Error using deltaMethod with nls object in R

Question

I'm trying to use the deltaMethod in the car library, but I'm getting an odd error.

library(car)
x=c(1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 8, 9, 9, 9, 9, 10, 10, 10, 10, 11, 11, 11, 11)
y=c(-0.78636545259908996, -0.48499513735893701, -0.61035206318152102, -0.60156864486986295, -0.61323703005521701, -0.33091952573467498, -0.269895273193686, -0.32222378534205598, -0.53183084634683997, -0.96631869084439304, -0.77105781684519603, -0.524039870915605, -0.41181303531095498, -0.27581842299642001, -0.72085673574325404, -0.35874718580022702, -0.30752543764527501, -0.090745334342823197, -0.465889655296298, -0.20115970219526799, -0.0511742487116199, 0.0100170907454752, -0.176138595601495, 0.042138062483845398, 0.00081247733328697303, -0.0045220167465173499, 0.57326735553016905, 0.116862163616526, 0.0072264835163109399, 0.48714531471859701, 0.83738659120408598, 0.83740077959237003, 0.48762419789728001, 0.20072016467283199, 0.56916547038663201, 0.14651949468445999, 0.575517323481333, 0.72715907067082697, 0.99958886855260898, 0.36070109242748599, 0.49335611371191601, 0.27098248212991599, 1.28001727666798, 0.36192955257384501)
gdat=data.frame(x,y)
fit=nls(y~a+b*(exp(-exp(s*(x-m)))),data=gdat,start=list(a=-0.5,b=1,s=-0.6,m=5))
deltaMethod(fit,"m-s*log((1/0.05)-1)")

The error I get is: Error in eval(expr, envir, enclos) : object 'ParaParam43' not found

While trying to solve this I noticed a couple things:

• the number after ParaParam changes depending on the parameters asked
• it works with lm objects, although I haven't tried the other supported objects
• in the above example, deltaMethod(fit,"m") works, but none of the other parameters do. This might be due to only m being in names(fit). Apparently the default method uses names(fit) instead of coef(fit) (as for other types of objects), so could it be that the method isn't recognising fit as a nls object? Do I need to tell deltaMethod what type of object to expect?

I used this a while back with the now deprecated alr3 delta.method without problems, but it now gives the same error.

Anyone have any ideas?

I'm using R version 2.13 and car version 2.0-10

Thanks

Answer 1

That's a bug in the deltaMethod.default method, and should be reported to the R developer team.

The reason for the bug is in following lines in the code :

for (i in seq(along = para.names)) {
    g <- gsub(para.names[i], std.names.ordered[i], g)
}

This rather amazing way of replacing things makes that all parameters that are names with any letter from Param will cause the code to return something that is not wanted. What it is supposed to do, is change the para.name ("s" in this case) into "Paramx". As it loops over it, it will change "s" to "Param3", and then it will change "Param3" to "ParaParam43", as it replaces the "m" by "Param4".

Obvious solution is to rename your parameters :

library(car)
x=c(1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 8, 9, 9, 9, 9, 10, 10, 10, 10, 11, 11, 11, 11)
y=c(-0.78636545259908996, -0.48499513735893701, -0.61035206318152102, -0.60156864486986295, -0.61323703005521701, -0.33091952573467498, -0.269895273193686, -0.32222378534205598, -0.53183084634683997, -0.96631869084439304, -0.77105781684519603, -0.524039870915605, -0.41181303531095498, -0.27581842299642001, -0.72085673574325404, -0.35874718580022702, -0.30752543764527501, -0.090745334342823197, -0.465889655296298, -0.20115970219526799, -0.0511742487116199, 0.0100170907454752, -0.176138595601495, 0.042138062483845398, 0.00081247733328697303, -0.0045220167465173499, 0.57326735553016905, 0.116862163616526, 0.0072264835163109399, 0.48714531471859701, 0.83738659120408598, 0.83740077959237003, 0.48762419789728001, 0.20072016467283199, 0.56916547038663201, 0.14651949468445999, 0.575517323481333, 0.72715907067082697, 0.99958886855260898, 0.36070109242748599, 0.49335611371191601, 0.27098248212991599, 1.28001727666798, 0.36192955257384501)
gdat=data.frame(x,y)
fit=nls(y~d+b*(exp(-exp(s*(x-k)))),data=gdat,start=list(d=-0.5,b=1,s=-0.6,k=5))
deltaMethod(fit,"k-s*log((1/0.05)-1)")

runs without error.

Answer 2

Wow, that was a fun bug to track down...

If you look in deltaMethod.default, you'll see it altering the names of the parameters to things like 'Param1' using gsub:

g <- gsub(para.names[i], std.names.ordered[i], g)

in a loop. Since you have a parameter named 'm', you get a double substitution since there is an 'm' in 'Param1'! This explains the error you get. Changing your parameter from 'm' to 'z' allowed the code to run for me.

I suggest contacting the package author with this information.