MSc Thesis: Italian PVS Analysis – Reproducible Workflow

This repository contains the full Stata + Python workflow used for my MSc Thesis titles:

“Socioeconomic Inequalities in Perceived Healthcare Quality in Italy”

My thesis investigates disparities in individuals’ perceptions of healthcare quality in Italy across socio-economic classes. The analysis reveals a strong correlation between lower socioeconomic status and negative perceptions of the quality of healthcare, and explores potential mechanisms underlying this relationship.

The present code starts from the raw PVS micro‑data, cleans and enriches the dataset, imputes some missing values, derives explanatory indices, builds socio‑economic status clusters, runs regression models and mediation tests.

Methodology

Data

The analysis draws from the first wave of the People's Voice Survey (Italian sample). The original dataset can be downloaded for free here

Additional information about the dataset can be read here

Methods for analysis

• Cluster analysis to identify socio-economic clusters
• Principal Component Analysis to construct indices
• Multivariate linear regression and logit
• Sobel mediation tests

---

Project tree

.
├── data/                # raw‑data .dta files + intermediate outputs
├── codes/               # Stata do‑files
│   ├── 1 - clean_data.do
│   ├── 2 - impute_income.do
│   ├── 3 - clusters.do
│   ├── 4 - indices.do
│   └── 5 - regressions.do
├── master.do            # one‑click driver script
└── outputs/             # tables, figures, logs
    ├── figures/
    └── tables/             

Tip: All relative paths are built from the folder that contains master.do. Clone the repo, cd into it, open Stata, and just do master.do.

---

Prerequisites

Software	Version	Used for
Stata	17 (or newer 16)	Main data wrangling & econometrics
Python	3.9+ (via python‑end)	K‑NN income imputation
pandas	≥ 2.0	—
scikit‑learn	≥ 1.4	KNNImputer
matplotlib / numpy	any recent	plotting & numerics
Stata user packages	estout, pca, mca, missings, heatplot, palettes, tabout, sgmediation2

All Stata packages are installed automatically by master.do. For Python, create a virtual‑env and pip install -r requirements.txt with:

pandas scikit-learn matplotlib numpy

---

How the pipeline works

Stage	Script	What it does
0. Prepare environment	master.do (lines 1–70)	Installs any missing SSC packages, defines global folder macros ($data, $codes, $outputs, …).
1. Clean raw data	1 - clean_data.do	‑ Keeps the Italian sample (country==14).‑ Harmonises variable names, drops uneeded IDs.‑ Recodes geography, education & insurance variables.‑ Calls indices.do and regressions.do later in the chain.
2. Impute income	2 - impute_income.do	Runs inside a Python block:   1. Pulls current Stata dataset into pandas via SFI.  2. Uses KNNImputer twice – first on the most‑correlated variables, then on a user‑selected set – to fill missing income.  3. Writes the imputed columns to data/imputed_income.dta.  Back in Stata, master.do merges those columns on respid.
3. SES clusters	3 - clusters.do	Standardises education & (imputed) income, then applies k‑means to create three socio‑economic status clusters (ses_cluster). Includes exploratory graphs & basic validations.
4. Outcome indices	4 - indices.do	Builds three composite outcomes:   System Quality, Primary‑Care Coverage, Patient Satisfaction. Each block: z‑scores variables, checks α / inter‑item correlations, runs a PCA where justified, and labels the resulting index.
5. Main regressions	5 - regressions.do	Runs a battery of OLS, Logit & Ordered‑Logit models:   Model 0 – controls only  Model 1 – + education  Model 2 – + SES cluster dummies. Stores estimates with eststo, exports publication‑ready tables via esttab.
6. Wrap‑up	master.do (final lines)	Collects regression tables, saves graphs (under outputs/figures), and writes log files.

---

Running the entire study

1. Clone the repo

   git clone https://github.com/<user>/italy‑pvs.git
   cd italy‑pvs

2. Open Stata, set your working dir to the repo root or simply run

   cd "~path to current folder"
   do master.do

3. Grab a coffee — the whole run takes ≈ 3 min on a modern laptop (Python imputation ~30 s).

All intermediate .dta files and final outputs appear in /outputs. Logs (.smcl) for every script are also saved there for transparency.

---

Citation

If you use or adapt this code, please cite:

Ardito, E. (2025). “Socioeconomic Inequalities in Perceived Healthcare Quality in Italy”. Unpublished MSc Thesis.

Any addition, correction or suggestion is welcome!