# pef$s.dummy is first declared as a 629*12 matrix (i.e. a total of 629 observations in

# 12 children) with all elements set to missing (NA).

# The "for{}" loop then sets all elements in the k"1 column to 0 except those relating to

# child k which take the value of the corresponding element of pef$s.

# Fit model gee.mod.slopes <- gee(x=pef$s.dummy, y=pef$m, id=pef$child, varfun="gaussian", link="identity", corstr="exchangeable", intercept=T, summarize=F)

# The only change from Figure 3 is that x (denoting the explanatory variables) is now

# declared to be the matrix pef$s.dummy rather than the single variable pef$s

# To interpret output:


# Elements 2 to 13 of the vector gee.mod.slopes$regest now hold the slopes for individual

# children.

# Empirical mean of separate slopes print(mean(gee.mod.slopes$regest[2:13]))

# Empirical standard error of empirical mean slope print(sqrt(var(gee.mod.slopes$regest[2:13])/12)) # denotes a comment in Splus. Unbolded text represents active Splus code.

Figure 11. S-plus code required to fit a GEE model with an exchangeable correlation structure allowing a separate slope in each child setting, the response of portable PEF meters to changes in true PEF is relatively poor and that important falls in true PEF can be associated with declines in portable meter-based PEF that are markedly attenuated and might easily be obscured by random within-child variability. On the basis of the analysis described in this paper, we would now argue that although the overall response of portable PEF meters to changes in true PEF is poor, it is probable that this problem is more serious in some children than in others. This suggests that if it was possible to identify children in whom the response was good, regular portable PEF monitoring might well be useful in this subgroup.

Was this article helpful?

0 0

Post a comment