LOOIC value of brm model

Tiril Borge

unread,

Apr 13, 2018, 8:36:23 AM4/13/18

to brms-users

I am using the brm function with three different family specifications to estimate which model fits my data best. I have zero inflated count data and I am using the families beta (b), zero inflated beta (zib) and beta binomial2 (bb2) (for b and zib I have transformed the outcome variable to range between 0 and 1). When using the LOO function on the three different models I get negative LOOIC values for the b and zib models, and a positive LOOIC value for the bb2 model. I am a novice concerning the use of LOOIC so what I am wondering is if the LOOIC value can be both positive and negative (I have only seen positive values specified in the literature I have read)? If it can only be positive, do you have any suggestions for what I might have done wrong when specifying the b and zib model? Thanks so much in advance for your reply. Best regards, Tiril

Paul Buerkner

unread,

Apr 14, 2018, 4:29:12 AM4/14/18

to Tiril Borge, brms-users

The values of all information criteria might as well be negative. There is nothing special about it.

2018-04-13 14:36 GMT+02:00 Tiril Borge <tiril...@gmail.com>:

I am using the brm function with three different family specifications to estimate which model fits my data best. I have zero inflated count data and I am using the families beta (b), zero inflated beta (zib) and beta binomial2 (bb2) (for b and zib I have transformed the outcome variable to range between 0 and 1). When using the LOO function on the three different models I get negative LOOIC values for the b and zib models, and a positive LOOIC value for the bb2 model. I am a novice concerning the use of LOOIC so what I am wondering is if the LOOIC value can be both positive and negative (I have only seen positive values specified in the literature I have read)? If it can only be positive, do you have any suggestions for what I might have done wrong when specifying the b and zib model? Thanks so much in advance for your reply. Best regards, Tiril

--
You received this message because you are subscribed to the Google Groups "brms-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to brms-users+unsubscribe@googlegroups.com.
To post to this group, send email to brms-...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/brms-users/d2e7b33b-e363-414e-9aa6-b4380868c714%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Tiril Borge

unread,

Apr 14, 2018, 5:51:25 AM4/14/18

to Paul Buerkner, brms-users

Ok, thank you for the swift reply. I then have a follow- up question: Best model fit is dictated as the one which have the lowest IC value. As I have two models with neg IC and one with pos IC, does the lowest value = closest to zero? Thanks again, regards Tiril

Sendt fra min iPhone

Paul Buerkner

unread,

Apr 14, 2018, 5:54:19 AM4/14/18

to Tiril Borge, brms-users

lowest = lowest on the real line. That is most negative = best fit. Zero has no special meaning for information criteria.

Wiktor Soral

unread,

Apr 14, 2018, 2:27:27 PM4/14/18

to brms-users

I believe there is one additional issue in the case presented by Tiril. Outcome variable for b and zib models was transformed, but for bb2 not. Correct me if I am wrong, but the latter model is not comparable with the former ones in such case. This is probably the reason why LOOIC for bb2 is so different from LOOIC for b and zib.

W dniu sobota, 14 kwietnia 2018 11:54:19 UTC+2 użytkownik Paul Buerkner napisał:

lowest = lowest on the real line. That is most negative = best fit. Zero has no special meaning for information criteria.

2018-04-14 11:51 GMT+02:00 Tiril Borge <tiril...@gmail.com>:

Ok, thank you for the swift reply. I then have a follow- up question: Best model fit is dictated as the one which have the lowest IC value. As I have two models with neg IC and one with pos IC, does the lowest value = closest to zero? Thanks again, regards Tiril

Sendt fra min iPhone

14. apr. 2018 kl. 10:29 skrev Paul Buerkner <paul.b...@gmail.com>:

The values of all information criteria might as well be negative. There is nothing special about it.

2018-04-13 14:36 GMT+02:00 Tiril Borge <tiril...@gmail.com>:

I am using the brm function with three different family specifications to estimate which model fits my data best. I have zero inflated count data and I am using the families beta (b), zero inflated beta (zib) and beta binomial2 (bb2) (for b and zib I have transformed the outcome variable to range between 0 and 1). When using the LOO function on the three different models I get negative LOOIC values for the b and zib models, and a positive LOOIC value for the bb2 model. I am a novice concerning the use of LOOIC so what I am wondering is if the LOOIC value can be both positive and negative (I have only seen positive values specified in the literature I have read)? If it can only be positive, do you have any suggestions for what I might have done wrong when specifying the b and zib model? Thanks so much in advance for your reply. Best regards, Tiril

--
You received this message because you are subscribed to the Google Groups "brms-users" group.

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

Paul Buerkner

unread,

Apr 14, 2018, 3:11:27 PM4/14/18

to Wiktor Soral, brms-users

you are right. I haven't looked at the models actually. you can only compare models which use the exact same response variable.

To view this discussion on the web visit https://groups.google.com/d/msgid/brms-users/93bc525b-411c-447b-b603-93632a8186a0%40googlegroups.com.

Tiril Borge

unread,

Apr 16, 2018, 2:49:27 AM4/16/18

to Paul Buerkner, Wiktor Soral, brms-users

Thank you very much for your responses! Regards, Tiril

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

To post to this group, send email to brms-...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/brms-users/93bc525b-411c-447b-b603-93632a8186a0%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

You received this message because you are subscribed to the Google Groups "brms-users" group.

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

To post to this group, send email to brms-...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/brms-users/CAGoSky94nM1M0L1y6OH6RQMsGx8SKt-rcoynseJHUv%3DKFVOoKw%40mail.gmail.com.

Reply all

Reply to author

Forward