random treatment,50 times loop and store the estimate with p value for the variable of interest (POST) and display.

[Aditi Roy]

Hi,

I want to randomly assign treatment and run the same regression in the loop for 50 times and then I want to see how many times my variable of interest (POST) is significant. It's better if I can store the estimate with the p-value (or in a star representation). I tried the following. Where am I going wrong? Any help is appreciated.


drop post
drop random
foreach i of numlist 1/50 {

generate random=uniform()
sort random
generate post=0
replace post=1 in 1/50

tsset id year

set matsize 800
set more off

reg agriculture post i.country_id i.year i.country#c.line_time_trend [aw=ypop], cluster( country_id )
est sto b_1

esttab b_1,keep( post ) se star(* 0.10 ** 0.05 *** 0.001)
est clear //estimates clear


matrix p = J(`post',1,.)
matrix rownames p = `y'
matrix colnames p = "p"

matlist p

// for the computation of the p-values
local i = 1
foreach var of local y {
reg `var' post
matrix p[`i++',1] = 2*ttail(e(df_r), abs(_b[post]/_se[post]))
}

//
matlist p
}

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. You can also use a simpler model - the extra variables just add complexity. Without data, we can't replicate your problem to see what is going on.

Your regression saves a matrix called r(table). If you don't mess with the order of variables, you can refer to the probability of a given variable based on its location in r(table). I don't know if you need to copy r(table) into a different matrix name first.
Instead of the second loop, I'd just set up a larger matrix before the loops with all missing, and a counter for the row of the matrix, then I'd write the results into the right places in the matrix. Sometimes, instead of a matrix, I write the values into a variable so I can easily do statistics.

[Mike Lacy]

As an addition to Phil's comments, I'd suggest you take a look at the built-in -permute- command, which sounds close to what I think you want.

[Aditi Roy]

Thanks, Phil. However, it is still not working. I am getting the p-value of the last regression only for Post. I want all the "p values" for the variable post after the loop runs 50 times and then I want to count the of times p-values of the variable post was under 0.10.

[code delimiters][set matsize 5000
set more off
unab y : agriculture
local l: word count `post'
matrix p = J(100,5,.) //defined matrix
matrix rownames p = `y'
matrix colnames p = "p"
foreach i of numlist 1/50 {
set seed 2846895
generate random=uniform()
sort random
generate post=0
replace post=1 in 1/1600
reg agriculture post i.country_id i.year i.country_id#c.line_time_trend [aw=ypop], cluster( country_id )
matrix p = 2*ttail(e(df_r), abs(_b[post]/_se[post]))
drop random
drop post
}
//to display the value of p
matlist p
count if p<=0.10]

Can anyone please help!