Blavaan: Prior Predictive Check Error

[Gabe Avakian Orona]

Hi All,

Any help with how to obtain the prior predictive check in blavaan? I followed these steps on the ecmerkle page, but receive the following two errors: 

Stan model 'stanmarg' does not contain samples.

AND:

> bfit3 <- bsem(b3, data = osf, std.lv=TRUE, prisamp = T) Error in eval(mc, parent.frame()) : Exception: mismatch in dimension declared and found in context; processing stage=data initialization; variable name=YXbar; position=0; dims declared=(16,12); dims found=(1,12) (in 'string', line 513, column 2 to column 60) failed to create the sampler; sampling not done Computing post-estimation metrics (including lvs if requested)... Stan model 'stanmarg' does not contain samples. Error in rowvec[samppar] <- which(grepl(stanvec[j], names(b.est)) & !(grepl("_c\\[", : replacement has length zero

Here is my full model:

b3 <- '
RT1 =~
prior("normal(0.81, .30)")*CRT1 +  
prior("normal(0.50, .30)")*BBS1 +
prior("normal(0.72, .30)")*BRN1 +
prior("normal(0.37, .30)")*WAS1
# Time 2: Reflective Thinking
RT2 =~
prior("normal(0.81, .30)")*CRT2 +
prior("normal(0.81, .30)")*CRT4 +
prior("normal(0.81, .30)")*CRTv +
prior("normal(0.50, .30)")*BBS2 +
prior("normal(0.72, .30)")*BRN2 +
prior("normal(0.37, .30)")*WAS2
# Regression
RT2 ~ c*trt + prior("normal(0.65, .40)")*RT1 + prior("normal(0.35, .15)")*logDT + b*logDT
logDT ~ prior("normal(0.22, .20)")*trt + a*trt

# Estimated/Generated Quantities
IndirectEffect_ab := a*b
TotalEffect := (a*b)+c
# Correlated Uniqueness
BRN1 ~~ BRN2
WAS1 ~~ WAS2
BBS1 ~~ BBS2
CRT1 ~~ CRT2
'
#_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

The error occurs without the indirect effects, as well.

Thank you for any help you can offer!

Best,

Gabe

[Ed Merkle]

Gabe,

Thanks for the report. I haven't been able to reproduce this error using the CRAN version of blavaan (0.5-4) or the github version of blavaan. Is there a chance you are on an earlier version of blavaan, or perhaps an earlier version of lavaan (earlier than 0.6-17)?

Ed

--

You received this message because you are subscribed to the Google Groups "blavaan" group.

To unsubscribe from this group and stop receiving emails from it, send an email to blavaan+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/blavaan/267722c5-e4d4-4bb3-ab2f-e9b23bb41094n%40googlegroups.com.

[Gabe Avakian Orona]

Hi Ed, 

Thank you for looking into this. I'm using .5-2

However, the issue does NOT occur when I use different data, such as the one in your example online. 

Gabe

--

Gabe Avakian Orona, Ph.D.

Postdoctoral Research Fellow
Hector Research Institute of Education Sciences and Psychology
University of Tübingen
gabrie...@uni-tuebingen.de

[Ed Merkle]

Thanks for the info. Are you able to upgrade to 0.5-4 or the github version? I believe this error may have been fixed in more recent versions (or, at least, some related errors have been fixed).

Ed

On Tue, 2024-05-21 at 18:48 +0200, Gabe Avakian Orona wrote:

Hi Ed, 

Thank you for looking into this. I'm using .5-2

[Gabe Avakian Orona]

Hi Ed,

Yes, I was able to upgrade to 0.5-4. However, the same error occurs:

Any assistance would be greatly appreciated. (I also tried this on two different computers--same result.)

Thank you very much, Ed. 

Best,
Gabe

[Ed Merkle]

Thanks for following up. I dug deeper and can reproduce this when the dataset has missing values.

For now, I think the workaround is to eliminate missing data from your dataset when you use prisamp = TRUE. This shouldn't have any influence on results because you are sampling from priors, and the data are not involved in the sampling. But I should also be able to fix it soon to avoid the error.

Ed

On Wed, 2024-05-22 at 09:37 +0200, Gabe Avakian Orona wrote:

Hi Ed,

Yes, I was able to upgrade to 0.5-4. However, the same error occurs:

[Gabe Avakian Orona]

Hi Ed,

Thank you very much. That worked great. Appreciate your help.

Best,

Gabe