Binscatter2 - faster, enhanced binned scatterplots in Stata

[Michael Droste]

Hi all,

I wanted to make note of a program that I've had available on GitHub for a while now to generate binned scatterplots in Stata, like Michael Stepner's excellent -binscatter- package. Binscatter2 inherits all of the usage and syntax of binscatter, but runs substantially faster in large datasets by leveraging the functionality of -gtools-. Furthermore, binscatter2 offers a handful of new bells and whistles -- expanded options for saving, fit lines, plotting quantile intervals of the data, etc.

This project is still very much ongoing, and I hope to submit it to SSC very soon.

You can read more it here: https://github.com/mdroste/stata-binscatter2

[Michael Droste]

I also want to note that another binscatter alternative has popped up in the last few days, binsreg: https://sites.google.com/site/nppackages/binsreg. Binsreg offers some features that binscatter2 also adds - like more flexible fit lines. Binsreg has more robust options for inference (i.e. overlaying confidence intervals), but less support for looking at the conditional variance of y given x (i.e. overlaying quantiles). One key difference that distinguishes binscatter2 with big datasets is runtime:

. clear all 

. set obs 25000000
number of observations (_N) was 0, now 25,000,000

. gen x = runiform()

. gen y = 1 + 2*x + 3*x^2 + rnormal()

. set rmsg on
r; t=0.00 16:47:33

. binscatter y x
r; t=93.44 16:49:07

. binscatter2 y x
r; t=20.70 16:49:28

. binsreg y x, nbins(20)

Binscatter plot
Bin selection method: User-specified
Placement: Quantile-spaced
Derivative: 0

----------------------------------------------
# of observations             | 25000000
# of distinct values          | 16455446
# of clusters                 |       .
------------------------------+---------------
Bin selection:                | 
         Degree of polynomial |       .
  # of smoothness constraints |       .
                    # of bins |      20
----------------------------------------------

----------------------------------------
         |      p       s       df
---------+------------------------------
 dots    |      0       0       20
----------------------------------------
r; t=281.96 16:54:10

Binsreg also offers an alternative method of residualizing a set of controls that the authors argue is more principled than what binscatter and binscatter2 currently incorporate. I will look into adding this alternative residualization in the near future.