# Compute

Perhaps our best loved lab members...

### Shackleton

Dell T620 Server. 24 CPU cores @ 2.4 GHz, 36 TB disk, 96 GB memory.

### Sacagawea

Dell T630 Server. 48 CPU cores @ 3.0 GHz, 128 TB disk space, 256 GB of memory.

### Laika

4 x Dell Blade servers at Berkeley HPC. 96 CPU cores @ 2.3 GHz, 512 GB memory.

### Norgay

Synology DS1815+ NAS w/ 2x DX513 expansions. 88 TB disk space.

# Code

**Adjust least-squares standard errors for spatial correlation and serial correlation in panel data. **(Stata code here, Matlab code here)** **(The technique is GMM, as described by Conley 2008). Code is described here.

**Move data back and forth between Stata and Matlab. **Code here. Usage is described here.

**Partial-out two variables from a model and plot them with a single command.** (Stata) Code here.

**Non-parametric regression for one or two independent variables** (Matlab) Code here. described here.

**Generate and analyze random networks** (Matlab) Poisson, Watts-Strogatz, generate overlapping stars, plotting, computing empirical connection density, clustering coeff.** **Code here.

**Data visualization tools**

**Watercolor Regression**** **(MATLAB)** **Display a smooth nonparametric distribution of the probability density for a nonparametric regression estimate. Code is here. Watercolor regression (described here) is a type of visually-weighted regression (described here). Citation: *Hsiang (2012) "Visually-Weighted Regression" *(here).

**Visually-Weighted Regression**** **(STATA & MATLAB) Use the color-saturation of a regression line to denote the statistical confidence in the regression line at each point. Code is here. Described here with several options demonstrated here and the "watercolor regression" option described here. Citation: *Hsiang (2012) "Visually-Weighted Regression" *(here).

**Three-Dimensional Nonparametric Regression** (MATLAB) Flexibly estimate means conditional on two variables with bootstrapped confidence intervals. Code is here.

**Boxplot Regression** (MATLAB) Discretize an independent variable and use boxplots as a nonparametric regression. Code is here.