"Mitch" wrote:

I need to run regression analyses (linear & nonlinear) on many data sets. I
have observed that the resulting equations generated in a chart trend line
and that generated by the linest() function are frequently very different. I
have read Tushar Mehta's explanation of the problem, but need advice on which
is more accurate.

I also would be interested in this. I have heard of a program that
significantly increases Excels capabilities beyond the 15 or 16 significant
digits. Has anyone used these and had positive results? It was some time ago
when I first heard of them, about the time this post originated.

Thanks
--
David


"The Gourou" wrote:



"Mitch" wrote:

I need to run regression analyses (linear & nonlinear) on many data sets. I
have observed that the resulting equations generated in a chart trend line
and that generated by the linest() function are frequently very different. I
have read Tushar Mehta's explanation of the problem, but need advice on which
is more accurate.

A couple of Excel add-ins that support user specified numeric precision are
http://digilander.libero.it/foxes/index.htm
http://precisioncalc.com/

You can also get infinite or user specified precision from symbolic
manipulators
such as Maple, Mathematica, MacSyma and open source packages like Maxima
http://maxima.sourceforge.net/

As I mentioned last May, Excel's precision is no better or worse than that
of dedicated statistics packages. Precision issues between Excel and other
packages are more a function of the algorithm employed, than the inherent
precision.

Jerry

"David" wrote:

I also would be interested in this. I have heard of a program that
significantly increases Excels capabilities beyond the 15 or 16 significant
digits. Has anyone used these and had positive results? It was some time ago
when I first heard of them, about the time this post originated.

Thanks
--
David


"The Gourou" wrote:



"Mitch" wrote:

I need to run regression analyses (linear & nonlinear) on many data sets. I
have observed that the resulting equations generated in a chart trend line
and that generated by the linest() function are frequently very different. I
have read Tushar Mehta's explanation of the problem, but need advice on which
is more accurate.

Thank you Jerry

David

"Jerry W. Lewis" wrote:

A couple of Excel add-ins that support user specified numeric precision are
http://digilander.libero.it/foxes/index.htm
http://precisioncalc.com/

You can also get infinite or user specified precision from symbolic
manipulators
such as Maple, Mathematica, MacSyma and open source packages like Maxima
http://maxima.sourceforge.net/

As I mentioned last May, Excel's precision is no better or worse than that
of dedicated statistics packages. Precision issues between Excel and other
packages are more a function of the algorithm employed, than the inherent
precision.

Jerry

"David" wrote:

I also would be interested in this. I have heard of a program that
significantly increases Excels capabilities beyond the 15 or 16 significant
digits. Has anyone used these and had positive results? It was some time ago
when I first heard of them, about the time this post originated.

Thanks
--
David


"The Gourou" wrote:



"Mitch" wrote:

I need to run regression analyses (linear & nonlinear) on many data sets. I
have observed that the resulting equations generated in a chart trend line
and that generated by the linest() function are frequently very different. I
have read Tushar Mehta's explanation of the problem, but need advice on which
is more accurate.