[R] test of significance for nlme coefficients

Lindsey Root lroot at umiami.ir.miami.edu
Fri Mar 4 19:49:54 CET 2005


I am using the nlme package to fit an exponential decay model to a
longitudinal data set:

> ##Model a variable called RASCH16A as an exponential function ## of an
initial value "A" and a decay rate "K"
> fm2 <- nlme(model = RASCH16A ~  A* (exp(-K*DAYS)), 
+ ###A has nonzero fixed and random effects, K has nonzero fixed and
random effects
+ fixed = A + K ~ 1,
+ data = rasch,
+ random = A + K ~ 1 | SUBJECT,
+ ##Give a start value for the fixed effects, expressed using the
concatenate function
+ start = c(60, 0.007))

However, 20 of the 161 estimates decay rates are NOT decay rates, but
are very small positive growth rates (all less than a standard deviation
(.03) from zero).  I would like to run a significance test to see how
many of these positive growth rates are significantly different from
zero. However, the nlme package only offers confidence intervals for the
model parameters, not the person-specific coefficients. I would welcome
any and all suggestions. 

Thanks if you can help,
Lindsey Root

More information about the R-help mailing list