using btm function with eps = 0 returns error

[SanVerhavert]

Executing the following line of code (with data from data.txt with specs: sep = ";", dec = "." )
Ability_sirt <- btm( data = Data, fix.eta = 0, fix.delta = 0, eps = 0 )
returns the error and warning

Error in while ((iter < maxiter) & (max.change > conv)) { : 
  missing value where TRUE/FALSE needed
In addition: Warning message:
In max(theta_ch, na.rm = TRUE) :
  no non-missing arguments to max; returning -Inf

The problem is located in the file btm.r on line 62. Calculating the mean of a vector containing INF values results in an INF value. If there is no epsilon correction (i.e. eps = 0) line 60 produces an -INF value (in the example data) because there is a 0 score.

A possible solution, in my opinion, is to insert the block from line 65 to line 76 between line 60 and 61 and add the line of code theta[elim_persons_index] <- NA after the line theta_elim <- theta[ elim_persons_index ] in the moved block.

That part of code should look as follows

theta <- 0.8 * stats::qlogis(propscore)
elim_persons <- FALSE
if (sum(propscore %in% c(0, 1)) > 0) {
  elim_persons <- TRUE
  elim_persons_index <- which(propscore %in% c(0, 1))
  theta_elim <- theta[elim_persons_index]
  theta[elim_persons_index] <- NA
  indicator_elim <- dat0
  indicator_elim[, 3] <- 0
  indicator_elim[indicator_elim[, 1] %in% elim_persons_index, 
                 3] <- 1
  indicator_elim[indicator_elim[, 2] %in% elim_persons_index, 
                 3] <- 1
}
if (center.theta) {
  theta <- theta - mean(theta, na.rm = T )
}
some.fixed.theta <- FALSE

data.txt

[alexanderrobitzsch]

I decided to replace the line for calculating initial values by theta <- stats::qlogis( ( propscore + .1 ) / 1.2 ) which avoids inifinite estimates. I submitted a new package version on Github.

[SanVerhavert]

That is indeed a more elegant solution.
I am unfornunately unable to test the change myself because the devtools::install_github( "alexanderrobitzsch/sirt" ) cannot install the package and returns the following output (last 8 lines)

init.o:init.c:(.text+0x0): multiple definition of `R_init_sirt'
RcppExports.o:RcppExports.cpp:(.text+0x1f0): first defined here
collect2.exe: error: ld returned 1 exit status
no DLL was created
ERROR: compilation failed for package 'sirt'
* removing 'C:/Users/SVerhavert/Documents/R/win-library/3.5/sirt'
In R CMD INSTALL
Installation failed: Command failed (1)

[alexanderrobitzsch]

Yes, I agree. This was an incorrect version. I just updated a new version in which the problem should have been solved.

[SanVerhavert]

The change works fine! Thanks!