Update README.md

Author: luna983

README.md

@@ -1,4 +1,4 @@
-# Measuring the Impacts of Poverty Alleviation Programs with Satellite Imagery and Deep Learning
+# Using Satellite Imagery and Deep Learning to Evaluate Anti-Poverty Programs
 
 _Luna Yue Huang_
 
@@ -8,23 +8,6 @@ _Luna Yue Huang_
 
 ----
 
-__Abstract__
-
-Household surveys are expensive. In this project, I argue that housing, a strong correlate of wealth, can be accurately and cheaply measured with high-resolution satellite imagery and deep learning models, and can be used to conduct impact evaluations with much lower costs. In Western Kenya, I evaluate the GiveDirectly randomized controlled trial, a large unconditional cash transfer program, with satellite imagery, and observe statistically significant and economically sizeable increases in overall building footprint and roof reflectance, a proxy for housing quality. Using an Engel curve approach, I infer overall treatment effects from observed increases in the consumption of housing, and obtain consistent results with extensive in-person surveys.
-
-----
-
-__Figure 1__: The figure below shows 10 sets of randomly sampled prediction results generated by the Mask RCNN model. The model performs well with a recall of 0.79 and a precision of 0.80.
-
-![fig-chips](/docs/fig-chips.jpg)
-
-__Figure 2__: The figure below reports estimated average treatment effects of the GiveDirectly unconditional cash transfer on the total square footage of buildings, roof reflectance, and nightlight values.
-_The x-axis represents the intensity of the cash inflow (in nominal USD) to each pixel (covering an area of 0.012 square kilometers). The y-axis represents the magnitude of the treatment effects. Red points represent the point estimates, error bars correspond to the 95% confidence intervals. Gray lines indicate estimated treatment effects from 100 placebo simulations.  Panel title reports pooled estimates with a linear effect assumption, along with 95% confidence intervals._
-
-![fig-ate](/docs/fig-ate.jpg)
-
-----
-
 ## Instructions for Replication
 
 To pretrain the deep learning model (required for replication of all the subsequent analyses)