Chloride {NRAIA} | R Documentation |
The Chloride
data frame has 54 rows and 2 columns representing
measurements of the chloride ion concentration in blood cells suspended
in a salt solution.
This data frame contains the following columns:
There is a noticeable serial correlation in the residuals from a fit of
a simple asymptotic regression model to these data, as described in
section 3.8 of Bates and Watts (Wiley, 1988). We can either use a
modified model with nls
, as described in the reference, or model
it directly with the gnls
function.
Bates and Watts (1998), Nonlinear Regression Analysis and Its Applications, Wiley (Appendix A4.1).
Originally from Sredni (1970), Ph.D. Thesis, University of Wisconsin - Madison.
## Fit nonlinear regression model (fm1 <- nls(conc ~ Asym*(1 - prop*exp(-exp(lrc)*Time)), data = Chloride, start = c(Asym = 50, prop = 0.6, lrc = log(0.25)), trace = TRUE)) plotfit(fm1, xlab = "Time (min)", ylab = "Chloride ion concentration (%)") plot(resid(fm1) ~ fitted(fm1), # plot shows patterns in the residuals xlab = "Least squares fitted values", ylab = "Least squares residuals", main = "Residuals from least squares fit") abline(h = 0, lty = 2, lwd = 0) plot(resid(fm1)[-1], resid(fm1)[-length(resid(fm1))], # lag plot xlab = "Least squares residual", ylab = "Lagged least squares residual", main = "Lag plot of least squares residuals") abline(h = 0, lty = 2, lwd = 0) abline(v = 0, lty = 2, lwd = 0) ## Use conditional linearity in another formulation of the model (fm2 <- nls(conc ~ cbind(1, exp(-exp(lrc)*Time)), data = Chloride, start = c(lrc = log(0.25)), algorithm = "plinear", trace = TRUE)) ## Using a self-Starting model in yet another formulation (fm3 <- nls(conc ~ SSasympOff(Time, Asym, c0, lrc), data = Chloride))