lavaan2mirt unexpectedly estimates a covariance matrix

[simonfelding]

Could you add an option to stop lavaan2mirt from estimating a covariance matrix when one is not specified in the lavaan model?

It seems like an odd default and it caused problems for me, exemplified by this cryptic error:

> lavaan2mirt(data, lavaan_model, poly.itemtype="graded", est.mirt=TRUE)
Error in chol.default(gp$gcov) : 
  the leading minor of order 1 is not positive definite

where lavaan_model is simply

[1] "# Latent variables\nML2 =~ dss_3 + dss_10 + dss_11 + dss_12 + dss_13 + dss_15 + dss_16 + dss_18 + dss_19 + dss_20\nML4 =~ dss_4 + dss_6 + dss_7 + dss_14 + dss_17\nML3 =~ dss_1 + dss_2 + dss_8 + dss_9\nML1 =~ dss_5"

The $mirt.syntax with est.mirt=FALSE is

> lavaan2mirt(data, lavaan_model %>% gsub("_", "", .), est.mirt=FALSE)$mirt.syntax
$x
     Type  Parameters                       
[1,] "ML2" "1,2,3,4,5,6,7,8,9,10"           
[2,] "ML4" "11,12,13,14,15"                 
[3,] "ML3" "16,17,18,19"                    
[4,] "ML1" "20"                             
[5,] "COV" "ML2*ML2,ML4*ML4,ML3*ML3,ML1*ML1"

I added the %>% gsub("_", "", .) to both the model and the dataset because I get Error in TAM::lavaanify.IRT(lavmodel, data = dat) : Please do not use _ in variable names in tamaan() function! without it.

Now that I look at it, it also seems odd that ML3 for example is [3,] "ML3" "16,17,18,19", when it is ML3 =~ dss_1 + dss_2 + dss_8 + dss_9 in lavaan form. The dataset has them ordered from col 1 to 20, so I would expect it to be [3,] "ML3" "1,2,8,9". Wouldn't it be easier to just the variable names since mirt.model accepts this? Then itemnames= could be passed as an argument too.

If it's any use, I generated the above lavaan model using psych::efa_to_cfa()

Thanks for the great and otherwise very useful package!