Prev: 3D Scatter Plot w. 45degree plane
Next: Standard N(0,1) test and PROC CAPABILITY NORMALTEST
From: Sid N on 13 Dec 2009 22:10
Matthew, thank you for your reply. Yes, you are right. Parameter A cannot be
equal to -min(column1). It can only be greater. In my existing model in
Excel, I do not have an upper bound on A. Do you think it is solvable in SAS
by changing the lower bound to A > -min(column1)?

Sid

On Sun, Dec 13, 2009 at 3:39 PM, Zack, Matthew M. (CDC/CCHP/NCCDPHP) <
mmz1(a)cdc.gov> wrote:

> I'm not sure that your model has a solution. According to your
> specifications,
>
> log(column1 + A) = C * column2 + B
>
> log(column1 + A) = C * column2 + max(log(column1 + A)) - max(column2)
> * C
> = C * (column2 - max(column2)) + max(log(column1 +
> A))
>
> where A >= - min(column1).
>
> Actually, in this last specification, A must exceed -min(column1)
> because, if A equals -min(column1),
> the program must take the logarithm of zero, which is inadmissible, at
> the observation where
> column1 = min(column1).
>
> Since A has no upper bound, log(column1 + A) and max(log(column1 + A))
> increase without bound.
> However, the difference between column2 and max(column2) for each
> observation remains constant
> no matter what happens to A but never exceeds zero. Therefore, as A
> increases without bound,
> C decreases towards zero. The error sum of squares decreases towards
> zero so that r-squared
> increases to 1.00.
>
> Matthew Zack
>
> -----Original Message-----
> From: SAS(r) Discussion [mailto:SAS-L(a)LISTSERV.UGA.EDU] On Behalf Of Sid
> N
> Sent: Friday, December 11, 2009 5:47 PM
> To: SAS-L(a)LISTSERV.UGA.EDU
> Subject: Re: PROC NLP and R^2
>
> Please note this correction on the parameter "B" constraint:
> B = max(log(column1 + A)) - max(column2) * C
>
> Sid
> On Fri, 11 Dec 2009 16:41:57 -0600, Sid N <nsid31(a)GMAIL.COM> wrote:
>
> >Hi,
> >
> >This is a follow-up question regarding non-linear modeling from my post
>
> >last month. Although I have received some useful input at that time, I
> >am still not able to replicate the results that I am getting with the
> Excel Solver.
> >
> >I may have left out some details in my original post (below) which may
> >be pertinent to the question at hand. The original equation is of the
> form:
> >
> >log(column1 + A) = C * column2 + B
> >
> >Here are the constraints on the paramaters:
> >
> >A >= -(min(column1))
> >B = max(log(column1 + A)) - max(column2) * B C is the slope of the
> >predicted log(column1 + A) values and the original
> >column2 values.
> >
> >I would like to know how I could set the initial parameters and bounds
> >on parameters B and C, which are dependent on other parameter(s) and
> >variables while using PROC NLIN/PROC NLMIXED/PROC OPTMODEL?
> >
> >Thank you again in advance.
> >
> >Sid
> >
> >On Thu, Nov 12, 2009 at 5:03 PM, Sid N <nsid31(a)gmail.com> wrote:
> >
> >> Hi,
> >>
> >> I would like to find the maximum R^2 that can be attained between
> >> log(column1 + a) and (c * column2 + b) for the below dataset by
> >> changing the values for a, b and c. Previously, I have used the
> >> 'Solver' add-in in Excel.
> >>
> >> data have;
> >> input column1 column2;
> >> datalines;
> >> 30000 6000000
> >> 50000 10000000
> >> 65000 13000000
> >> 70000 13500000
> >> 76000 13525000
> >> 113000 13925000
> >> 115500 15925000
> >> 119500 16925000
> >> 121500 17425000
> >> 122500 17725000
> >> ;
> >> run;
> >>
> >> Based on what I have read from the SAS-L posts, I understand that
> >> PROC NLP can help me do the same in SAS. Can someone please assist me
>
> >> with using this procedure? Thank you for your time.
> >>
> >> Sid
> >>
>
>
>
First  |  Prev  | 
Pages: 1 2
Prev: 3D Scatter Plot w. 45degree plane
Next: Standard N(0,1) test and PROC CAPABILITY NORMALTEST