using the boxTidwell function in the CAR package and getting a bizarre error

I am trying to use the boxTidwell function in the CAR package in r to run a number of tests on continuous data. My data looks something like this:

Gender Age    X1      X2   Outcome
  M    20.1   1.23   4.43     1
  F    19.5   2.33   3.21     0
  M    18.0   1.33   7.55     1
  M    17.2   3.22   6.44     0
  M    12.5   4.15   8.99     1
  F    14.2   5.15  10.22     0
  F    13.9   6.12  12.34     1 
  F     9.4   7.12   3.21     1

When I use boxTidwell on the dataframe, I get an error

library(car)    
gender<-c("M","F","M","M","M","F","F","F")
    age<-c(20.1, 19.5, 18.0, 17.2, 12.5, 14.2, 13.9, 9.4)
    X1<-c(1.23,2.33,1.33,3.22,4.15,5.15,6.12,7.12)
    X2<-c(4.43,3.21,7.55,6.44,8.99,10.22,12.34,3.21)
    outcome<-c(1,0,1,0,1,0,1,1)
    df<-cbind(gender,age,X1,X2,outcome)
    as.data.frame(df)
    boxTidwell(outcome~age+X1+X2, ~gender, data=df)

Error in boxTidwell.default(y, X1, X2, max.iter = max.iter, tol = tol, : the variables to be transformed must have only positive values In addition: Warning message: In model.response(mf, "numeric") : using type = "numeric" with a factor response will be ignored

I am not sure what the problem is, I assume it is because I am using a binary outcome. Any suggestion would be much appreciated

1 answer

  • answered 2018-02-13 02:59 akrun

    The data is insufficient for the algorithm to come up with a solution

    boxTidwell(outcome~age+X1+X2, ~gender, data=df)
    #     Score Statistic   p-value MLE of lambda
    #age      -0.3575862 0.7206530      4.339394
    #X1        0.3081380 0.7579773      3.377788
    #X2       -0.9979096 0.3183232     29.886634
    

    It is noticeable if we subset the data created below to mimic the OP's data (of 9 rows)

    boxTidwell(outcome~age+X1+X2, ~gender, data=df[1:8,])
    

    Error in lm.fit(cbind(1, x.log.x, x1.p, x2), y, ...) : NA/NaN/Inf in 'x'

    NOTE: In the OP's post, the data.frame is created after converting to matrix (with cbind). It is problematic as matrix can hold only a single class and all the columns convert to factor with as.data.frame (or character if stringsAsFactors = FALSE)

    data

    set.seed(24)
    df <- data.frame(gender = sample(c("M", "F"), 100, replace = TRUE),
        age = rnorm(100, 20, 1), X1 = rnorm(100, 4, 1), X2 = rnorm(100, 10, 1),
        outcome = sample(0:1, 100, replace = TRUE))