####################################################################
THIS REPOSITORY IS NO LONGER BEING MAINTAINED. PLEASE USE THE InterventionEvaluatR PACKAGE https://github.com/weinbergerlab/InterventionEvaluatR
#######################################################################
##NOTE: UPDATED VERSION The code here is the same as that used for Bruhn et al. PNAS and uses the BoomSpikeSlab sampler for MCMC. However, it performs linear regression on the log-transformed counts, rather than Poisson, which might not be ideal for some count data, particularly sparse data. We have an updated version of the code here: https://github.com/weinbergerlab/synthetic-control-poisson that uses a Poisson model with overdispersion and also performs STL+PCA, which is ideal for sparse data (Shioda et al. Epidemiology) and performs K-fold cross validation and Bayesian stacking.
#GETTING STARTED: #Save all of the files to your desktop and open the synthetic_control_run.R file. Set the working directory as the "main analysis components" folder. To use the sample Brazil data (found in Datasets for PNAS), use all of the default settings, or specficy your data file. Change the dates as needed to set the training and evaluation periods.
##Analysis files If you want to dig deeper, the functions and analysis code are found in the folder "main analysis components." The analysis code is split into several different files. The code in "synthetic_control_run.R" provides a template for defining initial constants, running analyses, and generating reports. The minimum files needed to run an analysis are the "synthetic_control_analysis.R" and "synthetic_control_functions.R", which contain the analysis code and functions necessary to run the analysis. Make sure these files, along with any additional code files, are in the same directory. See the table below for a short description of each file.
#Can also run stl-pca method of shioda et al https://www.biorxiv.org/content/early/2018/04/27/302224 #TROUBLESHOOTING 1)Make sure date formaton .csv file is correct. It needs to be "YYYY-MM-01". 2) Check to make sure working directory is set correctly. It should point to the directory where the .Rmd file is found 3) make sure output directory, file names are specified correctly on the run.R file
| File Name | File Dependencies | Details |
|---|---|---|
| synthetic_control_run.R | synthetic_control_analysis.R, synthetic_control_functions.R, synthetic_control_plot.R, synthetic_control_write_results.R, synthetic_control_report.R, synthetic_control_report.Rmd | A template for running analyses and generating an html report. User specifies variable names, dates, etc, and the program calls dependent files. |
| synthetic_control_analysis.R | synthetic_control_functions.R | The main analysis code. |
| synthetic_control_functions.R | N/A | Contains functions for the main analysis. |
| synthetic_control_plot.R | synthetic_control_analysis.R, synthetic_control_functions.R | Plots results from analysis. |
| synthetic_control_write_results.R | synthetic_control_analysis.R, synthetic_control_functions.R | Writes results from analysis to CSV files. |
| synthetic_control_report.R | synthetic_control_analysis.R, synthetic_control_functions.R, synthetic_control_plot.R, synthetic_control_write_results.R, synthetic_control_report.Rmd | Handles package dependencies and prepares arguments for "synthetic_control_report.Rmd". |
| synthetic_control_report.Rmd | synthetic_control_analysis.R, synthetic_control_functions.R, synthetic_control_plot.R | Generates an HTML report of analysis results containing tables and graphs. |
| synthetic_control_set_vars.R | synthetic_control_analysis.R, synthetic_control_functions.R, synthetic_control_plot.R, synthetic_control_write_results.R, synthetic_control_report.R, synthetic_control_report.Rmd | A more complicated example of how the constants can be defined for multiple regions. Demonstrates how reports can be generated in a loop. |
##Data
To prepare data that can be used for analysis, make sure the date column is in the format YYYY-MM-DD. There should be only column that designates group stratification. If you wish to have multiple layers of stratification, combine the group columns into one column (e.g., region and age group columns could combine to a single age group-region column). The outcome and covariate columns (i.e., every column other than the date and group columns) should be numeric. The file type must be .csv. Sample datasets are provided as well - see below for more details.
###Sample Data
The sample data, found in the folder "Datasets for PNAS", are CSV files formatted for use with the synthetic control analysis. There are five files, each containing data from different countries. The data are subsetted by age group, which can be translated using the list below. In addition to the analysis files listed in the previous sections, the applet files can be found in the folder labeled "synthetic_control".
| Numeric Representation | Age Group |
|---|---|
| 0 | < 3 months |
| 1 | 3-<12 months |
| 2 | 12-24 months |
| 3 | 24-59 months |
| 4 | 5-<18 years |
| 5 | 18-<40 years |
| 6 | 40-<65 years |
| 7 | 65-<80 years |
| 8 | 80+ years |
| 9 | <12 months |
For the US, cells with fewer than 10 counts are replaced with 9999 due to privacy considerations.
###SHINY app For a point-and-click interface, there is a Shiny app (ShinySC) that can be either run in RStudio (ShinySC_Local_Deploy folder) or deployed on the web (ShinySC_Web_Deploy). Open the server.R or ui.R files in Rstudio and click "Run App" button Note: the Shiny version of the code is non-parallel, so it is computationally slower than the synthetic_control_run.R version. You can access a version of the app at https://weinbergerlab.shinyapps.io/ShinySC/, but please note that this deployment is often unstable.