Paul Bernhardt wrote:

> I've never really thought of it in these terms, so, to see if I
> understand correctly, let me state it differently. There is one point
> through which you are sure the regression line passe, X-bar,Y-bar The
> terms for slope and intercept are estimates based on the data and have
> associated error. The divergence of the 'confidence bands' is a
> consequence of the combination of extremes of slope and intercept. Is
> that what you were saying? Therefore the divergence has nothing to do
> with the relative paucity of data at the extremes, nor would
> 'oversampling of the extremes' remove this divergence characteristic.
> Right?

Exactly. Of course oversampling at the extremes will reduce the error
of the slope, and hence the divergence of the confidence band will be
reduced as well.





.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to