Date: Thu, 7 Jul 2005 14:51:13 -0400
Reply-To: Peter Flom <flom@NDRI.ORG>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Peter Flom <flom@NDRI.ORG>
Subject: Re: Modeling Yes/No variable
Content-Type: text/plain; charset=US-ASCII
Thanks for this clarification
Here, I see no reason to use other than the canonical (usual) link
function......but I may be missing something
Peter
>>> "Cook, Cody" <Cody.Cook@ERIEINSURANCE.COM> 7/7/2005 11:50:29 AM
>>>
Okay... I am trying to quantify the impacts of several policyholder
characteristics on renewal ratios (renew with us or not?). I can do
this by modeling the probability they will stay with us (0 means they
didn't... 1 means they did). I have over 11 million observations... I
have characteristics such as how long have they been with our
company...
geodemographic data... prior loss experience... etc. By understanding
what has happened in the past, I hope to shed some light on the near
future, given predicted x-combination outcomes.
My brothers, a computer programmer and a high school math teacher,
would
probably each be asleep by now... or giving me lip about how boring my
job is... but, given they paid attention, I believe they'd understand
my
objective at this point.
Cody Cook
-----Original Message-----
From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of
Peter Flom
Sent: Thursday, July 07, 2005 11:43 AM
To: SAS-L@LISTSERV.UGA.EDU
Subject: Re: Modeling Yes/No variable
>>> "Cook, Cody" <Cody.Cook@ERIEINSURANCE.COM> 7/7/2005 11:35:18 AM
>>>
<<<
Forgive me for my limited knowledge and the relatively vague e-mail.
>>>
Well, we all have limited knowledge. :-)
<<<
Yes/No is the answer to a given question... for example, lets say the
class variable I am speaking of is State (which is one of several
other
descriptive variables within the analysis, for example, gender,
salary,
etc). We might expect a similar probability of a person saying yes
from
state-to-state, but we want to quantify the impact of the participant
being from a particular state (say the countrywide probability is 90%
say yes -- we might expect dummy variables for Alabama and New York
tohave slightly different means -- we want to be able to quantify this
difference, given our data).
>>>
Independent variables of this type can be handled in LOGISTIC and also
in GENMOD.
<<<
I didn't use logistic because I have never used it... bad reason, I
know, but I have only been using SAS for about a year. I tend to
operate with procs that I am functional with... and only deviate when
it
is completely necessary to invest the time to switch. I am
comfortable
with learning LOGISTIC if you think this would be a better
approach/procedure.
>>>
Nope, GENMOD should be fine, I just think LOGISTIC is easier, probably
because it's what I am used to.
<<<
> what are the IVs? What is N?
I believe you mean what is N in the dependent variable... which would
beanswered above (yes/no to given question)... however, I'm not
certain
I follow... have I answered above?
>>>
By 'what is N' I meant what's the total sample size?
But giving sensible answers to this sort of question always requires
context. Explain what you are trying to do, as if you were explaining
it to your brother who knows no statistics.
<<<
I apologize for the confusion of adding Poisson and Gamma to the
discussion as they obviously are not distributions useful in this
particular study and set of dependent variables. I spoke of them in
reference to error distribution / link function combinations -- not as
options for this particular analysis. Again, sorry for the
confusion...
>>>>
Oh, OK. I think that's partly my fault.....maybe I didn't read your
original post carefully enough
Peter
Peter L. Flom, PhD
Assistant Director, Statistics and Data Analysis Core
Center for Drug Use and HIV Research
National Development and Research Institutes
71 W. 23rd St
www.peterflom.com
New York, NY 10010
(212) 845-4485 (voice)
(917) 438-0894 (fax)
-----------------------------------------
Disclaimer: This message (and any attachments) is confidential and is
intended only for the addressee(s). This message may contain
information
that is protected by one or more legally recognized privileges. If the
reader of this message is not the intended recipient, I did not intend
to
waive, and I do not waive, any legal privilege or the confidentiality
of
the message. If you receive this message in error, please notify me
immediately by return e-mail and delete this message from your computer
and
network without saving it in any manner. The unauthorized use,
dissemination, distribution, or reproduction of this message,
including
attachments, is prohibited and may be unlawful.
