What does that mean?
More generally (mostly lme4 maintainers will care about this): I suspect that the PIRLS warning is most often thrown when an NaN occurs within the computations done in C++ code for *any* reason (predictions outside of the domain of the link function, complete separation, weirdly misspecified models, etc.). Figuring out how to better flag these at the point(s) of origin might help a lot with reporting useful errors.
On 2019-05-02 10:20 a.m., Martin Maechler wrote:
Ben Bolker
on Wed, 1 May 2019 15:40:44 -0400 writes:
> On 2019-04-29 8:09 p.m., DESPINA MICHAILIDOU wrote:
>> I am trying to run the following analysis and receive the following output
>>
>> glmm_Comb_PH_tod <- glmer(Comb_PH_tod~ CA_effect + (1 | ID/SCAN_DATE/Side),
>> data=TAK_data, family=binomial(link = "logit"))
what happens if you (Despina Michailidou) add 'verbose = 2,' in the above call to glmer() ? You should get output which may help us to help you .. (Note to lme4 maintainers: I'm working at improving our help pages so people are less likely to *not* notice the 'verbose' argument.) Best, Martin Maechler ETH Zurich
>> summary(glmm_Comb_PH_tod)
>>
>> Output
>> Error in length(value <- as.numeric(value)) == 1L :
>> (maxstephalfit) PIRLS step-halvings failed to reduce deviance in
>> pwrssUpdate
>>> summary(glmm_Comb_PH_tod)
>> Error in summary(glmm_Comb_PH_tod) : object 'glmm_Comb_PH_tod' not found
>>
>> How can I fix that? Any suggestions? I am very new to R.
>>
>> Thank you in advance.
>> Despina
> The second error is relatively easy to understand; since your first
> command (using glmer(...)) didn't work, no object "glmm_Comb_PH_tod" has
> been created for you to summarize.
> The first part is harder. My guess would be that there's complete
> separation ; I've added a few more notes about the "PIRLS step-halving"
> error to the GLMM FAQ
> <http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html>. But it's hard
> to know without seeing the actual data.
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models