Creating 1000 replications of regression with shuffled variable, and saving coefficients.

[Yaya Mensah]

Dear all,

Please forgive the rookie question.

I want to perform a placebo test to show that my IV results are not the result of a spurious correlation. To that end, I want to perform 1000 replications of my IV specification, shuffling the endogenous variable (within each year) each time. I want to save the coefficients and p-values on the endogenous variable so that I can depict them on a density plot.

I know that by using the -shufflevar- command with cluster(year) I can shuffle the endogenous variable by year and then run the regression. This would then be:

shufflevar endo, cluster(year)
xtivreg2 y (endo_shuffled = instrument) controls, fe

To do it once. What I don't know how to do is to run this 1000 times, saving the coefficient on endo_shuffled each time, so that I can produce the density plot. I don't know this because my knowledge of stata is not yet at a level where I know how to program loops etc.

Therefore, I would appreciate any help.

Thankyou.

[Andrew Musau]

shufflevar and xtivreg2 are from SSC (FAQ Advice #12). The following uses the official xtivreg and is not tested.

frame create results
frame results{
    set obs 1000
    g coef=.
    g pval=.
}
forval i=1/1000{
    shufflevar endo, cluster(year)
    xtivreg y (endo_shuffled = instrument) controls, fe
    frame results{
        replace coef= `=r(table)["b", "`e(instd)'"]' in `i'
        replace pval= `=r(table)["pvalue", "`e(instd)'"]' in `i'
    }
}
frame change results
browse

[Yaya Mensah]

Thank you, Andrew. Unfortunately it doesn't run for me, the problem seems to come here:

replace coef= `=r(table)["b", "`e(instd)'"]' in `i' replace pval= `=r(table)["pvalue", "`e(instd)'"]' in `i'

[Andrew Musau]

What version of Stata do you have?

[Yaya Mensah]

V 16.1,

But don't worry Andrew, I managed to write a solution myself using a loop and then the -simulate- command.

Let me know if you would like me to post it?

[Andrew Musau]

Let me know if you would like me to post it?

Yes, it should be useful for anyone following this thread.

[Joro Kolev]

-permute- should do the trick here as well. E.g.,

. sysuse auto, clear
(1978 automobile data)

. permute mpg _b, saving(bTEMP, replace): ivregress 2sls price (mpg = headroom turn weight)
(running ivregress on estimation sample)
(file bTEMP.dta not found)

Permutations (100): ..........10..........20..........30..........40..........50..........60......
> ....70..........80..........90..........100 done

Monte Carlo permutation results                    Number of observations =  74
Permutation variable: mpg                          Number of permutations = 100

      Command: ivregress 2sls price (mpg = headroom turn weight)

-------------------------------------------------------------------------------
             |                                               Monte Carlo error
             |                                              -------------------
           T |    T(obs)       Test       c       n      p  SE(p)   [95% CI(p)]
-------------+-----------------------------------------------------------------
         mpg | -323.0714      lower      45     100  .4500  .0497  .3503  .5527
             |                upper      55     100  .5500  .0497  .4473  .6497
             |            two-sided                  .9000  .0300  .8412  .9588
             |
       _cons |   13045.8      lower      55     100  .5500  .0497  .4473  .6497
             |                upper      45     100  .4500  .0497  .3503  .5527
             |            two-sided                  .9000  .0300  .8412  .9588
-------------------------------------------------------------------------------
Notes: For lower one-sided test, c = #{T <= T(obs)} and p = p_lower = c/n.
       For upper one-sided test, c = #{T >= T(obs)} and p = p_upper = c/n.
       For two-sided test, p = 2*min(p_lower, p_upper); SE and CI approximate.

. use bTEMP
(permute mpg : ivregress)

. hist _b_mpg
(bin=10, start=-5132.626, width=890.80593)

. summ _b_mpg

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
      _b_mpg |        100   -90.67279    1254.967  -5132.626   3775.433

.