BKRRW_Schools

This repository contains the code required to replicate Burlig, Knittel, Rapson, Reguant, and Wolfram (2020): "Machine learning from schools about energy efficiency," conditionally accepted at the Journal of the Association of Environmental and Resource Economists. The main text of the paper can be found in BKRRW_Schools.pdf, and the online appendix can be found in BKRRW_Schools_OnlineAppendix.pdf.

Required data and file structure

Due to a non-disclosure agreement between UC Berkeley and Pacific Gas and Electric (PG&E), we are unable to release the electricity data for this project publicly. Academic researchers wishing to replicate our results can submit a request through the Energy Data Request Program. Our data request can be found in this repository in BKRRW_PGE_data_request.pdf.

The file structure for this project is as follows:

MAIN PROJECT FOLDER
|-- Code
|   |-- Analyze
|   |-- Build
|   |-- Produce_output
|-- Data
|   |-- Intermediate
|       |--  School specific
|                    |-- forest
|                    |-- double lasso
|                    |-- prediction
|       |--  Matching
|   |-- Other data
|       |--  CA school info
|       |--  SunriseSunsetHoliday
|       |--  MesoWest FINAL
|   |-- Raw
|       |-- PGE_energy_combined
|                    |-- Customer info
|                    |-- Unzipped electric 15 min
|                    |-- Unzipped electric 60 min
|       |--  PGE_Oct_2016
|   |-- Temp
|-- Results
|   |-- Appendix

The full project data folder structure is available at the JAERE section of the Harvard Dataverse here, including the fully-populated Other data folder. All other folders are empty, per our NDA with PG&E. In order to run the code described below, researchers will need to acquire the following datasets, and place them according to the below filepaths:

• Meter lat/longs (gas and electric): [MASTER PROJECT FOLDER]/Data/Raw/
• Customer information data: [MASTER PROJECT FOLDER]/Data/Raw/PGE_energy_combined/Customer info/
• Fifteen-minute-interval AMI data: [MASTER PROJECT FOLDER]/Data/Raw/PGE_energy_combined/Unzipped electric 15 min/
• Sixty-minute-interval AMI data: [MASTER PROJECT FOLDER]/Data/Raw/PGE_energy_combined/Unzipped electric 15 min/
• Fifteen-minute-interval AMI data: [MASTER PROJECT FOLDER]/Data/Raw/PGE_energy_combined/Unzipped electric 15 min/
• Matches between schools and meters: [MASTER PROJECT FOLDER]/Data/Raw/PGE_Oct_2016/
• Energy efficiency measure data: [MASTER PROJECT FOLDER]/Data/Raw/PGE_Oct_2016/
• Energy efficiency measure data, pull 2: [MASTER PROJECT FOLDER]/Data/Raw/PGE_PGE_energy_combined/Customer info/

Code

Upon obtaining data from PG&E, and populating as per the above file structure, researchers can replicate our results by running the code in the following order:

1. BKRRW_Schools/00_MASTER_set_paths.do sets all paths for use in subsequent Stata .do files. Before using this, you will need to change the master paths to match your directory structure.

2. BKRRW_Schools/Build/B00_MASTER_build_all.do runs all code to build datasets in Stata. Note that some portions of this build are run in R. Researchers will have to run the 4 .R files in the BKRRW_Schools/Build folder at the appropriate time partway through the BOO_MASTER_build_all.do file. This code takes large amounts of memory and is quite slow (ie, may take several days to run), due to the use of interval electricity metering data.

3. BKRRW_Schools/Analyze/A00_MASTER_analyze_all.do runs all analysis code in Stata.

4. BKRRW_Schools/produce_output/PO00_MASTER_produce_output_all.do generates all tables (in LaTeX format) and figures (in PDF format) in Stata for both the main text and the appendix. Appendix Figure C.1: "Locations of untreated and treated schools" must be built in R using the file PO05_MASTER_make_map.R.

The BKRRW_Schools/Build, BKRRW_Schools/Analyze, and BKRRW_Schools/Produce_output folders contain all required sub-programs.

Contact

If you have remaining questions about the code described here, please contact Fiona Burlig or Mar Reguant.