home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
TPUG - Toronto PET Users Group
/
TPUG Users Group CD
/
TPUG Users Group CD.iso
/
PET
/
S-Super PET
/
(s)t5.d64
/
EDA.REGRESS.INF
< prev
next >
Wrap
Text File
|
2009-01-18
|
3KB
|
51 lines
PPPP multiple regression PPPP
@@@@@@@@@@@@@@@@@@@
introduction
@@@@@@@@@@@@ the function regress uses the modified gram_schmidt
procedure for performing multiple regression analyses.
this procedure is regarded as the best available in
terms of speed and numerical stability for fitting a
regression relation to a point cloud.
parameters there are four parameters which can be set by the user.
atom :default 10P@6" _prevents Kblowing upK when
one of the carriers is a linear
combination of previously swept
out carriers.
epsilon :default .01" _proportion by which sum of the
absolute residuals must be
reduced to continue iterating.
constant :default 1" _causes the constant carrier to
be included when set to 1 and
neglected when set to 0.
cutoff :default 4" _biweight constant.
command syntax
@@@@@@@@@@@@@@ regress dddd _where dddd is a numeric array
z[regress dddd _store residuals in z
z[regress dddd;<1 3 4 8' _regard col 8 as the response
and cols 1 3 4 as carriers.
comments
@@@@@@@@ the coefficients are displayed and stored in the
global variable cf.
care needs to be exercised in differentiating
between carriers and response variables. the response
variable is the last column named in dddd. the order
in which carriers are specified does not affect the
the result.
PPPP practice PPPP
@@@@@@@@
this workspace contains the data set called logcrimes. chapter 6 of
mcneil gives an analysis of this data set. try the following commands>_
logcrimes _display the data set.
PP n.b. col1 _ log murder rates
col2 _ log assult rates
col3 _ per cent urban
col4 _ log rape rates
z[regress logcrimes