Date: Tue, 31 Jul 2001 08:56:58 0400
ReplyTo: Steve Rowe <steverowe@EMAIL.MSN.COM>
Sender: "SAS(r) Discussion" <SASL@LISTSERV.UGA.EDU>
From: Steve Rowe <steverowe@EMAIL.MSN.COM>
Subject: Re: Growth rate
ContentType: text/plain; charset=ISO88591
Modeling the growth curve as suggested below assumes that the growth is
unbounded – in fact, the exponential model assumes that even the rate of
growth increases unboundedly. This model is appropriate for some phenomena
such as unconstrained population growth, but what about the growth of an
individual? Many growth curves are ‘sigmoidal’ – starting out slow,
increasing faster and faster up to a limit, then the rate of growth drops
off and finally reaches 0 when the growth cycle is complete. MOST IMPORTANT
TO NOTE HERE IS THAT YOU CAN EASILY USE AN INAPPROPRIATE MODEL TO GET
NUMBERS THAT MAKE YOU THINK YOU HAVE AN ANSWER, AND NEVER HAVE A CLUE THAT
YOU ARE ‘WAY OUT IN LEFT FIELD’. SAS is fantastic for database programming,
but using it for statistical inference requires mathematical and
statistical training. No harm if such mistakes really aren’t that
important, but then why do the analysis?
On Mon, 30 Jul 2001 21:10:53 0500, Peter Grandstaff <grandsta@SWBELL.NET>
wrote:
>My approach would be standard elementary (classroom) econometric:
>
>First choice:
>Fit separate exponential regressions for a, b, c. Record sum of squared
>residuals.
>Fit a "grand" regression for all the data together. Record sum of squared
>residuals.
>Compare whether the sums from the three regressions (added up) are a great
>deal less than the grand regression sum whether the separate fits reduce
>total unexplained variation a great deal.
>An F statistic based upon the reduction in variation can be judged in
>comparison to critical values for 6 and 3x(N2) degrees of freedom (if N
>equals the number of observations in each panel), I think.
>
>Second choice:
>Run the regression:
>ln(dependent)=a0+ra*(year)+rb*(dummyb*year)+rc*(dummyc*year)
>where dummyb=0 or 1 depending upon whether the data pertain to "panel" b
>where dummyc=0 or 1 depending upon whether the data pertain to "panel" c
>The rb and rc coefficients can be judged as significantly different from
>zero or not.
>
>Interested in other answers posted. Will provide references if the methods
>survive your and "peer" review.
>
>
>
> Original Message 
>From: "P. L" <li_9@HOTMAIL.COM>
>Newsgroups: bit.listserv.sasl
>To: <SASL@LISTSERV.UGA.EDU>
>Sent: Monday, July 30, 2001 7:23 PM
>Subject: Growth rate
>
>
>> I have a dada set like this:
>>
>> input group $ year sales;
>> datalines;
>> a 1998 1000
>> a 1999 2000
>> a 2000 3000
>> ...
>> c 1998 1200
>> c 1999 2300
>> c 2000 3400
>> ;
>>
>> My question is: What I should do if I want to examine if or not
>> differences of growth rates among groups are statistically significant.
>Any
>> help will be apprecaited.
>>
>> P. Lee
