S******y
发帖数: 1123 | 1 This seems to be a simple problem. But I just could not figure out -- |
A*******s
发帖数: 3942 | 2 Box-Cox transformation? How about transform the equation to
log(Y-a)=log(b)+X*log(c)
then try a series of values of a to see which one fits better? Just my two
cents, correct me please.
【在 S******y 的大作中提到】
: This seems to be a simple problem. But I just could not figure out --
|
S******y
发帖数: 1123 | 3 Thanks!
Is there a formal way to solve this?
such as optimization to minimize discrepancies between y and y-hat by
searching thru parm space? |
A*******s
发帖数: 3942 | 4 I think you can use box-cox transformation, but i'm not sure how to include
the term a.
【在 S******y 的大作中提到】
: Thanks!
: Is there a formal way to solve this?
: such as optimization to minimize discrepancies between y and y-hat by
: searching thru parm space?
|
W**********E
发帖数: 242 | 5 You can fit a nonlinear regression model using SAS. I think you can define
function form y=a+b(c^x) in nlin or model procedure.
When you choose initial values of parameters, you could either make a guess
or use the suggestions from the post above---transform y and x and fit
linear regression--get crude estimates of log(b) and log(c) and exponentiate
them.
The headaches of running non-linear regression always come from choosing
initial values.
【在 S******y 的大作中提到】
: Thanks!
: Is there a formal way to solve this?
: such as optimization to minimize discrepancies between y and y-hat by
: searching thru parm space?
|
P****D
发帖数: 11146 | 6 1. Feed your data and the relationship to SAS proc nlin. That proc is
usually helpful in finding initial values for estimates. After you have good
initial values you can improve them using nlmixed if needed.
2. Good luck... God bless you... |
S******y
发帖数: 1123 | 7 Thank both of you for help!
I only tried PROC MODEL. which seems working well! |
