Power estimates for clustered samples

samplsi-sampclus.do

//Power calcs start here
sum l_household_income_total if (treatment == 0 & round == 1)
local mean1 = r(mean) //mean in control
local sd1 = r(sd) 	  //SD in control

sum l_household_income_total if ((treatment == 1 | treatment == 2) & round == 1)
local mean2 = r(mean) //mean in AFSP
local sd2 = r(sd) 	  //SD in AFSP

loneway l_household_income_total hhid_dvdc //--> 0.13
local rho = r(rho) // intra-cluster correlation

local obsclus = 10 

local ratio = 1

reg l_household_income_total bl_w_total_income
local r0 = _b[bl_w_total_income]
local r1 = `r0'
local r01= `r0' 
//r0 is supposed to be correlation btw pre-T rounds
//r1 is supposed to be correlation btw post-T rounds
//r01 is supposed to be correlation btw pre and post
//Assuming all 3 to be equal for simplicity

local pre = 1 // Number of pre-treatment rounds
local post = 2 // Number of post-treatment rounds
di "mean1: `mean1' mdes rho: `rho' gamma: `r0'"

//Going to look for samp sizes for a number of potential effect sizes. The mean log-income is about 11.5, so going to work on that basis. 
foreach effect in 1 1.5 2 2.5 3 3.5 4 4.5{
qui {
local mdes = `effect' // MDES

di "mdes = " `mdes'

local mean2 = `mean1' + `mdes'
sampsi `mean1' `mean2', sd1(`sd1') sd2(`sd2') ratio(`ratio') pre(`pre') post(`post') r0(`r1') r1(`r1') r01(`r01') method(ancova)
}
di `effect'
sampclus, rho(`rho') obsclus(`obsclus')
}