# Re: st: ST: predict vs predcalc

 From Austin Nichols To statalist@hsphsun2.harvard.edu Subject Re: st: ST: predict vs predcalc Date Sun, 27 Sep 2009 09:14:27 -0400

```You might want to start by reading e.g.
http://www.stata.com/statalist/archive/2002-09/msg00052.html
http://www.stata.com/statalist/archive/2002-12/msg00193.html

On Sun, Sep 20, 2009 at 12:09 PM, Laura Platchkov <LMP881@bham.ac.uk> wrote:
> Dear all,
>
> Does anyone has a clue?
>
> I have a cross sectional dataset, and use OLS. This is the model:
>
> ln(Y) = b1 + b2X + b3Z + e
>
> I would like to do some predictions to compare the change in Y when the X has, let say 10 units more:
>
> ln(Y)* = b1 + b2(X+10) + b3Z + e
>
> and compare the change:
>
> Y*- Y
>
> I have 2 questions:
>
> 1) But I will have the change ln(y*) -ln(y). If I compute the percentage change between both log,it would n't be the same as the levels. I should first transpose it in level, right?
> Wooldridge says that to get rid of the bias, we should multply the exponential of the predicted value by an estimator of the expected value of the exponential of the error term.
>
> 2) In general, how can I do exactly? I have created a new dataset with the modified values of the regressors and was planning to do out of sample predictions with the predict command. But the results look strange. I am mistaken in the technique? Or should I use the predcalc command? What is the difference between the predict out-of-sample and the predcalc command?
>
>
> Thanks a lot!
>
> laura
>

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