/*		REPLICATION OF STATISTICAL ANALYSIS IN "MAPPING BLUE HELMETS: INTRODUCING THE GEO-PKO DATASET"
		Deniz Cil, Hanne Fjelde, Lisa Hultman, and Desirée Nilsson
		Journal of Peace Research
*/


clear
use "replication_Geo-PKO_analysis_JPR.dta"


sort id year month

*	Generate new variables
gen roaddens=.
replace roaddens=1 if pavedroad_density_90th==0
replace roaddens=0 if pavedroad_density_90th==1

gen troops1000=troops_total/1000

gen troops_roaddens=troops1000*roaddens
gen area_log=ln(area)
gen mountains_log=ln(mountains_mean)
gen pop_log=ln(pop_gpw_sum)
gen capdist_log=ln(capdist)

by id: gen troops_roaddens_lag=troops_roaddens[_n-1]
by id: gen troops1000_lag=troops1000[_n-1]

by id: egen conf_prov_max=max(battle_sum)
gen conf_prov=conf_prov_max
replace conf_prov=1 if conf_prov_max>0

*	Summary statistics
inspect battle_sum
sum troops_total


*	MODELS REPORTED IN TABLE II:

*	Model 1
nbreg battle_sum troops1000_lag roaddens area_log nlights_calib_mean mountains_log pop_log duration_conf conf_intensity_3mave, cl(id) nolog

*	Model 2
nbreg battle_sum troops1000_lag roaddens troops_roaddens_lag area_log nlights_calib_mean mountains_log pop_log duration_conf conf_intensity_3mave, cl(id) nolog

*	Model 3
nbreg battle_sum troops1000_lag roaddens troops_roaddens_lag area_log nlights_calib_mean mountains_log pop_log duration_conf conf_intensity_3mave if conf_prov==1, cl(id) nolog


*	Downsizing the sample: Results keeping 50% zeros on dependent variable
sample 50 if battle_sum==0
nbreg battle_sum troops1000_lag roaddens troops_roaddens_lag area_log nlights_calib_mean mountains_log pop_log duration_conf conf_intensity_3mave, cl(id) nolog