Skills and methods.

Machine learning and text analysis

I use neural-network ensembles to classify occupational descriptions and word embeddings to compare how people describe the same job. My coursework also covered support vector machine classification, random forests, bagging, and boosting.

Methods

Neural-network ensembles in TensorFlow
Word2Vec and GloVe embeddings
k-means and DBSCAN clustering

Metric construction

I enjoy constructing metrics to operationalize abstract concepts.

Measures developed

Sharp Boundary Index: combines the strength and relative direction of racial gradients with the local presence of both groups.
Content-weighted semantic similarity: compares two descriptions using both agreement in meaning and the amount of relevant detail.

Large datasets and computation

I regularly wrangle messy panel data, including roughly 250 million raw mobile-phone observations, and harmonize millions of municipal 311 service-request records.

Methods

Longitudinal data cleaning and linkage
Harmonizing administrative records
Matrix formulation of segregation estimates
Parallel computation in R and C++

Statistical and econometric modeling

I fit models across panel, survey, spatial, and frequency data. My training also includes time-series forecasting with ARIMA models.

Methods

Fractional logit and fixed-effects models
Poisson pseudo-maximum likelihood (PPML)
Multilevel and survey-weighted regression
Difference-in-differences
Propensity-score matching
Mediation analysis
Event-history analysis
Log-multiplicative models for frequency data

Spatial and network analysis

I address spatial autocorrelation with spatial regression and spatially robust standard errors. I work with mobile-phone coordinates and Census shapefiles. I used Daymet weather raster data to analyze the relationship between climate change and migration. I also implemented network sampling with memory in R for an immigrant survey, translating the original Stata code.

Methods

Spatial lag and spatial error models
Spatial HAC standard errors
Demographic surfaces and boundary measures
Community detection algorithms
Network sampling with memory (NSM)
Raster processing and geographic aggregation
Shapefiles and coordinate reference systems

Visual explanations.

I create visualizations that make abstract concepts and analysis results intuitive.

Tools

R: Data preparation, statistical modeling, spatial analysis, and graphics.
C++: Matrix and parallel computation called from R through Rcpp.
Python: TensorFlow models and text processing.
Additional tools: SQL; project experience with Julia and Stata.