Slack: #p-india-nfhs
Project Description: This D4D project aims to glean insights from India's National Family Health Survey datasets. Generally around women's empowerment and wealth index issues.
Project Co-Leads / Maintainers: @lukewolcott
Contributors: (Slack handles) @lukewolcott, @rsibi, @sai, @preet, @vsridhar, @fibinse
Data: https://data.world/data4democracy/india-nfhs, and in this repo.
Background on the data | Motivating questions | Repo contents | Deliverables
The NFHS is conducted about every 10 years. The 4th one, NFHS-4, was conducted 2015-2016 and the full datasets will be available "later this year". At the moment, state-level and district-level data are available in PDF factsheets. Debarghya Das has written scripts to scrape the district-level PDFs and pull the data into one place. This CSV file is available in our repo as nfhs-district-level.csv.
The NFHS-3, conducted 2005-2006, is available in full from the Demographic and Health Surveys (DHS) program. So are the two earler NFHSs. The datasets come from three questionnaires: female, male, and household. One must register to use this data, so we can't put it up on a public repo. Instead, it lives in a private dataset on the D4D data.world site. Contact @lukewolcott through Slack to get access to this dataset.
The NFHS includes a "Wealth Index" variable, but this oversimplifies the significant difference between rural and urban Indian households. Can we use clustering and PCA to construct multiple, more indicative wealth indices?
The Hindustan Times created a "Women's Empowerment Index" from questions 101--108 in the state-level data, and has done some analysis with it. Is this an accurate index, or could it be improved? What trends do we see? How does this index correlate with other variables (like alcohol consumption)?
What do the district-level NFHS-4 data indicate about geographic patterns for important variables related to wealth, empowerment, and quality of life? Can we use this aggregated data to anticipate the individual household-level data?
The NFHS-3 (from 2005-2006) household questionnaire gathers hundreds of variables on 100k households, including some health data. Can we use these household characteristics to predict if the household head is male or female? Or if the household is likely to have clean water? Can we predict if the household is at risk of alcoholism? How do household characteristics cluster the data?
The nfhs3-metadata folder has some metadata for the NFHS-3 household questionnaire. To access the data, contact @lukewolcott and he'll send you an invitation to the data.world site.
The nfhs3-women-emp-analysis folder has some initial exploratory plots, a jupyter notebook with some exploratory data analysis, and a notebook of machine learning models (decision trees, random forests) to predict the sex of the household head.
The nfhs3-wealth-index-analysis folder contains files and reports analyzing the 'wealth index' variable, and the difference between rural and urban households.
Independently Debarghya Das scraped the district-level NFHS-4 data into nfhs-district-level.csv. He has generated some heatmaps with some of the gender-related variables, and this is a good start for generating new questions to ask the data. Questions 101--108 in the data are related to women's empowerment.
The volatile-sex-ratio folder contains the files and code used to generate @fibinse's Volatile Sex Ratio data story.
The nfhs4-chloropleths folder contains files and reports related to district-level chloropleths of important variables.
EDA-lukewolcott.ipynb: a notebook with initial exploratory data analysis and plots for NFHS-3. (@lukewolcott)
Predicting-HV219.ipynb: using decision trees, random forests to predict sex of household head. (@lukewolcott, @vsridhar)
Wealth_Index_Initial_Report.html: analyzing the components of the wealth index, as they differ between geographic regions (rural, urban, large or small city, etc). (@sai, @preet)
Volatile Sex Ratio: a nice web report looking at change in each Indian state's sex ratio between NFHS-3 and NFHS-4. (@fibinse)
NFHS-4 Chloropleth maps: district-level chloropleth maps of several important variables from the NFHS-4. (@preet)
Clustering_by_residence_type.html: cluster analysis of separate geographic regions, looking at access to electricity, clean water, improved sanitation, and clean cooking fuel. (@lukewolcott, @sai, @preet)