Social Housing and Investment Improvement

A repeat of the analysis on the effect of social housing on overall well-being of individuals with improved link rates and an additional wave of the GSS survey.

Overview

This analysis repeats and improves on previous work by the Social Investment Agency (previously the Social Wellbeing Agency) to measure how placement in social housing impacts the wellbeing of people. In November 2018 we released a report Measuring the wellbeing impacts of public policy: social housing, and the accompanying GitHub repository social-housing-and-wellbeing. The analysis behind this previous report has been repeated with several improvements, including:

• Wellbeing indicators from the 2016 wave of the New Zealand General Social Survey (NZGSS) are used following its addition into the Integrated Data Infrastructure (IDI).
• Refinements to the spine linking process, using address information to resolve orphan identities, provide a higher link rate for the NZGSS respondents.

Both improvements increase the sample size, reducing uncertainty, and making the results and conclusions more robust. The code from this repository relates to the report Measuring the impact of social housing placement on wellbeing: Evidence from linked survey and administrative data.

Features

This repository contains several code features that may be of interest to researchers, even if they do not wish to run the entire analysis:

• Stats NZ provides survey weights for the GSS calculated for the entire respondent population. When working in the IDI a reweighting to account for imperfect linkage is recommended. 2_of_rewt_gss_person_replicates.R contains our methodology for updating the survey weights to account for non-linkage.
• Some identities can not be fully linked, resulting in orphan identities (partial identities that have not been linked to the spine - the list of core identities). si_improve_gss_linkage.sas contains our methodology for identifying orphan identities and reconnecting them to the spine.
• The IDI contains 5 waves of the NZGSS. Due to changes in the survey over time, combining these waves into a single dataset requires some cleaning. The code scripts si_create_of_gss_variables.sas contains the core of our methodology to combine these five waves into a single dataset.

Dependencies

• It is necessary to have an IDI project if you wish to replicate this work. Visit the Stats NZ website for more information about this.

• While we have attempted to capture all the code dependencies in this project, several other Agency repositories may be required to replicate this project. These repositories are:

  • social_investment_analytical_layer (SIAL)
  • social_investment_data_foundation (SIDF)
  • SIAtoolbox

• Instructions for the installation of each repository can be found in their respective readme files. Only the SIAL would need to be run prior to this analysis. Any dependency on the SIDF and tool box are via SAS macros and R functions contained in those repositories.

This analysis has been developed for the IDI_Clean_20181020 refresh of the IDI. As changes in database structure can occur between refreshes, access to, and use of, this refresh should be considered a dependency for the purpose of executing the code as it was created.

The R code makes use of several publicly available R packages. The version of some of these packages may be important. This analysis was conducted using dplyr version 0.7.2, srvyr version 0.2.2 and rlang version 0.2.1.

Folder descriptions

This repositry contains all the core code to assemble the data and run the analysis.

• sasautos: This folder contains SAS macros that are used during the processing.
• sasprogs: This folder contains SAS programs. The main script that builds the dataset is located in here as well as the control file where analysis parameters are entered.
• rprogs: This folder contains all the necessary R scripts that are required to perform the analysis on the dataset created by the SAS code.
• sql: Several auxiliary SQL scripts that create summary outputs for the final report are stored here.

Instructions to run the project

Step A: Create analysis population and variables

1. Start a new SAS session
2. Open sasprogs/si_control.sas. Go to the yellow datalines and update any of the parameters that need changing. The one that is most likely to change if you are outside the Agency is the si_proj_schema. In case the IDI version that you're pointing to needs to be updated, edit the corresponding variables as well- the variables are idi_version and si_idi_dsnname. Note that the results in the paper are based on IDI_Clean_20181020. If you have made changes save and close the file.
3. Open sasprogs/si_main.sas and change the si_source_path variable to your project folder directory. Once this is done, run the si_main.sas script, which will build the datasets that are needed to do the analysis.

Step B: Data Preparation & Analysis

There are 2 distinct streams of analysis for this project-

Survey-Weighted Descriptive Statistics

1. Start a new R session.

2. Open up rprogs/1_of_weighted_gss_analysis_wrapper.R. Edit the working directory by modifying with path specified at the first line of this file. In addition, also edit the variable schemaname to the appropriate schema that you are using. This is a wrapper script that runs all steps involved for generating the weighted descriptive statistics for the analysis. The script performs a linking of the GSS survey data with the IDI Spine, and reweights the survey to account for records that are unlinked with the IDI Spine. A comparison test between the distribution of the GSS variables is also performed pre and post-reweighting. This is to ensure that the IDI Spine linkage and the subsequent and re-weighting procedure does not bias the variables that is to be compared. The outputs of this operation can be obtained from the output folder.

Unweighted Before-After Analysis

1. Start a new R session.
2. Open up rprogs/1_run_analysis_treat_control.R. This script creates loads up all required libraries and generates all the Before-After analysis results. This analysis does not take into account the survey weights, and compares the group that was housed 12 months before GSS interview to the group housed 15 months after. Bootstrap sampling is used to get confidence intervals around the estimates here. In addition to the main analysis, this code also performs a validation, by comparing the group that was housed 12 months before GSS interview to the group housed 12 months after, and another validation using propensity matched groups. Additionally, this code also performs regression models for the outcome variables of interest. The outputs of this analysis can be obtained from the output folder.

Citation

Social Investment Agency (2019). Social housing and wellbeing improvement. Source code. https://github.com/nz-social-investment-agency/social-housing-and-wellbeing-improvement

Getting Help

If you have any questions email info@sia.govt.nz