Forecasting child welfare outcomes in the United States

This repository contains code to run projection models for the four following outcomes:

Entries into protective care
Children involved in maltreatment investigation
Exits from protective care with permanency
Exits from protective care without permanency

The outcomes are projected for six race/ethnicity groups

Non-hispanic white population
Non-hispanic black population
Non-hispanic Asian/Pacific Islander population
Non-hispanic American Indian/Alaskan Native population
Total population

The repository also contains code to create the Shiny application to explore projection results, currently hosted at: https://monica-alexander.shinyapps.io/foster_care/

Inputs

In the inputs folder there are several used to get data for the projection models.

explanatory_vars contains all potential covariates. This is created using the data processing scripts (written by Alex)
outcome_vars contains all outcomes. This is created using the data processing scripts (written by Alex)
covariates contains a list and description of covariates included in the model. This was manually created based on exploratory data analysis, discussions with Casey and model selection.
all_data_for_model contains all data required to run projection models and is an output of the script 1_load_prepare_data (see below)

Notes on data updates:

If there are data updates based on rerunning Alex's scripts then explanatory_vars and outcome_vars should be directly replaced.
If updates to the list of covariates included in the model are desired, then covariates should be updated.
If any of the above data files are updated, then all_data_for_model should be recreated (see below).

Code

In the code folder, the main scripts that need to be executed are prefixed with numbers.

1_load_prepare_data processes the data to create a file for modeling (called all_data_for_model). If there have been updates to the data, this script needs to be run.
2_run_models runs projection models for all outcome and race/ethnicity combinations and saves the results to the results folder.
3_prepare_results_for_app reads in the results and prepares them for the Shiny app, saving the output too the app folder.

If there have been updates to the data, the covariates included in the model, these scripts should be run in numerical order.

The remainder of the scripts in the scripts folder are either helper functions or the model code.

Changing the years of the model run

Currently, the model is set to run using data from 2005 to 2018, with a projection period of 5 years. If these years need to be changed, then lines 26 and 27 of the 2_run_models should be edited. For example, if 2019 is made available and is updated in the data files, the line 26 should change to

years <- 2005:2019

If you would like to project forward 7 years instead of 5 years, for example, then line 27 becomes

P <- 7

A note on the model

The statistical model is written in the Stan programming langauge, and the file is contained in the code/models folder. Changes to the model will require editing this file and probably also the two functions (contained in the code/functions folder), prepare_stan_data and write_save_model_results. This would take a more advanced knowledge of R and Stan compared to making data updates.

Shiny App

All code and inputs required to make the Shiny App are contained in the app folder. To run the app, open the make_app script and click the "Run App" button that is display at the top of the screen in RStudio. This runs the app locally. To publish the app (e.g. on the shinyapps.io website), click the "Publish" button on top of the app. If there have been data or model updated you will need to rerun and republish the Shiny app.

y-chu / child_welfare_projections