binomial mixed effects model BIC - R vs SPSS

Question

I'm trying to calculate Bayes Factor from my data and I'm getting very different results in R and SPSS for my mixed effects model. It's fine for a linear one, but not binomial. Here is the R code:

``memory.model = glmer(correct ~ (1|ps) + (1|item), data=memorystudy, family=binomial, glmerControl(optimize = "bobyqa"))
memory.model2 = glmer (correct ~ encoding + retrieval + (1|ps) + (1|item), data=memorystudy, family=binomial, glmerControl(optimize = "bobyqa"))
memory.model3 = glmer (correct ~ encoding*retrieval + (1|ps) + (1|item), data=memorystudy, family=binomial, glmerControl(optimize = "bobyqa"))``

and the SPSS syntax:

``GENLINMIXED
/FIELDS TARGET = correct
/DATA_STRUCTURE SUBJECTS = ps*item COVARIANCE_TYPE = VARIANCE_COMPONENTS
/TARGET_OPTIONS DISTRIBUTION = BINOMIAL LINK = LOGIT REFERENCE = 0
/FIXED EFFECTS = encoding retrieval
/RANDOM USE_INTERCEPT = TRUE SUBJECTS = ps*item.``

    ``GENLINMIXED
/FIELDS TARGET = correct
/DATA_STRUCTURE SUBJECTS = ps*item COVARIANCE_TYPE = VARIANCE_COMPONENTS
/TARGET_OPTIONS DISTRIBUTION = BINOMIAL LINK = LOGIT REFERENCE = 0
/FIXED EFFECTS = encoding retrieval encoding*retrieval
/RANDOM USE_INTERCEPT = TRUE SUBJECTS = ps*item.``

The BIC in R is about 2280 for each model, and in SPSS it's 13973... And SPSS doesn't give me any results such as effect sizes, so I can't compare it with R. Is there a way to change something in SPSS so it gives me consistent results?

Answer 1

I'll admit that I'm not an SPSS user, so if there are SPSS users out there, feel free to correct this answer. I notice that in your R models, ps and item are separate random intercepts. Skimming the SPSS manual (see e.g. this link, under the section "linear mixed model"), it appears these should be specified separately, e.g.

``GENLINMIXED
/FIELDS TARGET = correct
/DATA_STRUCTURE SUBJECTS = ps*item COVARIANCE_TYPE = VARIANCE_COMPONENTS
/TARGET_OPTIONS DISTRIBUTION = BINOMIAL LINK = LOGIT REFERENCE = 0
/FIXED EFFECTS = encoding retrieval
/RANDOM USE_INTERCEPT = TRUE SUBJECTS = ps
/RANDOM USE_INTERCEPT = TRUE SUBJECTS = item.``