s-news
[Top] [All Lists]

help with a local regression model

To: "SPLUS_NEWS" <s-news@lists.biostat.wustl.edu>
Subject: help with a local regression model
From: "Bill Shipley" <bill.shipley@usherbrooke.ca>
Date: Thu, 24 Feb 2005 13:37:34 -0500
Importance: Normal

Hello.  I am analysing data using local (smoother) regression in which I have to constrain the first derivatives of two groups over a section of the independent data.  I don’t know how to do this, or if it is possible.  Here is the experimental design (simplified to a single species):

 

N1, and N2 plants are grown in a growth chamber under two different treatments.  The “control” treatment consists of sequential harvests of individuals over time under a constant nutrient supply rate and the biomass of the plants are estimated.  I estimate the 1st derivative of log(biomass) ~ time (called the relative growth rate) using the cubic spline smoother (smooth.spline() ).  The “treatment” consists of the same sequential harvests of individuals over the same time period. The plants during the first 6 harvests are grown under the same conditions as in the control treatment.  Following this the nutrient supply rate is reduced to a new lower level.  The objective of the analysis is to estimate the way the slope changes as the plants adjust to the new conditions.

Because the growth conditions are identical in the two sets of data (treatment/control) for the first 6 harvests, I want to be able to constrain the smooth.spline to estimate the same 1st derivatives in the two sets over these first 6 harvests; however the actual values will not be the same since the plants were of slightly different sizes at the beginning of the experiments.  After the 6th harvest, I want the smooth.spline to freely estimate the 1st derivatives separately in the two groups of data.

 

Essentially, I want to fit a local regression separately in these two data sets but keep the 1st derivatives equal in these two data sets until time t; after time t I want to let the 1st derivatives be different.

 

Can this be done somehow (perhaps using gam?).  If so, how can I do it?

 

Thanks.  Any suggestions would be helpful.

 

Bill Shipley

 

<Prev in Thread] Current Thread [Next in Thread>
  • help with a local regression model, Bill Shipley <=