/binscatter

binscatter: now with confidence intervals!

Primary LanguageStataOtherNOASSERTION

binscatter: Binned Scatterplots...

binscatter is a Stata program which generates binned scatterplots. These are a convenient way of observing the relationship between two variables, or visualizing OLS regressions. They are especially useful when working with large datasets.

See the project webpage for more information.

... now with confidence intervals

This version adds support for confidence intervals!

Add confidence intervals around the conditional means with the option ci(bins), add confidence intervals around fit lines (both qfit and lfit supported) with ci(model). Technically, both options can be selected at the same time (though this doesn't usually look good) with ci(bins model).

Install

Install from within stata

net install binscatter, from("https://raw.githubusercontent.com/dballaelliott/binscatter/pkg/") replace

OR 2. Download/clone/fork this repo

Dependencies

This program uses reghdfe to allow for extended vce options for standard errors.

ssc install reghdfe

ci(model)

The ci() option class provides a suboption to bound linear and quadratic fits.

The syntax is exactly the same in either case; binscatter knows to bound the appropriate model type, whether it is linear...

binscatter price mpg, ci(model) linetype(lfit)

Example Figure: Linear Fit

or quadratic....

binscatter price mpg, ci(model) linetype(qfit)

Example Figure: Quadratic Fit

ci(bins)

The ci() option class also provides an option to bound the binned y-means.

sysuse auto, clear 
binscatter price mpg, ci(bins) linetype(none)

Example Figures: Many Bins

This helps to highlight the bias-variance tradeoff that binscatter shares with other non-parametric conditional means estimators,

sysuse auto, clear 
binscatter price mpg, ci(bins) linetype(none) n(5)

Example Figures: (Maybe) Too Few Bins

Aesthetically, this can look quite nice when we add linetype(connect)

binscatter price mpg, ci(bins) linetype(connect) n(5)

Example Figures: Connecting the Dots

Technically, these options can be specified together, but it often leads to a crowded figure...

binscatter price mpg, ci(model bins) linetype(lfit)

Example Figure: Too Much!

Changelog/ Coming Soon...

  • Alternative implementation of ci(bins) that doesn't rely on tranparency for backwards compatability with versions before Stata 15.0.

  • Quality of life update to ci(model) The option n_draws() is now available. This takes an integer argument, and can be used to set the number of draws from the bivariate normal distribution that are used to approximate the confidence bands on the model (more precisely, the bounds on y-hat). The default is 100,000.

  • Installing via net install also updates the stata help file so that the internal documentation provided by help binscatter includes documentation for the ci() option.

in line compiled externally by codecogs