ViolenceAgainstWomen
Project by Nicholas Neuteufel while working for Tremendous Hearts (Cape Town, South Africa)
Written in R language.
Uses UN Women data (http://www.endvawnow.org/uploads/browser/files/vawprevalence_matrix_june2013.pdf) and the following R packages: randomForest, quantregForest, bartMachine, WDI, rworldmap, ggmap, lattice, Hmisc, sp, spgwr, ape, and countrycode.
NEXT STEPS:
a) Political, conflict, and cultural indicators
b) Model comparison (BART v Random Forests)
VERSION 1.3
Bayesian Additive Regression Trees (BART) analysis performed.
VERSION 1.2
Extreme weather (drought/flooding) and corruption control data added.
VERSION 1.1
Geography-weighted regression (GWR) -- COMPLETE (http://imgur.com/a/dp53F) -Including spatial autocorellation (Moran's I) analysis
Region analysis (Statistically analyzing trends by continent and regions within continents)
VERSION 1.0
STEP 1: Mapping Out Violence (see Mapping) -- COMPLETE
STEP 2: The "Missing" Statistics--trying to predict missing data
A) Experimentally -- COMPLETE
a) Random Forests (RF) -- complete (see ExperimentingDataImputation)
B) Empirically -- COMPLETE-ish
a) Using older data from WDI to fill-in incomplete cases (NAs) -- COMPLETE
b) Comparing empirical data to RF predicted imputation (pending C(a))
STEP 3: Predicting key countries with missing data (with both 95% & 90% prediction intervals)
No particular order:
a) Asia: China, South Korea, Saudi Arabia, Iraq, Iran, Pakistan, Indonesia.
b) Latin America: Argentina.
c) Africa: South Africa.
d) Europe: Spain, France.
STEP 4: Mapping the world; project evaluation.