Policy Research Project: Fair Play, Over Fair Pay?

A data science capstone research project for the Data & Decisions course taught by Abhilash Mishra, at Applied Data Institute by Equitech Futures, exploring whether perceptions of social fairness are associated with how many children young Chinese adults say they want.

A slide from my capstone presentation in December 2025.

--

Using the 2023 Chinese General Social Survey (CGSS, N = 4,889 respondents born 1980+), the analysis finds that procedural fairness — whether people feel the system is fair and that effort is rewarded — is positively associated with ideal number of children, while distributive fairness (perceived income equality) is not. The project also documents a sharp generational gap: Gen Z respondents report an average of 1.22 ideal children compared to 1.79 for Millennials.

These are correlational findings from cross-sectional survey data, not causal claims. The paper discusses limitations in detail.

Interactive data story

An accompanying scrollytelling website presents the key findings in a more accessible format:

Fair Play, Over Fair Pay? — a data-driven narrative exploring why young China is saying no to babies, built with D3.js and Scrollama.

Summary of findings

Finding	Detail
Procedural fairness is associated with fertility intentions	General fairness β = 0.100 (p < 0.001); meritocracy β = 0.118 (p = 0.024)
Distributive fairness is not	Income fairness β = −0.031 (n.s.)
Generational gap	Gen Z 1.22 vs Millennials 1.79 ideal children (d = 0.56)
Urban–rural differences	Urban residents respond to general fairness (β = 0.117); rural to meritocracy (β = 0.201)
Gender differences	General fairness effect is stronger for men (0.119) than women (0.076)

Methods

• Sample: CGSS 2023 respondents born 1980 or later (Millennials + Gen Z).
• Missing data: the split-questionnaire design leaves 53–73% missing on fairness items; addressed with multiple imputation (M = 20) pooled via Rubin's rules.
• Models: OLS with HC3 robust standard errors; Poisson and Negative Binomial as robustness checks.
• Heterogeneity: stratified analyses by gender and urban/rural residence, with Benjamini–Hochberg FDR correction on interaction tests.

Reproduce

pip install -r requirements.txt
# Place the CGSS 2023 microdata at data/CGSS2023.csv — see data/README.md
python run_analysis.py          # full pipeline (M = 20)
python run_analysis.py --quick  # smoke test (M = 3)

The full pipeline reproduces the appendix tables in outputs/ to three decimal places.

Repository structure

src/                  Modular analysis pipeline
  config.py             paths, constants, variable lists
  data_prep.py          load CGSS + construct variables
  imputation.py         multiple imputation (M = 20, Rubin's rules)
  descriptives.py       descriptive + bivariate statistics
  models.py             pooled OLS / Poisson / NegBin, stratified, interactions
  figures.py            editorial figures (driven by model results)
  export_web.py         pre-aggregated JSON for the website
run_analysis.py       one-command orchestrator
notebooks/            Narrative notebook
outputs/              Figures + appendix tables
site/                 Interactive data-story website
paper/                Capstone paper (markdown)
data/                 Codebooks + questionnaires (microdata not included)

Data access

The CGSS microdata requires free individual registration and is not redistributed here. Codebooks and questionnaires are included. See data/README.md for how to obtain the data.

Citation

If you reference this work, please cite the paper in paper/ and cite CGSS per the survey provider's requirements.