je-codes291/kap-survey-lead-poisoning-nigeria

Baseline KAP survey analysis (2,586 respondents, 4 Nigerian states) — data QA, bivariate stats, multivariable logistic regression, and thematic coding of open-ended responses in Python. # Baseline KAP Survey Analysis — Lead Poisoning Awareness (Nigeria) > **Note on this repo:** This is an anonymized portfolio version of a commissioned client project. The commissioning organization's name, the internal project name, and all verbatim respondent quotes have been withheld or removed for client confidentiality. The statistical methodology, code, and aggregate findings below reflect the analysis as actually conducted. A full quantitative analysis of a baseline **Knowledge, Attitudes, Practices and Behaviours (KAPB)** household survey on lead poisoning awareness and prevention, covering **2,586 respondents across 4 Nigerian states** (Kano, Kwara, Nasarawa, Lagos). I designed and ran the entire analytical pipeline: data reconciliation and quality assurance, descriptive and bivariate statistics, multivariable logistic regression, thematic coding of open-ended responses, and a full state/LGA-level breakdown, delivered as a client-facing report with metric-bound, timebound recommendations. ## What this project demonstrates - **End-to-end survey data pipeline**: raw KoboToolbox export → cleaned analytic dataset → statistical analysis → client-ready findings and recommendations - **Real inferential statistics**, not just descriptive summaries: chi-square tests of independence, one-way ANOVA, and two multivariable logistic regression models (`statsmodels.formula.api`) reporting adjusted odds ratios with 95% confidence intervals - **Rigorous data QA**: reconciled a 50-column analytic dataset against the raw 75-column export, matching all retained variables 1:1 by respondent UUID before running any analysis - **Transparent handling of a real data gap**: six planned survey items were missing from the dataset entirely; rather than silently working around this, the analysis documents the gap, substitutes two directionally-informative regression models, and flags exactly what would need to be recovered before the original planned model could be run - **Thematic an …