**Cite as:**
Stachl, C., Au, Q., Schoedel, R., Samuel. D. Gosling, Gabriella. M. Harari., Buschek, D., Völkel, S., Schuwerk, T., . . . Bühner, M. (2020). Predicting Personality from Patterns of Behavior Collected with Smartphones. Proceedings of the National Academy of Sciences of the United States of America (PNAS). https://doi.org/10.1073/pnas.1920484117
### Interactive website and data-dictionary:
[**>> Go to the website (1-2 minutes loading time) <<**][1]
If you experience problems with the website please send us a notice (<clemens.stachl@unisg.ch>) so we can fix the issue as soon as possible.
Disclaimer: the website needs to cache a lot of data in the background. This may take some minutes.
If the website fails to load, or you are disconnected from the server - you can also run the website locally with this file in RStudio: OSF-Repo/Website/website.Rmd
#### Website Content:
- Behavioral Patterns
- Random Forest
- Permutation Importance and ALE Plots
- Pattern Plot (Interactive)
- Elastic Net
- Beta-Coefficients
- Descriptive Statistics
- Pairwise Correlations
- Demographic and Personality Correlation
- Results Benchmark
- Results R2
- Results Pearson Correlation
- Detailed Results
- Data Dictionary
- Music
- Time
- Categories of App-Usage
- Communication Behavior
- General Phone Usage Behavior
- Mobility Behavior
- Statistical Estimators
- Predictor Overview
- Machine Learning Gist
- Random Forest
- Linear Regression and the Elastic Net
- Performance Evaluation
- Resampling
- Interpretable Machine Learning
### Available Files:
Download and unzip the "**Personality_Prediction.zip**" file to re-run the analyses.
Start the ```Personality_Prediction.Rproj```(RStudio users) or make sure to set the working directory to this folder and run ```paper.Rmd``` (to compile the pdf version of the manuscript).
This file allows to re-create the pre-print version of the manuscript.
The folder tree after unzipping:
- OSF-Repo
- **analysis**: Contains files with regard to the analysis of the obtained results from machine learning. These files will create descriptive statistics, ALE plots, and tables with aggregated results.
- **data**: Data for the benchmark and analyses.
- modeling
- external_data (app categories from google sheets)
- results
- benchmark
- feature_importance (batchtools registries and single csvs with variable importance values)
- grouped_importance (results from the calculation of semantic/grouped importance)
- selfreports (raw questionnaire item values and aggregated person parameter scores)
- sensing_data (all computed variables)
- raw_data: this folder is empty (cannot be shared due to the high sensitivity of the raw data)
- **feature_extraction** (this code cannot be run! Raw data is not openly available)
- features: syntax for each feature category
- helper_functions: necessary, additional functions
- preprocessing: additional pre-processing scripts for app and music data
- **modeling**: syntax files for machine learning benchmark, variable importance and grouped variable importance
- Packrat (contains a packrat lock-file, with all packages including version numbers)
- phonestudyLMU (contains a custom made, unreleased r-package to aid with feature extraction from the smartphone logs)
- Website (all files to reproduce the interactive website)
#### Revision 1
Download and unzip the "**Personality_Prediction_rev.zip**" file to re-run the additional analyses from the first round of revisions.
- **PNAS_paper/Revision/**: Additional code for unique and combined class importance, significance tests with PIMP, t-tests for the benchmark.
#### Revision 2
Download and unzip the "**Personality_Prediction_rev2.zip**" file to re-run the additional analyses from the second round of revisions.
- **PNAS_paper/Revision2/**: Additional code for linear mixed models and nested best subset search via greedy forward search.
[1]: https://compstat-lmu.shinyapps.io/Personality_Prediction/
[2]: http://stachl@stanford.edu
