# Replication package: Acolin, Decter-Frain, and Hall 2022

The data processing and analysis for this article were produced using R 4.2.1 and Stata 15.3.
The raw consumer data cannot be shared but we share the processed tract, ZCTA and county estimates along with the Census and ACS
variables used as part of the analysis.
The data shared for replicability are free from non-disclosure or other confidentiality constraints.

The raw data is available for purchase from Data Axle (https://www.data-axle.com/) under commercial license.
For this analysis the following variables from the raw data from 2006 to 2020 were used:
-Consumer Unit ID: identify each individual consumer unit
-Location ID: identify each individual housing unit
-Latitude and longitude of the unit

The shared files include:
Under Data Folder: 
Controls Subfolder: ACS and Census data used as control to adjust the raw estimates at tract, ZCTA, county and state level
Data Axle Subfolder: consumer unit and household estimates based on Data Axle data at tract, ZCTA, and county level

Under Code Folder:
descriptives_replication: the code used to produce Fig. 1-2, Fig. 7 and Table 2.

Under Section 5 and 6:

This folder contains the code and data for replicating sections 5 and 6 of the paper, along with all related appendices

The folder has the following structure:

data/
	- Contains all raw and intermediate data
	- Subfolders:
		* acs/
			- All raw ACS data
		* decennial/
			- Contains all raw data from the 2010 Decennial Census
		* errors/
			- Contains the errors from all the adjustment models fit in Section 5
		* infogroup/
			- Contains all counts of consumer records from Data Axle
		* lookups/county_name_lookup.csv
			- A lookup file to get county names from FIPS codes for plotting
		* prepped_data/
			- Contains outputs from scripts for cleaning raw data
		* projections/
			- Contains estimates from applying calibration models in section 5 to Data Axle data from all years.

functions/utils.R
	- Contains utility functions for fitting calibration models and projecting them

models/
	- Contains fitted multilevel calibration models produced by brms. We store them here because it takes a long time to re-fit them

notebooks/
	- Contains jupyter notebooks with all the core analyses
	- Files:
		* adjust_counties.ipynb
			* Contains code for the county-level calibration model presented in section 5
		* adjust_counties_oos.ipynb
			* Contains code for the county-level calibration model with out-of-sample validation as presented in the Appendix
		* adjust_tract.ipynb
			* Contains code for the tract-level calibration model presented in section 5
		* adjust_tract_oos.ipynb
			* Contains code for the tract-level calibration model with out-of-sample validation as presented in the Appendix
		* adjust_zcta.ipynb
			* Contains code for the zcta-level calibration model presented in section 5
		* adjust_zcta_oos.ipynb
			* Contains code for the zcta-level calibration model with out-of-sample validation as presented in the Appendix
		* check_benchmarking_property.ipynb
			* Contains code for checking the benchmarking property as presented in the Appendix
		* nowcasting.ipynb
			* Contains code for implementing 2-stage nowcasting as presented in section 6
		* plot_***
			* Notebooks for producing different plots presented in the main text and appendix

scripts/
	- Contains three scripts for producing datasets at the county, zcta, and tract level that are ready for further analysis