Re: Logistic regression and non-gaussian variables

From: juan cristobal (xxx_at_xxx.xx)
Date: 01/29/05 

Date: Sat, 29 Jan 2005 15:28:56 +0100

Thanks.
My thoughts were confused about these transformations. By "smoothing"
slightly the initial distributions, i have been thinking that the aim was to
"normalize" them, ok it's stupid.:)
I should have think it over.Bye.

"Richard Ulrich" <Rich.Ulrich@comcast.net> a écrit dans le message de
news:vaokv09g8201ckaqrt0qfs8qhaegf3af33@4ax.com...
> On Fri, 28 Jan 2005 12:55:15 +0100, "juan cristobal" <xxx@xxx.xx>
> wrote:
>
> > Hi!
> > If I take the "amount of a credit" as a variable for my logistic
regression,
> > tests for normality aren't ok.
> > Indeed, people don't ask for small amounts of money, so that the
> > distribution has a tail on the right and is equal to 0 in the small
values.
> >
> > Question :
> > Do i have to transform the variable before computing?What are the common
> > transformations ?(in this case and generally : log?sqrt?...)
>
> What you want to think about is linearity.
> - What makes equal steps from small to large?
> Does "$1000" belong in the same equation with a million?
>
> The "power family" is usual, because taking square
> root, log, reciprocal, etc., has the good chance for
> preserving logical relations. That is *usually* desirable.
>
> Taking the log is the common one for money.
> If you playing around, compare other results.
>
>
> --
> Rich Ulrich, wpilib@pitt.edu
> http://www.pitt.edu/~wpilib/index.html

Relevant Pages

  • Re: Question about null space
    ... zero vector property the magnitude of the vector is unimportant. ... This is also common when ... So if you multiply your vector by -1, and then normalize it you will ...
    (comp.soft-sys.matlab)
  • Which is More Efficient - Lots of One Field Tables or Larger Multi-Field Tables
    ... I am trying to decide how to "normalize" a large database. ... 25 fields with many common values like State, County, Street Address, MapID, ... those common values, but that seems even less efficient. ...
    (microsoft.public.access.tablesdbdesign)
  • Re: Advice needed on programming style
    ... I assume that by normalize, you mean merge the tables so that the common ... Due to the design of the program (which is ... argument without having or needing prior knowledge of which type is ...
    (microsoft.public.dotnet.languages.vb)
  • Re: Which is More Efficient - Lots of One Field Tables or Larger Multi-Field Tables
    ... 25 fields with many common values like State, County, Street Address, MapID, etc. ... I can normalize this table and create many linked tables with each of those common values, but that seems even less efficient. ... If you do lots of updates, then you might not want to have many copies of any one field, so you'd want more linked lists. ...
    (microsoft.public.access.tablesdbdesign)