# RE: st: Bell Curve of Predicted Values

 From "Nick Cox" To Subject RE: st: Bell Curve of Predicted Values Date Mon, 17 Apr 2006 15:33:17 +0100

There are several issues here:

1. As a matter of Statalist procedure, your reference
to Woolridge presupposes that people can recognise that
reference, which in many cases will be wrong. Name and
date references are inadequate, as the FAQ makes clear.

2. You can draw a normal curve using -twoway function-
plugging in the mean and standard deviation you desire,
which will be accessible via -predict-.

3. This will tell you about the sensitivity of predictions
only in the very limited sense that it assumes that
the model assumptions hold exactly, which is unlikely.

Nick
n.j.cox@durham.ac.uk

Randall

> Maarten, thanks.  This seems very close to what I
> would like to do.  I had been calculating my predicted
> values using matrix language, just because I had
> particular x-values I was interested in obtaining a
> prediction for.  Using your example, I did something
> like this:
>
> reg price mpg foreign
> matrix yhat=_b[_cons] + _b[mpg]*12 + _b[foreign]*1
> matrix list yhat
>
> So in essence, I wanted a prediction for mpg=12,
> foreign=1.
>
> So my question is this:  is there a way of generating
> a distribution based on particular criteria (i.e.
> mpg=12, foreign=1)?

Maarten buis

> > I this what you are looking for?
> >
> > *--------begin example----------
> > sysuse auto, clear
> > reg price mpg foreign
> > predict yhat, xb
> > hist yhat, normal
> > *-------end example------------

Randall

> > > I have a multivariate OLS regression, and I am
> > only
> > > interested in the predicted value of the outcome
> > > (y-hat).  I am also interested in the sensitivity
> > of
> > > this predicted outcome, and would like to show
> > this in
> > > a graph, specifically a bell curve.  Does anyone
> > know
> > > a way to create a graph which can show a bell
> > curve of
> > > predicted values? (i.e. probability on the y-axis
> > and
> > > predicted values on the x-axis)
> > >
> > > I can get the predicted value, and I can get 95%
> > > confidence intervals for predicted values in Stata
> > > using the method of Woolridge, 2nd Ed., p. 206.
> > This
> > > way calls for adjusting the data and re-running
> > the
> > > regression so that the constant's 95% confidence
> > > intervals are those of the predicted values.  But
> > this
> > > is merely two more points in the bell curve; I am
> > > looking for the entire distribution graphically.

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/