# OSF repository for manuscript "Followers Forever - Prior Commitment Predicts Post-Scandal Support of a Social Media Celebrity" This repository contains all code and data necessary to computationally replicate the analysis provided in the manuscript. Please note that, for data privacy reasons, we cannot provide the raw data, as these contain possibly identifiying information (e.g. links to author youtube channels). For this reason, we anonymized all authorIDs. Analysis and data is structured within the following folders. ## data The "data" folder contains the following subfolders and data files: - Cleaned: cleaned datasets of comments on the apology video (LP_AP_clean.csv) and the pre-scandal videos (LP_pre_combined). - ddr_results: output from the DDR analysis for the apology video (LP_AP_NCFL_YTcorpus.tsv), the pre-scandal videos (LP_PRE_YTcorpus.tsv). - Hand_Coding: the hand coded sample of 1000 apology video comments, with the individual coder results, as well as the combined coding (LP_post_1000_coding_composite_FINAL.xlsx) - Processed: this contains data files that resulting from various steps of the analysis We provide a **codebook** for the main datasets in the scripts folder. ## DDR contains all ddr scripts, within two subfolders ### DDR-master Contains the ddr source code used for creating the word vectors files, (see specific readme). Also contains the folder RUN_DDR, which contains all dictionaries used for analysis, as well as the different word vector models, and output generated from the DDR analysis. This folder also holds the Run_DDR.ipynb jupyter notebook file allowing to rerun the ddr analysis. Please note: we do not provide the Google News corpus used for the benchmark analysis, as it is too big to host it here on OSF. It can be accessed and downloaded here: ### word2vec contains a data folder with the full (cleaned) dataset of youtube comments from all videos, used to create the youtube word vector model, and two jupyter notebook files: - Create_Word2Vec_from_text.ipynb: used to create the word vector model from the youtube corpus - Word2Vec_view.ipynb: used to visualize and inspect the created word vector model. ## plots Contains plots created in the analysis ## results Contains results tables from the analysis in .html format ## scripts Contains all R scripts (in .rmd format) used for analysis. Scripts are ordered by order of execution. Please note that the first script (LP_Raw_to_clean_data.rmd) cannot be run, as raw data is not provided. The script is provided for transparency reasons, so that all data manipulations in this analysis can be retraced. Each script is commented, aiming to provide a thorough description of the analysis process. In the following we provide a short overview: 1. Cleaning - uses raw data (crawled youtube comments of the apology video, and seven pre-scandal videos) to create cleaned datasets. As outlined above, this script cannot be run, as raw data is not provided for data protection reasons. 2. DDR prep - creates the corpora of youtube comments used in the DDR analysis (produces two .tsv files, one for the apology video, one for the pre-scandal videos) 3. DDR_benchmark - used for the benchmark comparison of the Youtube corpus vs. the Google News corpus. Shows that Youtube corpus word vector model performs better than Google News model 4. Intercoder Reliability - calculates intercoder reliability for the handcoded dataset 5. DDR classification - classifies comments according to the dictionary loadings and hand coded dataset, creates datasets used for final analysis 6. Analysis - main analysis script. Runs all models reported in the manuscript, as well as creating all plots and summary tables Please note a few translational changes from R scripts to manuscript: - "devotion" is used to refer to the concept of commitment in the manuscript - "Fan Language", or "LP_fan", refers to the concept of social identity or fan identity in the manuscript - "sentiment" is refers to liking, as discussed in the manuscript