under the University of Chicago Institutional Review Board (Social Sciences Division), but we cannot share anonymized user-level data. Using a federated learning strategy, we share: (1) code for how to turn the original raw data into models, (2) the fitted models, and (3) code for how to load them to produce the main analyses in the published study. See the details for file navigation and use below. All scripts contain inline comments for easy use.
**Details**
There are four folders in the OSF Storage section:
- 00_Song2Vec_model
- 01_analysis_taste_exploration
- 02_analysis_taste_adaptation
- 03_analyslis_DiD_DDD
***00_Song2Vec_model***
The first folder (00_Song2Vec_model) contains two Python scripts: Song2Vec_Model.ipynb and Song2Vec_Validation.ipynb. The Song2Vec_Model script demonstrates how we trained our S2V model using playlists data provided by Deezer. Although we are not authorized to share the playlist data, we provide the trained S2V model (see the Dropbox folder, "s2v_2018_2019_2020"). A separate Python script, Song2Vec_Validataion.ipynb, enables loading of the model and demonstrates the validation examples for our S2V model, as presented in Figure S5. One can replicate the examples by running this script.
***01_analysis_taste_exploration***
The second folder (01_analysis_taste_exploration) contains two STATA data and program files: analysis_01_a.dta and analysis_01_script.do. The .dta (data) file contains user-month level panel data, and the .do (program) script runs the main regression models regarding our analysis of `taste exploration`, `taste resonance`, `granger-causality`, and `radical exploration` using the panel data.
***02_analysis_taste_adaptation***
The third folder (02_analysis_taste_adaptation) contains two STATA data and program files: analysis_02_a.dta and analysis_02_script.do. The .dta (data) file contains user level data that shows the taste distance and geographical distance between a user's home city and visited cities in a given month. The .do (program) script runs the main regression models regarding our analysis of `taste adaptation` to cities.
***03_analyslis_DiD_DDD***
The fourth folder (03_analysis_DiD_DDD) contains two STATA data and program files: analysis_03_a.dta and analysis_03_script.do. The .dta (data) file contains user-week level variables of those users from South Africa and Australia during our observation period (i.e., the first 24 weeks of 2020). The .do (program) script runs the main DiD and DDD models regarding our analysis of `taste exploration during COVID-19 lockdowns` between South Africa and Australia. Furthermore, this script allows one to replicate the statistics and figures for our pararllel-trands and granger causality tests.
