This page illustrates the use of bbandits.ado for Stata. See Stata. To download the bbandits package including the ado file type in Stata
-
. net from https://rostam-afschar.de/bbandits
. net install bbandits.pkg
See Kemper and Rostam-Afschar (2024) "Earning While Learning: How to Run Batched Bandit Experiments" for more information. Check out the slides.
Example 1a: Estimate average treatment effects using batched OLS
-
Six call methods to enroll rice farmers from (Kasy and Sautmann, 2021):
. use "example data\kasy_sautmann_2021.dta", clear
. bbandits outcome treatment date
Example 1b: Show cumulative weights assigned by Exploration sampling
Example 2: Estimate average treatment effects using batched OLS and pooled OLS
-
32 invitation messages for business surveys (Gaul, Keusch, Rostam-Afschar, Simon, 2024):
. use "example data\gaul_et_al_2024.dta", clear
. bbandits reward selected trial
Example 3: Monte Carlo simulation of OLS and BOLS estimates with small margin using epsilon-greedy
-
. bbandit_sim 1 1, monte_carlo greedy reference_arm(0) arm(1) test_value(0) n(500) eps(0.2)
Example 4: Finding best of three arms (stacked and not stacked) using Thompson sampling
-
download Stata code here
Example 5: Monte Carlo simulation of OLS estimates with small and large margin at batch size 20 using Thompson sampling
-
download Stata code here
Example 6: Monte Carlo simulation of OLS estimates with zero margin using Thompson sampling
-
download Stata code here
Authors
Jan Kemper
ZEW Mannheim
jkemper@zew.de
Davud Rostam-Afschar
Universität Mannheim
rostam-afschar@uni-mannheim.de