Calculating Wildbootstrap P-Value with jwdid

[Fredie Didier]

Hello, I am trying to calculate the Wildbootstrap P-Value after I perform an analysis with jwdid.

I am running: jwdid taxa_homicidios_total_por_100m_1 [aw= population_2000_muni], ivar(municipality_code) tvar(year) gvar(treatment_year) never cluster(state_code)

estat group

and then when I ran: boottest _b[group], weighttype(webb) cluster(state_code) reps(9999) seed(123) this appears: "Overriding estimator's cluster/robust settings with cluster(state_code)
note: constraint _b[group] caused error r(111)"

If I put the parameter boottype(wild) there is also an error

Can someone help me please?

[FernandoRios]

I believe boottest works based on the influence functions.
estat group after jwdid however does not estimate any influence functions. So I don’t think that will work.

[Fredie Didier]

Thanks for replying! Do you think that if I do the following below it will be correct?:

drop if municipality_code == 2300000 | municipality_code == 2600000
drop if population_2000_muni == .
gen log_population = log(population_muni)

* Create treatment variables
gen treated = 0
replace treated = 1 if state == "PE" & year >= 2007
replace treated = 1 if state == "BA" & year >= 2011
replace treated = 1 if state == "PB" & year >= 2011
replace treated = 1 if state == "CE" & year >= 2015
replace treated = 1 if state == "MA" & year >= 2016

* Create treatment year (staggered treatment)
gen treatment_year = 0
replace treatment_year = 2011 if state == "BA" | state == "PB"
replace treatment_year = 2015 if state == "CE"
replace treatment_year = 2016 if state == "MA"
replace treatment_year = 2007 if state == "PE"


// 1. Original JWDID estimation
jwdid taxa_homicidios_total_por_100m_1 [aw= population_2000_muni], ///
ivar(municipality_code) tvar(year) gvar(treatment_year) never cluster(state_code)

// 2. Get group effects and store them
estat group, esave(group_effects)

// 3. Create a new dataset for bootstrap testing
clear all

// Create matrix with the group effects
matrix G = (2007, -20.34896 \ ///
2011, -5.983892 \ ///
2015, -3.803404 \ ///
2016, -4.865206)

// Convert matrix to dataset properly
set obs 4
generate treatment_year = .
generate effect = .

forvalues i = 1/4 {
replace treatment_year = G[`i',1] in `i'
replace effect = G[`i',2] in `i'
}

// Generate group indicators
gen g2007 = (treatment_year == 2007)
gen g2011 = (treatment_year == 2011)
gen g2015 = (treatment_year == 2015)
gen g2016 = (treatment_year == 2016)

// Bootstrap tests for each group
// 2007 group
regress effect g2007
boottest g2007, weighttype(webb) reps(9999) boottype(wild) seed(1234)

// 2011 group
regress effect g2011
boottest g2011, weighttype(webb) reps(9999) boottype(wild) seed(1234)

// 2015 group
regress effect g2015
boottest g2015, weighttype(webb) reps(9999) boottype(wild) seed(1234)

// 2016 group
regress effect g2016
boottest g2016, weighttype(webb) reps(9999) boottype(wild) seed(1234)

// Joint test of all groups
regress effect g2007 g2011 g2015 g2016
boottest {g2007} {g2011} {g2015} {g2016}, ///
weighttype(webb) reps(999) seed(1234) madjust(sidak)

[FernandoRios]

I dont think it will, because now the regression is not carrying over the "noise" of the estimation.
At the same time, i have no idea what would be a good way to simulate jwdid so that boottest works.

[Fredie Didier]

I see. Thanks for replying. I also don't have an idea on how to use boottest with jwdid