[R-sig-ME] How to interpret the output for nested random factors in glmer and what to write in a table
N o s t a l g i a
kenj|ro @end|ng |rom @ho|n@@c@jp
Tue Aug 27 11:09:36 CEST 2024
Hi, this is something I posted on Cross Validated a few days ago, but
since I failed to get any substantial answers, I am posting a question
here, hoping someone could help me, as I am getting desperate.
I am looking at an alternation of a character in a word using a Japanese
congressional record. I have a binary outcome dependent variable (kuni),
the date of the meeting expressed as the standardized number of days
since 1949/05/31 (days_cnt2) as the only fixed variable, and two random
factors for meeting (meetid) and word (morphid). Because there are
several meetings in a given day, and a meeting is specific to a specific
day, the meeting variable is nested under days_cnt2. Thus, the model
looks like kuni~days_cnt2+(1+days_cnt2|morphid)+(1|days_cnt2/meetid). I
ran a binomial logistic regression with glmer and got the following output:
Command:
result <-
glmer(kuni~days_cnt2+(1+days_cnt2|morphid)+(1|days_cnt2/meetid),data=glmmdata6,family=binomial,
control=glmerControl(optimizer="bobyqa",optCtrl=list(maxfun=2e5))
A part of the output (summary):
Random effects:
Groups Name Variance Std.Dev. Corr
meetid:days_cnt2 (Intercept) 14.0795 3.7523
days_cnt2 (Intercept) 26.7335 5.1704
morphid (Intercept) 3.3889 1.8409
days_cnt2 0.2283 0.4778 0.71
Number of obs: 748186, groups:
meetid:days_cnt2, 12034; days_cnt2, 1189; morphid, 301
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -9.2793 0.3416 -27.16 <2e-16 ***
days_cnt2 -11.8274 0.3045 -38.84 <2e-16 ***
The ICC values are:
Adjusted ICC: 0.931 Unadjusted ICC: 0.237
Now I have several questions for this result:
1. Am I right in saying word (morphid) has less effect than meeting
(meetid) on kuni, based on the fact that the former has only about half
SD (1.8409) of the latter (3.7523)?
2. How should I interpret the two days_cnt2 parameter values in the
random effects section?
3. When publishing the result in a paper, should I include the values
for all of the four effects (meetid:days_cnt2, days_cnt2, morphid, and
days_cnt2) or should I skip days_cnt2?
Thanks in advance,
Kenjiro Matsuda
More information about the R-sig-mixed-models
mailing list