One policy, two questions.
How is the Swachh Bharat Mission framed, and does the data match what was declared?
How the policy is framed.
How do government, academics and journalists frame the same policy differently?
A Computational Text Analysis of Stakeholder Framing in the Swachh Bharat Mission
I studied how different sources describe the same sanitation policy, using computational text analysis to compare stakeholder framing.
Mentors: Amrita Saib (Director, EY) and Gaurav Thapak (Consultant, EY).
From documents to patterns.
The corpus contains 55 documents across 7 source categories, including academic papers, government reports, journalism and policy briefings. I classified sources into 15 stakeholder categories.
- 01Python preprocessing
- 02VADER sentiment
- 03TF-IDF
- 04LDA
The pipeline identifies 5 thematic clusters and compares sentiment across government, academic and journalistic sources.
Different sources, different frames.
The analysis revealed contrasting perspectives on sanitation policy across source types. Automated sentiment analysis, however, struggles to capture complex social issues. A score is not a substitute for reading the context.
Under review at the Journal of Emerging Investigators (JEI) · CREST Gold Award, October 2026.
Framing tells you what people say.
I wanted to know what the data says.
Declared vs. Measured.
When a district is declared “open defecation free”, does the data agree?
I compared official open-defecation-free (ODF) declarations with NFHS-5 sanitation data across 704 Indian districts.
of districts below
70% sanitation coverage.
42% of districts below 70% sanitation coverage. This arrangement illustrates the reported proportion; it does not identify individual districts.
Beyond the declaration.
I used first-difference regression to examine child-health outcomes. The follow-up asks how official declarations relate to measured sanitation coverage, rather than assuming they describe the same thing.
I am now rewriting the analysis code myself and developing it into a standalone paper.
Code & data · Declared vs. Measured