CODEWITHSUNDEEP
rglm
Jakob
Jakob

Reputation: 1463

R: Debug glm "subscript out of bounds" in ExtremeBounds command

I'm running code using ExtremeBounds and glm (logit) through several specifications and sometimes I run into the following problem:

Data <- data.frame(var1=rbinom(30,1,0.2),var2=rbinom(30,1,0.2),var3=rnorm(30),coll1=rep(3:32),coll2=rep(1:30))
eba <- eba(data=Data, y="var1", 
       free="var2", 
       doubtful=c("var3","coll1","coll2"),
       reg.fun=glm, k=1, vif=7, draws=50, weights = "lri", family = binomial(logit))

where the error message is:

Error in reg.summary$coefficients[variable.label, 1] : 
subscript out of bounds

Since my dataset is quite big, I have no idea which variable causes this problem in my actual application. Does anyone know one of the answers to these questions:

  1. how can I see which variable causes it?

  2. how can I drop it automatically (maybe with a message like: "coll2 omitted due to collinearity")

  3. Am I right that this problem is caused by glm?

Upvotes: 1

Views: 1147

Answers (1)

Ben Bolker
Ben Bolker

Reputation: 226761

Posting as an answer, a bit long for a comment:

  • set options(error=recover)
  • running the code above (without se.fun=se.robust) gives
1: eba(data = Data, y = "var1", free = "var2", doubtful = c("var3", "coll1", "
2: .eba.wrap(formula = formula, data = data, y = y, free = free, doubtful = do
3: run.eba(y = y, v = focus, x = free, combinations = comb, data = data, level
4: run.regressions(y, v, x, combinations, data, level, vif.max, vars.labels, v

I choose option #4 (the lowest level)

I can run ls() to see all the variables that exist within this scope. Since I know the problem occurs with reg.summary$coefficients[variable.label, 1] I can try

Browse[1]> variable.label
#   coll2 
# "coll2" 
Browse[1]> str(reg.summary$coefficients)
#  num [1:3, 1:4] -0.9154 0.9436 -0.0504 1.1159 0.9728 ...
#  - attr(*, "dimnames")=List of 2
#   ..$ : chr [1:3] "(Intercept)" "var2" "coll1"
#   ..$ : chr [1:4] "Estimate" "Std. Error" "z value" "Pr(>|z|)"

That shows me that reg.summary$coefficients has apparently dropped the coll2 variable ...

Unfortunately I don't know enough about what eba() is doing to know how to proceed farther with diagnosis ... if you do, you can inspect the reg.summary object further to try to figure out what's going wrong ...

Upvotes: 4

Related Questions