Dear All, I am trying to specify the following fixed effects model for lme: y ~ constant1 - beta1*(x - beta2) where y is the response, x is the independent variable, and the operators above are real arithmetic operations of addition, subtraction, and multiplication. I realize that this model is just a reparameterization of y=beta0+beta1*x, but I am using this parameterization because I am specifically interested in confidence bounds for beta2. I have looked at the help, but the closest hint I find is the I() function, and that does not seem to work this way. I confess that I am actually using S-plus, but there does not seem to be a resource like this list for S-plus. Any help would be greatly appreciated. Kind regards, Eric [[alternative HTML version deleted]]
On 6/9/05, Eric Hack <hack at tera.org> wrote:> Dear All, > > > > I am trying to specify the following fixed effects model for lme:If you have a linear fixed-effects model you should use lm, not lme.> > y ~ constant1 - beta1*(x - beta2) > > where y is the response, x is the independent variable, and the > operators above are real arithmetic operations of addition, subtraction, > and multiplication. I realize that this model is just a > reparameterization of y=beta0+beta1*x, but I am using this > parameterization because I am specifically interested in confidence > bounds for beta2.You would need to fit that as a nonlinear model. In reference to such models "linear" means "linear in the parameters" and that model isn't.> I have looked at the help, but the closest hint I find is the I() > function, and that does not seem to work this way. > > > > I confess that I am actually using S-plus, but there does not seem to be > a resource like this list for S-plus.Look for the S-news email list (http://www.biostat.wustl.edu/s-news/)
Thanks for the response. It is actually a repeated measures study, I just mention the fixed effects specification because I think I know the random effect specification, i.e.: Random = ~ 1|subject And thanks for the tip about the nonlinear model and the S-plus list. I will check out nlme and the other list. Eric On 6/9/05, Eric Hack <hack at tera.org> wrote:> Dear All, > > > > I am trying to specify the following fixed effects model for lme:If you have a linear fixed-effects model you should use lm, not lme.> > y ~ constant1 - beta1*(x - beta2) > > where y is the response, x is the independent variable, and the > operators above are real arithmetic operations of addition,subtraction,> and multiplication. I realize that this model is just a > reparameterization of y=beta0+beta1*x, but I am using this > parameterization because I am specifically interested in confidence > bounds for beta2.You would need to fit that as a nonlinear model. In reference to such models "linear" means "linear in the parameters" and that model isn't.> I have looked at the help, but the closest hint I find is the I() > function, and that does not seem to work this way. > > > > I confess that I am actually using S-plus, but there does not seem tobe> a resource like this list for S-plus.Look for the S-news email list (http://www.biostat.wustl.edu/s-news/)