Re: st: simulated value of a ratio does not compare to manually calculated value

[Austin Nichols <austinnichols@gmail.com>]

Ariel Linden, DrPH <ariel.linden@gmail.com>:

The mean of a ratio generically does not equal the ratio of the means.
Put
 sc mse_t*
 corr mse_t*
 g r=mse_tr/mse_tt
 su r mse_true mse_tt mse_ratio

On Fri, Apr 26, 2013 at 4:28 PM, Ariel Linden, DrPH
<ariel.linden@gmail.com> wrote:
> Hi All,
>
> Below is a simulation that I am running in which I generate a ratio
> (mse_ratio) which is simply the ratio of the mse_true / mse_tt
>
> When running the simulation, the mse_ratio does not calculate correctly,
> even though each component does come through correctly...
>
> More specifically, you can see from the output below that the mse_true =
> .0038127, and the mse_tt = .0352618  , which should
> elicit a ratio of .10812551
>
> However, the mse_ratio value is  91.79034.
>
> I cannot figure out what in my code is not working here... Any help is
> appreciated... See my code below...
>
> Thanks in advance!
>
> Ariel
>
>
> * this is the results of the simulation code from below
> . sum mse_true mse_tt mse_ratio
>
>     Variable |       Obs        Mean    Std. Dev.       Min        Max
> -------------+--------------------------------------------------------
>     mse_true |        50    .0038127    .0050518   9.49e-07   .0231653
>       mse_tt |        50    .0352618    .0266384   4.62e-07   .1037486
>    mse_ratio |        50    91.79034    619.6749   .0000267   4382.196
>
>
> *********************************
> * Start code
> *******************************
> program drop _all
>
> program define sim, rclass
>
> drop _all
>
> set obs 1000
> generate x=1
> replace x=0 if _n>1000/2
> generate ybase=invnorm(uniform())
> generate e1=invnorm(uniform())
> generate pi=(1-x)/(1+exp(-1+0.1*ybase)) + x/(1+exp(+0.9-0.1*ybase))
> generate u1=uniform()
> gen a=0
> replace a=1 if pi>=u1
> gen ysqr=ybase^2
> gen intxy=x*ybase
> gen intxy2=x*ysqr
> generate y=0.5*a-0.45+0.35*ybase+0.1*x+0.3*intxy+0.3*intxy2+0.25*ysqr+e1
>
> * true estimates of Y
> regress y a x ybase intxy ysqr intxy2
> scalar tx = _b[a]
> return scalar tx = tx
>
> scalar mse_true = (_b[a]-0.50)^2
> return scalar mse_true = mse_true
>
> *ttest estimates - naïve
> ttest y, by(a)
> scalar ttest1 = r(mu_2)-r(mu_1)
> return scalar ttest1 = ttest1
>
> scalar mse_tt = (ttest1-0.50)^2
> return scalar mse_tt = mse_tt
>
> scalar mse_ratio = mse_true / mse_tt
> return scalar mse_ratio = mse_ratio
>
> end
>
> set seed 1234
> simulate true=r(tx) ttest1=r(ttest1) mse_true=r(mse_true) ///
> mse_tt=r(mse_tt) mse_ratio=r(mse_ratio), reps(50): sim
>
> sum mse_true mse_tt mse_ratio

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/