| multiple regression [message #30507] |
Mon, 29 April 2002 06:09  |
Klaus Scipal
Messages: 45
Registered: November 1997
|
Member |
|
|
Hi,
I am using the regress function to perform a multiple linear regression fit.
However I miss statistics to determine the significance of each coefficient
in the regression model.
Pv-wave for example offers the functionality of calculating a two tailed
t-test for each coefficient. Does anyone know of a similar thing in IDL (or
possibly a work around)?
Thanks,
Klaus
|
|
|
|
|
|
| Re: multiple regression [message #30686 is a reply to message #30507] |
Fri, 10 May 2002 00:46  |
Klaus Scipal
Messages: 45
Registered: November 1997
|
Member |
|
|
Sorry for not replying Andrews message in time
Andrew is actually right, T=B/sigma is what I need.
Stepwise sound interesting and implementation shouldn't be to difficult.
Using the regress function, calculate the T statistics, exclude variables
that are not signifcant and repeat that until a certain level of
confidence. However for the time being I am happy with Andrews advice.
Cheers,
Klaus
"William Gallery" <wgallery@aer.com> wrote in message
news:c6dd2c1c.0205080826.3d8e1bcc@posting.google.com...
> Andrew Noymer <noymer@socrates.Berkeley.EDU> wrote in message
news:<yx6hhelp5gnl.fsf@socrates.Berkeley.EDU>...
>> True, no t-stat in IDL.
>>
>> But t=B/sigma, and you have both of those I believe.
>>
>> Andrew
>
> There is an obsolete idl procedure stepwise.pro that performs a
> stepwise multiple regression: it ranks each variable by significance
> and removes variables with a significance below a threshold. This may
> be helpful to you.
>
> BTW, does anyone know why this routine is obsolete? There is nothing
> to replace it.
>
> Cheers,
>
> Bill Gallery
> AER, Inc.
|
|
|
|
| Re: multiple regression [message #30703 is a reply to message #30507] |
Thu, 09 May 2002 01:04 |
Klaus Scipal
Messages: 45
Registered: November 1997
|
Member |
|
|
Sorry for not replying
Andrew is actually right, T=B/sigma is what I need.
Stepwise sound interesting and it shouldn't be to difficult to implement it
using a simple while not statement and run the loop until a certain level of
confidence. However for the time being I am happy with Andrews advice.
Cheers,
Klaus
"William Gallery" <wgallery@aer.com> wrote in message
news:c6dd2c1c.0205080826.3d8e1bcc@posting.google.com...
> Andrew Noymer <noymer@socrates.Berkeley.EDU> wrote in message
news:<yx6hhelp5gnl.fsf@socrates.Berkeley.EDU>...
>> True, no t-stat in IDL.
>>
>> But t=B/sigma, and you have both of those I believe.
>>
>> Andrew
>
> There is an obsolete idl procedure stepwise.pro that performs a
> stepwise multiple regression: it ranks each variable by significance
> and removes variables with a significance below a threshold. This may
> be helpful to you.
>
> BTW, does anyone know why this routine is obsolete? There is nothing
> to replace it.
>
> Cheers,
>
> Bill Gallery
> AER, Inc.
|
|
|
|
| Re: multiple regression [message #30712 is a reply to message #30539] |
Wed, 08 May 2002 09:26 |
wgallery
Messages: 32
Registered: December 1998
|
Member |
|
|
Andrew Noymer <noymer@socrates.Berkeley.EDU> wrote in message news:<yx6hhelp5gnl.fsf@socrates.Berkeley.EDU>...
> True, no t-stat in IDL.
>
> But t=B/sigma, and you have both of those I believe.
>
> Andrew
There is an obsolete idl procedure stepwise.pro that performs a
stepwise multiple regression: it ranks each variable by significance
and removes variables with a significance below a threshold. This may
be helpful to you.
BTW, does anyone know why this routine is obsolete? There is nothing
to replace it.
Cheers,
Bill Gallery
AER, Inc.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
