Can the boxTidwell function handle binary outcome variables? - r

Question

I initially wanted to run a boxTidwell() (found in the "car" package) analysis on my prospective Logistic Regression model (BinaryOutcomeVar ~ ContinuousPredVar + ContinuousPredVar^2 + ContinuousPredVar^3). I ran into issues:
Error in x - xbar : non-numeric argument to binary operator
In addition: Warning message:
In mean.default(x) : argument is not numeric or logical: returning NA
So, I created a reproducable example for demonstrating the error:
Doesn't work:
boxTidwell(formula = Treatment ~ uptake, other.x = ~ poly(x = colnames(CO2)[c(1,2,4)], degree = 2), data = CO2)
boxTidwell(y = CO2$Treatment, x = CO2$uptake)
Works:
boxTidwell(formula = prestige ~ income + education, other.x = ~ poly(x = women , degree = 2), data = Prestige)
I've been goofing around with the other.x parameter and am guessing that's the issue.
Question
So, does anyone know if 1. the boxTidwell() function works with binary outcome variables 2. the logic behind the other.x, because I can't get my dummy example to work either.

Answer 1

After further searching, it looks like the car:::boxTidwell can't handle the binary outcome variable in the formula, but it can be hand coded:
require(MASS)
require(car)
d1<-read.csv("path for your csv file",sep=',',header=TRUE)
x<-d1$explanatory variable name
y<-d1$dependent variable name
#FIT IS DONE USING THE glm FUNCTION
m1res <- glm(y ~ x,family=binomial(link = "logit"))
coeff1<- coefficients(summary(m1res))
lnx<-x*log(x)
m2res <- glm(y ~ x+lnx ,family=binomial(link = "logit"))
coeff2<- coefficients(summary(m2res))
alpha0<-1.0
pvalue<-coeff2[3,4]
pvalue
beta1<-coeff1[2,1]
beta2<-coeff2[3,1]
iter<-0
err<-1
while (pvalue<0.1) {
alpha <-(beta2/beta1)+alpha0
err<-abs(alpha-alpha0)
alpha0<-alpha
mx<-x^alpha
m1res <- glm(y ~ mx,family=binomial(link = "logit"))
coeff1<- coefficients(summary(m1res))
mlnx<-mx*log(x)
m2res <- glm(y ~ mx+mlnx ,family=binomial(link = "logit"))
coeff2<- coefficients(summary(m2res))
pvalue<-coeff2[3,4]
beta1<-coeff1[2,1]
beta2<-coeff2[3,1]
iter<- iter+1
}
# PRINT THE POWER TO CONSOLE
alpha
above code taken from:
https://sites.google.com/site/ayyalaprem/box-tidwelltransform