/* This .do file computes the pre-treatment fit index proposed by Adhikari and Alm (2016) using the Stata sample dataset provided by Abadie et al. (2010). Author: Bibek Adhikari. */ clear all ** download data and declare panel sysuse smoking, clear xtset state year ** define synth parameters local outcome "cigsale" local treatment_year = 1989 local treatment_unit = 3 local first_year = 1970 ** run synth synth `outcome' `outcome' cigsale(1988) cigsale(1980) cigsale(1975) beer lnincome retprice age15to24, trunit(`treatment_unit') trperiod(`treatment_year') fig ** store rmspe in scalar scalar rmspe = e(RMSPE)[1,1] ** create benchmark rmspe gen outcome_square = `outcome'*`outcome' sum outcome_square if year<`treatment_year' & year >= `first_year' & state==`treatment_unit' scalar benchmark_rmspe = sqrt(r(mean)) ** calculate fit index scalar fit_index = round(rmspe/benchmark_rmspe,0.001) ** Display them all di "actual rmspe is " rmspe di "benchmark_rmspe is " benchmark_rmspe di "pre-treatment fit index is " fit_index